Searched +hist:0 +hist:b61f8a4 (Results 1 - 25 of 194) sorted by last modified time
/linux-master/fs/xfs/ | ||
H A D | xfs_inode.c | diff e23d7e82 Thu Mar 14 11:07:02 MDT 2024 Andrey Albershteyn <aalbersh@redhat.com> xfs: allow cross-linking special files without project quota There's an issue that if special files is created before quota project is enabled, then it's not possible to link this file. This works fine for normal files. This happens because xfs_quota skips special files (no ioctls to set necessary flags). The check for having the same project ID for source and destination then fails as source file doesn't have any ID. mkfs.xfs -f /dev/sda mount -o prjquota /dev/sda /mnt/test mkdir /mnt/test/foo mkfifo /mnt/test/foo/fifo1 xfs_quota -xc "project -sp /mnt/test/foo 9" /mnt/test > Setting up project 9 (path /mnt/test/foo)... > xfs_quota: skipping special file /mnt/test/foo/fifo1 > Processed 1 (/etc/projects and cmdline) paths for project 9 with recursion depth infinite (-1). ln /mnt/test/foo/fifo1 /mnt/test/foo/fifo1_link > ln: failed to create hard link '/mnt/test/testdir/fifo1_link' => '/mnt/test/testdir/fifo1': Invalid cross-device link mkfifo /mnt/test/foo/fifo2 ln /mnt/test/foo/fifo2 /mnt/test/foo/fifo2_link Fix this by allowing linking of special files to the project quota if special files doesn't have any ID set (ID = 0). Signed-off-by: Andrey Albershteyn <aalbersh@redhat.com> Reviewed-by: "Darrick J. Wong" <djwong@kernel.org> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff 0e24ec3c Thu Feb 22 01:33:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: remember sick inodes that get inactivated If an unhealthy inode gets inactivated, remember this fact in the per-fs health summary. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> |
H A D | xfs_super.c | diff d8d222e0 Mon Jan 15 21:33:07 MST 2024 Dave Chinner <dchinner@redhat.com> xfs: read only mounts with fsopen mount API are busted Recently xfs/513 started failing on my test machines testing "-o ro,norecovery" mount options. This was being emitted in dmesg: [ 9906.932724] XFS (pmem0): no-recovery mounts must be read-only. Turns out, readonly mounts with the fsopen()/fsconfig() mount API have been busted since day zero. It's only taken 5 years for debian unstable to start using this "new" mount API, and shortly after this I noticed xfs/513 had started to fail as per above. The syscall trace is: fsopen("xfs", FSOPEN_CLOEXEC) = 3 mount_setattr(-1, NULL, 0, NULL, 0) = -1 EINVAL (Invalid argument) ..... fsconfig(3, FSCONFIG_SET_STRING, "source", "/dev/pmem0", 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "ro", NULL, 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "norecovery", NULL, 0) = 0 fsconfig(3, FSCONFIG_CMD_CREATE, NULL, NULL, 0) = -1 EINVAL (Invalid argument) close(3) = 0 Showing that the actual mount instantiation (FSCONFIG_CMD_CREATE) is what threw out the error. During mount instantiation, we call xfs_fs_validate_params() which does: /* No recovery flag requires a read-only mount */ if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) { xfs_warn(mp, "no-recovery mounts must be read-only."); return -EINVAL; } and xfs_is_readonly() checks internal mount flags for read only state. This state is set in xfs_init_fs_context() from the context superblock flag state: /* * Copy binary VFS mount flags we are interested in. */ if (fc->sb_flags & SB_RDONLY) set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate); With the old mount API, all of the VFS specific superblock flags had already been parsed and set before xfs_init_fs_context() is called, so this all works fine. However, in the brave new fsopen/fsconfig world, xfs_init_fs_context() is called from fsopen() context, before any VFS superblock have been set or parsed. Hence if we use fsopen(), the internal XFS readonly state is *never set*. Hence anything that depends on xfs_is_readonly() actually returning true for read only mounts is broken if fsopen() has been used to mount the filesystem. Fix this by moving this internal state initialisation to xfs_fs_fill_super() before we attempt to validate the parameters that have been set prior to the FSCONFIG_CMD_CREATE call being made. Signed-off-by: Dave Chinner <dchinner@redhat.com> Fixes: 73e5fff98b64 ("xfs: switch to use the new mount-api") cc: stable@vger.kernel.org Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff d8d222e0 Mon Jan 15 21:33:07 MST 2024 Dave Chinner <dchinner@redhat.com> xfs: read only mounts with fsopen mount API are busted Recently xfs/513 started failing on my test machines testing "-o ro,norecovery" mount options. This was being emitted in dmesg: [ 9906.932724] XFS (pmem0): no-recovery mounts must be read-only. Turns out, readonly mounts with the fsopen()/fsconfig() mount API have been busted since day zero. It's only taken 5 years for debian unstable to start using this "new" mount API, and shortly after this I noticed xfs/513 had started to fail as per above. The syscall trace is: fsopen("xfs", FSOPEN_CLOEXEC) = 3 mount_setattr(-1, NULL, 0, NULL, 0) = -1 EINVAL (Invalid argument) ..... fsconfig(3, FSCONFIG_SET_STRING, "source", "/dev/pmem0", 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "ro", NULL, 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "norecovery", NULL, 0) = 0 fsconfig(3, FSCONFIG_CMD_CREATE, NULL, NULL, 0) = -1 EINVAL (Invalid argument) close(3) = 0 Showing that the actual mount instantiation (FSCONFIG_CMD_CREATE) is what threw out the error. During mount instantiation, we call xfs_fs_validate_params() which does: /* No recovery flag requires a read-only mount */ if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) { xfs_warn(mp, "no-recovery mounts must be read-only."); return -EINVAL; } and xfs_is_readonly() checks internal mount flags for read only state. This state is set in xfs_init_fs_context() from the context superblock flag state: /* * Copy binary VFS mount flags we are interested in. */ if (fc->sb_flags & SB_RDONLY) set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate); With the old mount API, all of the VFS specific superblock flags had already been parsed and set before xfs_init_fs_context() is called, so this all works fine. However, in the brave new fsopen/fsconfig world, xfs_init_fs_context() is called from fsopen() context, before any VFS superblock have been set or parsed. Hence if we use fsopen(), the internal XFS readonly state is *never set*. Hence anything that depends on xfs_is_readonly() actually returning true for read only mounts is broken if fsopen() has been used to mount the filesystem. Fix this by moving this internal state initialisation to xfs_fs_fill_super() before we attempt to validate the parameters that have been set prior to the FSCONFIG_CMD_CREATE call being made. Signed-off-by: Dave Chinner <dchinner@redhat.com> Fixes: 73e5fff98b64 ("xfs: switch to use the new mount-api") cc: stable@vger.kernel.org Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff d8d222e0 Mon Jan 15 21:33:07 MST 2024 Dave Chinner <dchinner@redhat.com> xfs: read only mounts with fsopen mount API are busted Recently xfs/513 started failing on my test machines testing "-o ro,norecovery" mount options. This was being emitted in dmesg: [ 9906.932724] XFS (pmem0): no-recovery mounts must be read-only. Turns out, readonly mounts with the fsopen()/fsconfig() mount API have been busted since day zero. It's only taken 5 years for debian unstable to start using this "new" mount API, and shortly after this I noticed xfs/513 had started to fail as per above. The syscall trace is: fsopen("xfs", FSOPEN_CLOEXEC) = 3 mount_setattr(-1, NULL, 0, NULL, 0) = -1 EINVAL (Invalid argument) ..... fsconfig(3, FSCONFIG_SET_STRING, "source", "/dev/pmem0", 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "ro", NULL, 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "norecovery", NULL, 0) = 0 fsconfig(3, FSCONFIG_CMD_CREATE, NULL, NULL, 0) = -1 EINVAL (Invalid argument) close(3) = 0 Showing that the actual mount instantiation (FSCONFIG_CMD_CREATE) is what threw out the error. During mount instantiation, we call xfs_fs_validate_params() which does: /* No recovery flag requires a read-only mount */ if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) { xfs_warn(mp, "no-recovery mounts must be read-only."); return -EINVAL; } and xfs_is_readonly() checks internal mount flags for read only state. This state is set in xfs_init_fs_context() from the context superblock flag state: /* * Copy binary VFS mount flags we are interested in. */ if (fc->sb_flags & SB_RDONLY) set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate); With the old mount API, all of the VFS specific superblock flags had already been parsed and set before xfs_init_fs_context() is called, so this all works fine. However, in the brave new fsopen/fsconfig world, xfs_init_fs_context() is called from fsopen() context, before any VFS superblock have been set or parsed. Hence if we use fsopen(), the internal XFS readonly state is *never set*. Hence anything that depends on xfs_is_readonly() actually returning true for read only mounts is broken if fsopen() has been used to mount the filesystem. Fix this by moving this internal state initialisation to xfs_fs_fill_super() before we attempt to validate the parameters that have been set prior to the FSCONFIG_CMD_CREATE call being made. Signed-off-by: Dave Chinner <dchinner@redhat.com> Fixes: 73e5fff98b64 ("xfs: switch to use the new mount-api") cc: stable@vger.kernel.org Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff d8d222e0 Mon Jan 15 21:33:07 MST 2024 Dave Chinner <dchinner@redhat.com> xfs: read only mounts with fsopen mount API are busted Recently xfs/513 started failing on my test machines testing "-o ro,norecovery" mount options. This was being emitted in dmesg: [ 9906.932724] XFS (pmem0): no-recovery mounts must be read-only. Turns out, readonly mounts with the fsopen()/fsconfig() mount API have been busted since day zero. It's only taken 5 years for debian unstable to start using this "new" mount API, and shortly after this I noticed xfs/513 had started to fail as per above. The syscall trace is: fsopen("xfs", FSOPEN_CLOEXEC) = 3 mount_setattr(-1, NULL, 0, NULL, 0) = -1 EINVAL (Invalid argument) ..... fsconfig(3, FSCONFIG_SET_STRING, "source", "/dev/pmem0", 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "ro", NULL, 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "norecovery", NULL, 0) = 0 fsconfig(3, FSCONFIG_CMD_CREATE, NULL, NULL, 0) = -1 EINVAL (Invalid argument) close(3) = 0 Showing that the actual mount instantiation (FSCONFIG_CMD_CREATE) is what threw out the error. During mount instantiation, we call xfs_fs_validate_params() which does: /* No recovery flag requires a read-only mount */ if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) { xfs_warn(mp, "no-recovery mounts must be read-only."); return -EINVAL; } and xfs_is_readonly() checks internal mount flags for read only state. This state is set in xfs_init_fs_context() from the context superblock flag state: /* * Copy binary VFS mount flags we are interested in. */ if (fc->sb_flags & SB_RDONLY) set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate); With the old mount API, all of the VFS specific superblock flags had already been parsed and set before xfs_init_fs_context() is called, so this all works fine. However, in the brave new fsopen/fsconfig world, xfs_init_fs_context() is called from fsopen() context, before any VFS superblock have been set or parsed. Hence if we use fsopen(), the internal XFS readonly state is *never set*. Hence anything that depends on xfs_is_readonly() actually returning true for read only mounts is broken if fsopen() has been used to mount the filesystem. Fix this by moving this internal state initialisation to xfs_fs_fill_super() before we attempt to validate the parameters that have been set prior to the FSCONFIG_CMD_CREATE call being made. Signed-off-by: Dave Chinner <dchinner@redhat.com> Fixes: 73e5fff98b64 ("xfs: switch to use the new mount-api") cc: stable@vger.kernel.org Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff d8d222e0 Mon Jan 15 21:33:07 MST 2024 Dave Chinner <dchinner@redhat.com> xfs: read only mounts with fsopen mount API are busted Recently xfs/513 started failing on my test machines testing "-o ro,norecovery" mount options. This was being emitted in dmesg: [ 9906.932724] XFS (pmem0): no-recovery mounts must be read-only. Turns out, readonly mounts with the fsopen()/fsconfig() mount API have been busted since day zero. It's only taken 5 years for debian unstable to start using this "new" mount API, and shortly after this I noticed xfs/513 had started to fail as per above. The syscall trace is: fsopen("xfs", FSOPEN_CLOEXEC) = 3 mount_setattr(-1, NULL, 0, NULL, 0) = -1 EINVAL (Invalid argument) ..... fsconfig(3, FSCONFIG_SET_STRING, "source", "/dev/pmem0", 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "ro", NULL, 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "norecovery", NULL, 0) = 0 fsconfig(3, FSCONFIG_CMD_CREATE, NULL, NULL, 0) = -1 EINVAL (Invalid argument) close(3) = 0 Showing that the actual mount instantiation (FSCONFIG_CMD_CREATE) is what threw out the error. During mount instantiation, we call xfs_fs_validate_params() which does: /* No recovery flag requires a read-only mount */ if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) { xfs_warn(mp, "no-recovery mounts must be read-only."); return -EINVAL; } and xfs_is_readonly() checks internal mount flags for read only state. This state is set in xfs_init_fs_context() from the context superblock flag state: /* * Copy binary VFS mount flags we are interested in. */ if (fc->sb_flags & SB_RDONLY) set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate); With the old mount API, all of the VFS specific superblock flags had already been parsed and set before xfs_init_fs_context() is called, so this all works fine. However, in the brave new fsopen/fsconfig world, xfs_init_fs_context() is called from fsopen() context, before any VFS superblock have been set or parsed. Hence if we use fsopen(), the internal XFS readonly state is *never set*. Hence anything that depends on xfs_is_readonly() actually returning true for read only mounts is broken if fsopen() has been used to mount the filesystem. Fix this by moving this internal state initialisation to xfs_fs_fill_super() before we attempt to validate the parameters that have been set prior to the FSCONFIG_CMD_CREATE call being made. Signed-off-by: Dave Chinner <dchinner@redhat.com> Fixes: 73e5fff98b64 ("xfs: switch to use the new mount-api") cc: stable@vger.kernel.org Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff d8d222e0 Mon Jan 15 21:33:07 MST 2024 Dave Chinner <dchinner@redhat.com> xfs: read only mounts with fsopen mount API are busted Recently xfs/513 started failing on my test machines testing "-o ro,norecovery" mount options. This was being emitted in dmesg: [ 9906.932724] XFS (pmem0): no-recovery mounts must be read-only. Turns out, readonly mounts with the fsopen()/fsconfig() mount API have been busted since day zero. It's only taken 5 years for debian unstable to start using this "new" mount API, and shortly after this I noticed xfs/513 had started to fail as per above. The syscall trace is: fsopen("xfs", FSOPEN_CLOEXEC) = 3 mount_setattr(-1, NULL, 0, NULL, 0) = -1 EINVAL (Invalid argument) ..... fsconfig(3, FSCONFIG_SET_STRING, "source", "/dev/pmem0", 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "ro", NULL, 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "norecovery", NULL, 0) = 0 fsconfig(3, FSCONFIG_CMD_CREATE, NULL, NULL, 0) = -1 EINVAL (Invalid argument) close(3) = 0 Showing that the actual mount instantiation (FSCONFIG_CMD_CREATE) is what threw out the error. During mount instantiation, we call xfs_fs_validate_params() which does: /* No recovery flag requires a read-only mount */ if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) { xfs_warn(mp, "no-recovery mounts must be read-only."); return -EINVAL; } and xfs_is_readonly() checks internal mount flags for read only state. This state is set in xfs_init_fs_context() from the context superblock flag state: /* * Copy binary VFS mount flags we are interested in. */ if (fc->sb_flags & SB_RDONLY) set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate); With the old mount API, all of the VFS specific superblock flags had already been parsed and set before xfs_init_fs_context() is called, so this all works fine. However, in the brave new fsopen/fsconfig world, xfs_init_fs_context() is called from fsopen() context, before any VFS superblock have been set or parsed. Hence if we use fsopen(), the internal XFS readonly state is *never set*. Hence anything that depends on xfs_is_readonly() actually returning true for read only mounts is broken if fsopen() has been used to mount the filesystem. Fix this by moving this internal state initialisation to xfs_fs_fill_super() before we attempt to validate the parameters that have been set prior to the FSCONFIG_CMD_CREATE call being made. Signed-off-by: Dave Chinner <dchinner@redhat.com> Fixes: 73e5fff98b64 ("xfs: switch to use the new mount-api") cc: stable@vger.kernel.org Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff d8d222e0 Mon Jan 15 21:33:07 MST 2024 Dave Chinner <dchinner@redhat.com> xfs: read only mounts with fsopen mount API are busted Recently xfs/513 started failing on my test machines testing "-o ro,norecovery" mount options. This was being emitted in dmesg: [ 9906.932724] XFS (pmem0): no-recovery mounts must be read-only. Turns out, readonly mounts with the fsopen()/fsconfig() mount API have been busted since day zero. It's only taken 5 years for debian unstable to start using this "new" mount API, and shortly after this I noticed xfs/513 had started to fail as per above. The syscall trace is: fsopen("xfs", FSOPEN_CLOEXEC) = 3 mount_setattr(-1, NULL, 0, NULL, 0) = -1 EINVAL (Invalid argument) ..... fsconfig(3, FSCONFIG_SET_STRING, "source", "/dev/pmem0", 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "ro", NULL, 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "norecovery", NULL, 0) = 0 fsconfig(3, FSCONFIG_CMD_CREATE, NULL, NULL, 0) = -1 EINVAL (Invalid argument) close(3) = 0 Showing that the actual mount instantiation (FSCONFIG_CMD_CREATE) is what threw out the error. During mount instantiation, we call xfs_fs_validate_params() which does: /* No recovery flag requires a read-only mount */ if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) { xfs_warn(mp, "no-recovery mounts must be read-only."); return -EINVAL; } and xfs_is_readonly() checks internal mount flags for read only state. This state is set in xfs_init_fs_context() from the context superblock flag state: /* * Copy binary VFS mount flags we are interested in. */ if (fc->sb_flags & SB_RDONLY) set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate); With the old mount API, all of the VFS specific superblock flags had already been parsed and set before xfs_init_fs_context() is called, so this all works fine. However, in the brave new fsopen/fsconfig world, xfs_init_fs_context() is called from fsopen() context, before any VFS superblock have been set or parsed. Hence if we use fsopen(), the internal XFS readonly state is *never set*. Hence anything that depends on xfs_is_readonly() actually returning true for read only mounts is broken if fsopen() has been used to mount the filesystem. Fix this by moving this internal state initialisation to xfs_fs_fill_super() before we attempt to validate the parameters that have been set prior to the FSCONFIG_CMD_CREATE call being made. Signed-off-by: Dave Chinner <dchinner@redhat.com> Fixes: 73e5fff98b64 ("xfs: switch to use the new mount-api") cc: stable@vger.kernel.org Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff d8d222e0 Mon Jan 15 21:33:07 MST 2024 Dave Chinner <dchinner@redhat.com> xfs: read only mounts with fsopen mount API are busted Recently xfs/513 started failing on my test machines testing "-o ro,norecovery" mount options. This was being emitted in dmesg: [ 9906.932724] XFS (pmem0): no-recovery mounts must be read-only. Turns out, readonly mounts with the fsopen()/fsconfig() mount API have been busted since day zero. It's only taken 5 years for debian unstable to start using this "new" mount API, and shortly after this I noticed xfs/513 had started to fail as per above. The syscall trace is: fsopen("xfs", FSOPEN_CLOEXEC) = 3 mount_setattr(-1, NULL, 0, NULL, 0) = -1 EINVAL (Invalid argument) ..... fsconfig(3, FSCONFIG_SET_STRING, "source", "/dev/pmem0", 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "ro", NULL, 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "norecovery", NULL, 0) = 0 fsconfig(3, FSCONFIG_CMD_CREATE, NULL, NULL, 0) = -1 EINVAL (Invalid argument) close(3) = 0 Showing that the actual mount instantiation (FSCONFIG_CMD_CREATE) is what threw out the error. During mount instantiation, we call xfs_fs_validate_params() which does: /* No recovery flag requires a read-only mount */ if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) { xfs_warn(mp, "no-recovery mounts must be read-only."); return -EINVAL; } and xfs_is_readonly() checks internal mount flags for read only state. This state is set in xfs_init_fs_context() from the context superblock flag state: /* * Copy binary VFS mount flags we are interested in. */ if (fc->sb_flags & SB_RDONLY) set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate); With the old mount API, all of the VFS specific superblock flags had already been parsed and set before xfs_init_fs_context() is called, so this all works fine. However, in the brave new fsopen/fsconfig world, xfs_init_fs_context() is called from fsopen() context, before any VFS superblock have been set or parsed. Hence if we use fsopen(), the internal XFS readonly state is *never set*. Hence anything that depends on xfs_is_readonly() actually returning true for read only mounts is broken if fsopen() has been used to mount the filesystem. Fix this by moving this internal state initialisation to xfs_fs_fill_super() before we attempt to validate the parameters that have been set prior to the FSCONFIG_CMD_CREATE call being made. Signed-off-by: Dave Chinner <dchinner@redhat.com> Fixes: 73e5fff98b64 ("xfs: switch to use the new mount-api") cc: stable@vger.kernel.org Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff d8d222e0 Mon Jan 15 21:33:07 MST 2024 Dave Chinner <dchinner@redhat.com> xfs: read only mounts with fsopen mount API are busted Recently xfs/513 started failing on my test machines testing "-o ro,norecovery" mount options. This was being emitted in dmesg: [ 9906.932724] XFS (pmem0): no-recovery mounts must be read-only. Turns out, readonly mounts with the fsopen()/fsconfig() mount API have been busted since day zero. It's only taken 5 years for debian unstable to start using this "new" mount API, and shortly after this I noticed xfs/513 had started to fail as per above. The syscall trace is: fsopen("xfs", FSOPEN_CLOEXEC) = 3 mount_setattr(-1, NULL, 0, NULL, 0) = -1 EINVAL (Invalid argument) ..... fsconfig(3, FSCONFIG_SET_STRING, "source", "/dev/pmem0", 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "ro", NULL, 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "norecovery", NULL, 0) = 0 fsconfig(3, FSCONFIG_CMD_CREATE, NULL, NULL, 0) = -1 EINVAL (Invalid argument) close(3) = 0 Showing that the actual mount instantiation (FSCONFIG_CMD_CREATE) is what threw out the error. During mount instantiation, we call xfs_fs_validate_params() which does: /* No recovery flag requires a read-only mount */ if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) { xfs_warn(mp, "no-recovery mounts must be read-only."); return -EINVAL; } and xfs_is_readonly() checks internal mount flags for read only state. This state is set in xfs_init_fs_context() from the context superblock flag state: /* * Copy binary VFS mount flags we are interested in. */ if (fc->sb_flags & SB_RDONLY) set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate); With the old mount API, all of the VFS specific superblock flags had already been parsed and set before xfs_init_fs_context() is called, so this all works fine. However, in the brave new fsopen/fsconfig world, xfs_init_fs_context() is called from fsopen() context, before any VFS superblock have been set or parsed. Hence if we use fsopen(), the internal XFS readonly state is *never set*. Hence anything that depends on xfs_is_readonly() actually returning true for read only mounts is broken if fsopen() has been used to mount the filesystem. Fix this by moving this internal state initialisation to xfs_fs_fill_super() before we attempt to validate the parameters that have been set prior to the FSCONFIG_CMD_CREATE call being made. Signed-off-by: Dave Chinner <dchinner@redhat.com> Fixes: 73e5fff98b64 ("xfs: switch to use the new mount-api") cc: stable@vger.kernel.org Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff d8d222e0 Mon Jan 15 21:33:07 MST 2024 Dave Chinner <dchinner@redhat.com> xfs: read only mounts with fsopen mount API are busted Recently xfs/513 started failing on my test machines testing "-o ro,norecovery" mount options. This was being emitted in dmesg: [ 9906.932724] XFS (pmem0): no-recovery mounts must be read-only. Turns out, readonly mounts with the fsopen()/fsconfig() mount API have been busted since day zero. It's only taken 5 years for debian unstable to start using this "new" mount API, and shortly after this I noticed xfs/513 had started to fail as per above. The syscall trace is: fsopen("xfs", FSOPEN_CLOEXEC) = 3 mount_setattr(-1, NULL, 0, NULL, 0) = -1 EINVAL (Invalid argument) ..... fsconfig(3, FSCONFIG_SET_STRING, "source", "/dev/pmem0", 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "ro", NULL, 0) = 0 fsconfig(3, FSCONFIG_SET_FLAG, "norecovery", NULL, 0) = 0 fsconfig(3, FSCONFIG_CMD_CREATE, NULL, NULL, 0) = -1 EINVAL (Invalid argument) close(3) = 0 Showing that the actual mount instantiation (FSCONFIG_CMD_CREATE) is what threw out the error. During mount instantiation, we call xfs_fs_validate_params() which does: /* No recovery flag requires a read-only mount */ if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) { xfs_warn(mp, "no-recovery mounts must be read-only."); return -EINVAL; } and xfs_is_readonly() checks internal mount flags for read only state. This state is set in xfs_init_fs_context() from the context superblock flag state: /* * Copy binary VFS mount flags we are interested in. */ if (fc->sb_flags & SB_RDONLY) set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate); With the old mount API, all of the VFS specific superblock flags had already been parsed and set before xfs_init_fs_context() is called, so this all works fine. However, in the brave new fsopen/fsconfig world, xfs_init_fs_context() is called from fsopen() context, before any VFS superblock have been set or parsed. Hence if we use fsopen(), the internal XFS readonly state is *never set*. Hence anything that depends on xfs_is_readonly() actually returning true for read only mounts is broken if fsopen() has been used to mount the filesystem. Fix this by moving this internal state initialisation to xfs_fs_fill_super() before we attempt to validate the parameters that have been set prior to the FSCONFIG_CMD_CREATE call being made. Signed-off-by: Dave Chinner <dchinner@redhat.com> Fixes: 73e5fff98b64 ("xfs: switch to use the new mount-api") cc: stable@vger.kernel.org Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> |
H A D | xfs_buf.c | diff 0b3a76e9 Mon Jan 15 15:59:46 MST 2024 Dave Chinner <dchinner@redhat.com> xfs: use GFP_KERNEL in pure transaction contexts When running in a transaction context, memory allocations are scoped to GFP_NOFS. Hence we don't need to use GFP_NOFS contexts in pure transaction context allocations - GFP_KERNEL will automatically get converted to GFP_NOFS as appropriate. Go through the code and convert all the obvious GFP_NOFS allocations in transaction context to use GFP_KERNEL. This further reduces the explicit use of GFP_NOFS in XFS. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: "Darrick J. Wong" <djwong@kernel.org> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff 0a97c01c Thu Nov 30 12:40:18 MST 2023 Nhat Pham <nphamcs@gmail.com> list_lru: allow explicit memcg and NUMA node selection Patch series "workload-specific and memory pressure-driven zswap writeback", v8. There are currently several issues with zswap writeback: 1. There is only a single global LRU for zswap, making it impossible to perform worload-specific shrinking - an memcg under memory pressure cannot determine which pages in the pool it owns, and often ends up writing pages from other memcgs. This issue has been previously observed in practice and mitigated by simply disabling memcg-initiated shrinking: https://lore.kernel.org/all/20230530232435.3097106-1-nphamcs@gmail.com/T/#u But this solution leaves a lot to be desired, as we still do not have an avenue for an memcg to free up its own memory locked up in the zswap pool. 2. We only shrink the zswap pool when the user-defined limit is hit. This means that if we set the limit too high, cold data that are unlikely to be used again will reside in the pool, wasting precious memory. It is hard to predict how much zswap space will be needed ahead of time, as this depends on the workload (specifically, on factors such as memory access patterns and compressibility of the memory pages). This patch series solves these issues by separating the global zswap LRU into per-memcg and per-NUMA LRUs, and performs workload-specific (i.e memcg- and NUMA-aware) zswap writeback under memory pressure. The new shrinker does not have any parameter that must be tuned by the user, and can be opted in or out on a per-memcg basis. As a proof of concept, we ran the following synthetic benchmark: build the linux kernel in a memory-limited cgroup, and allocate some cold data in tmpfs to see if the shrinker could write them out and improved the overall performance. Depending on the amount of cold data generated, we observe from 14% to 35% reduction in kernel CPU time used in the kernel builds. This patch (of 6): The interface of list_lru is based on the assumption that the list node and the data it represents belong to the same allocated on the correct node/memcg. While this assumption is valid for existing slab objects LRU such as dentries and inodes, it is undocumented, and rather inflexible for certain potential list_lru users (such as the upcoming zswap shrinker and the THP shrinker). It has caused us a lot of issues during our development. This patch changes list_lru interface so that the caller must explicitly specify numa node and memcg when adding and removing objects. The old list_lru_add() and list_lru_del() are renamed to list_lru_add_obj() and list_lru_del_obj(), respectively. It also extends the list_lru API with a new function, list_lru_putback, which undoes a previous list_lru_isolate call. Unlike list_lru_add, it does not increment the LRU node count (as list_lru_isolate does not decrement the node count). list_lru_putback also allows for explicit memcg and NUMA node selection. Link: https://lkml.kernel.org/r/20231130194023.4102148-1-nphamcs@gmail.com Link: https://lkml.kernel.org/r/20231130194023.4102148-2-nphamcs@gmail.com Signed-off-by: Nhat Pham <nphamcs@gmail.com> Suggested-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Tested-by: Bagas Sanjaya <bagasdotme@gmail.com> Cc: Chris Li <chrisl@kernel.org> Cc: Dan Streetman <ddstreet@ieee.org> Cc: Domenico Cerasuolo <cerasuolodomenico@gmail.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Muchun Song <muchun.song@linux.dev> Cc: Roman Gushchin <roman.gushchin@linux.dev> Cc: Seth Jennings <sjenning@redhat.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Shuah Khan <shuah@kernel.org> Cc: Vitaly Wool <vitaly.wool@konsulko.com> Cc: Yosry Ahmed <yosryahmed@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> diff e33c267a Tue May 31 21:22:24 MDT 2022 Roman Gushchin <roman.gushchin@linux.dev> mm: shrinkers: provide shrinkers with names Currently shrinkers are anonymous objects. For debugging purposes they can be identified by count/scan function names, but it's not always useful: e.g. for superblock's shrinkers it's nice to have at least an idea of to which superblock the shrinker belongs. This commit adds names to shrinkers. register_shrinker() and prealloc_shrinker() functions are extended to take a format and arguments to master a name. In some cases it's not possible to determine a good name at the time when a shrinker is allocated. For such cases shrinker_debugfs_rename() is provided. The expected format is: <subsystem>-<shrinker_type>[:<instance>]-<id> For some shrinkers an instance can be encoded as (MAJOR:MINOR) pair. After this change the shrinker debugfs directory looks like: $ cd /sys/kernel/debug/shrinker/ $ ls dquota-cache-16 sb-devpts-28 sb-proc-47 sb-tmpfs-42 mm-shadow-18 sb-devtmpfs-5 sb-proc-48 sb-tmpfs-43 mm-zspool:zram0-34 sb-hugetlbfs-17 sb-pstore-31 sb-tmpfs-44 rcu-kfree-0 sb-hugetlbfs-33 sb-rootfs-2 sb-tmpfs-49 sb-aio-20 sb-iomem-12 sb-securityfs-6 sb-tracefs-13 sb-anon_inodefs-15 sb-mqueue-21 sb-selinuxfs-22 sb-xfs:vda1-36 sb-bdev-3 sb-nsfs-4 sb-sockfs-8 sb-zsmalloc-19 sb-bpf-32 sb-pipefs-14 sb-sysfs-26 thp-deferred_split-10 sb-btrfs:vda2-24 sb-proc-25 sb-tmpfs-1 thp-zero-9 sb-cgroup2-30 sb-proc-39 sb-tmpfs-27 xfs-buf:vda1-37 sb-configfs-23 sb-proc-41 sb-tmpfs-29 xfs-inodegc:vda1-38 sb-dax-11 sb-proc-45 sb-tmpfs-35 sb-debugfs-7 sb-proc-46 sb-tmpfs-40 [roman.gushchin@linux.dev: fix build warnings] Link: https://lkml.kernel.org/r/Yr+ZTnLb9lJk6fJO@castle Reported-by: kernel test robot <lkp@intel.com> Link: https://lkml.kernel.org/r/20220601032227.4076670-4-roman.gushchin@linux.dev Signed-off-by: Roman Gushchin <roman.gushchin@linux.dev> Cc: Christophe JAILLET <christophe.jaillet@wanadoo.fr> Cc: Dave Chinner <dchinner@redhat.com> Cc: Hillf Danton <hdanton@sina.com> Cc: Kent Overstreet <kent.overstreet@gmail.com> Cc: Muchun Song <songmuchun@bytedance.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> diff b9b3fe15 Wed Apr 20 16:44:59 MDT 2022 Dave Chinner <david@fromorbit.com> xfs: convert buffer flags to unsigned. 5.18 w/ std=gnu11 compiled with gcc-5 wants flags stored in unsigned fields to be unsigned. This manifests as a compiler error such as: /kisskb/src/fs/xfs/./xfs_trace.h:432:2: note: in expansion of macro 'TP_printk' TP_printk("dev %d:%d daddr 0x%llx bbcount 0x%x hold %d pincount %d " ^ /kisskb/src/fs/xfs/./xfs_trace.h:440:5: note: in expansion of macro '__print_flags' __print_flags(__entry->flags, "|", XFS_BUF_FLAGS), ^ /kisskb/src/fs/xfs/xfs_buf.h:67:4: note: in expansion of macro 'XBF_UNMAPPED' { XBF_UNMAPPED, "UNMAPPED" } ^ /kisskb/src/fs/xfs/./xfs_trace.h:440:40: note: in expansion of macro 'XFS_BUF_FLAGS' __print_flags(__entry->flags, "|", XFS_BUF_FLAGS), ^ /kisskb/src/fs/xfs/./xfs_trace.h: In function 'trace_raw_output_xfs_buf_flags_class': /kisskb/src/fs/xfs/xfs_buf.h:46:23: error: initializer element is not constant #define XBF_UNMAPPED (1 << 31)/* do not map the buffer */ as __print_flags assigns XFS_BUF_FLAGS to a structure that uses an unsigned long for the flag. Since this results in the value of XBF_UNMAPPED causing a signed integer overflow, the result is technically undefined behavior, which gcc-5 does not accept as an integer constant. This is based on a patch from Arnd Bergman <arnd@arndb.de>. Reported-by: Geert Uytterhoeven <geert@linux-m68k.org> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Chandan Babu R <chandan.babu@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff b9b3fe15 Wed Apr 20 16:44:59 MDT 2022 Dave Chinner <david@fromorbit.com> xfs: convert buffer flags to unsigned. 5.18 w/ std=gnu11 compiled with gcc-5 wants flags stored in unsigned fields to be unsigned. This manifests as a compiler error such as: /kisskb/src/fs/xfs/./xfs_trace.h:432:2: note: in expansion of macro 'TP_printk' TP_printk("dev %d:%d daddr 0x%llx bbcount 0x%x hold %d pincount %d " ^ /kisskb/src/fs/xfs/./xfs_trace.h:440:5: note: in expansion of macro '__print_flags' __print_flags(__entry->flags, "|", XFS_BUF_FLAGS), ^ /kisskb/src/fs/xfs/xfs_buf.h:67:4: note: in expansion of macro 'XBF_UNMAPPED' { XBF_UNMAPPED, "UNMAPPED" } ^ /kisskb/src/fs/xfs/./xfs_trace.h:440:40: note: in expansion of macro 'XFS_BUF_FLAGS' __print_flags(__entry->flags, "|", XFS_BUF_FLAGS), ^ /kisskb/src/fs/xfs/./xfs_trace.h: In function 'trace_raw_output_xfs_buf_flags_class': /kisskb/src/fs/xfs/xfs_buf.h:46:23: error: initializer element is not constant #define XBF_UNMAPPED (1 << 31)/* do not map the buffer */ as __print_flags assigns XFS_BUF_FLAGS to a structure that uses an unsigned long for the flag. Since this results in the value of XBF_UNMAPPED causing a signed integer overflow, the result is technically undefined behavior, which gcc-5 does not accept as an integer constant. This is based on a patch from Arnd Bergman <arnd@arndb.de>. Reported-by: Geert Uytterhoeven <geert@linux-m68k.org> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Chandan Babu R <chandan.babu@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 01728b44 Thu Mar 17 10:09:13 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_is_shutdown vs xlog_is_shutdown cage fight I've been chasing a recent resurgence in generic/388 recovery failure and/or corruption events. The events have largely been uninitialised inode chunks being tripped over in log recovery such as: XFS (pmem1): User initiated shutdown received. pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS (pmem1): Unmounting Filesystem XFS (pmem1): Mounting V5 Filesystem XFS (pmem1): Starting recovery (logdev: internal) XFS (pmem1): bad inode magic/vsn daddr 8723584 #0 (magic=1818) XFS (pmem1): Metadata corruption detected at xfs_inode_buf_verify+0x180/0x190, xfs_inode block 0x851c80 xfs_inode_buf_verify XFS (pmem1): Unmount and run xfs_repair XFS (pmem1): First 128 bytes of corrupted metadata buffer: 00000000: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000010: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000020: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000030: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000040: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000050: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000060: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ 00000070: 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 ................ XFS (pmem1): metadata I/O error in "xlog_recover_items_pass2+0x52/0xc0" at daddr 0x851c80 len 32 error 117 XFS (pmem1): log mount/recovery failed: error -117 XFS (pmem1): log mount failed There have been isolated random other issues, too - xfs_repair fails because it finds some corruption in symlink blocks, rmap inconsistencies, etc - but they are nowhere near as common as the uninitialised inode chunk failure. The problem has clearly happened at runtime before recovery has run; I can see the ICREATE log item in the log shortly before the actively recovered range of the log. This means the ICREATE was definitely created and written to the log, but for some reason the tail of the log has been moved past the ordered buffer log item that tracks INODE_ALLOC buffers and, supposedly, prevents the tail of the log moving past the ICREATE log item before the inode chunk buffer is written to disk. Tracing the fsstress processes that are running when the filesystem shut down immediately pin-pointed the problem: user shutdown marks xfs_mount as shutdown godown-213341 [008] 6398.022871: console: [ 6397.915392] XFS (pmem1): User initiated shutdown received. ..... aild tries to push ordered inode cluster buffer xfsaild/pmem1-213314 [001] 6398.022974: xfs_buf_trylock: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 16 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_inode_item_push+0x8e xfsaild/pmem1-213314 [001] 6398.022976: xfs_ilock_nowait: dev 259:1 ino 0x851c80 flags ILOCK_SHARED caller xfs_iflush_cluster+0xae xfs_iflush_cluster() checks xfs_is_shutdown(), returns true, calls xfs_iflush_abort() to kill writeback of the inode. Inode is removed from AIL, drops cluster buffer reference. xfsaild/pmem1-213314 [001] 6398.022977: xfs_ail_delete: dev 259:1 lip 0xffff88880247ed80 old lsn 7/20344 new lsn 7/21000 type XFS_LI_INODE flags IN_AIL xfsaild/pmem1-213314 [001] 6398.022978: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 17 pincount 0 lock 0 flags DONE|INODES|PAGES caller xfs_iflush_abort+0xd7 ..... All inodes on cluster buffer are aborted, then the cluster buffer itself is aborted and removed from the AIL *without writeback*: xfsaild/pmem1-213314 [001] 6398.023011: xfs_buf_error_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_ioend_fail+0x33 xfsaild/pmem1-213314 [001] 6398.023012: xfs_ail_delete: dev 259:1 lip 0xffff8888053efde8 old lsn 7/20344 new lsn 7/20344 type XFS_LI_BUF flags IN_AIL The inode buffer was at 7/20344 when it was removed from the AIL. xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_item_relse: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_done+0x31 xfsaild/pmem1-213314 [001] 6398.023012: xfs_buf_rele: dev 259:1 daddr 0x851c80 bbcount 0x20 hold 2 pincount 0 lock 0 flags ASYNC|DONE|STALE|INODES|PAGES caller xfs_buf_item_relse+0x39 ..... Userspace is still running, doing stuff. an fsstress process runs syncfs() or sync() and we end up in sync_fs_one_sb() which issues a log force. This pushes on the CIL: fsstress-213322 [001] 6398.024430: xfs_fs_sync_fs: dev 259:1 m_features 0x20000000019ff6e9 opstate (clean|shutdown|inodegc|blockgc) s_flags 0x70810000 caller sync_fs_one_sb+0x26 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x0 caller xfs_fs_sync_fs+0x82 fsstress-213322 [001] 6398.024430: xfs_log_force: dev 259:1 lsn 0x5f caller xfs_log_force+0x7c <...>-194402 [001] 6398.024467: kmem_alloc: size 176 flags 0x14 caller xlog_cil_push_work+0x9f And the CIL fills up iclogs with pending changes. This picks up the current tail from the AIL: <...>-194402 [001] 6398.024497: xlog_iclog_get_space: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x0 flags caller xlog_write+0x149 <...>-194402 [001] 6398.024498: xlog_iclog_switch: dev 259:1 state XLOG_STATE_ACTIVE refcnt 1 offset 0 lsn 0x700005408 flags caller xlog_state_get_iclog_space+0x37e <...>-194402 [001] 6398.024521: xlog_iclog_release: dev 259:1 state XLOG_STATE_WANT_SYNC refcnt 1 offset 32256 lsn 0x700005408 flags caller xlog_write+0x5f9 <...>-194402 [001] 6398.024522: xfs_log_assign_tail_lsn: dev 259:1 new tail lsn 7/21000, old lsn 7/20344, last sync 7/21448 And it moves the tail of the log to 7/21000 from 7/20344. This *moves the tail of the log beyond the ICREATE transaction* that was at 7/20344 and pinned by the inode cluster buffer that was cancelled above. .... godown-213341 [008] 6398.027005: xfs_force_shutdown: dev 259:1 tag logerror flags log_io|force_umount file fs/xfs/xfs_fsops.c line_num 500 godown-213341 [008] 6398.027022: console: [ 6397.915406] pmem1: writeback error on inode 12621949, offset 1019904, sector 12968096 godown-213341 [008] 6398.030551: console: [ 6397.919546] XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/ And finally the log itself is now shutdown, stopping all further writes to the log. But this is too late to prevent the corruption that moving the tail of the log forwards after we start cancelling writeback causes. The fundamental problem here is that we are using the wrong shutdown checks for log items. We've long conflated mount shutdown with log shutdown state, and I started separating that recently with the atomic shutdown state changes in commit b36d4651e165 ("xfs: make forced shutdown processing atomic"). The changes in that commit series are directly responsible for being able to diagnose this issue because it clearly separated mount shutdown from log shutdown. Essentially, once we start cancelling writeback of log items and removing them from the AIL because the filesystem is shut down, we *cannot* update the journal because we may have cancelled the items that pin the tail of the log. That moves the tail of the log forwards without having written the metadata back, hence we have corrupt in memory state and writing to the journal propagates that to the on-disk state. What commit b36d4651e165 makes clear is that log item state needs to change relative to log shutdown, not mount shutdown. IOWs, anything that aborts metadata writeback needs to check log shutdown state because log items directly affect log consistency. Having them check mount shutdown state introduces the above race condition where we cancel metadata writeback before the log shuts down. To fix this, this patch works through all log items and converts shutdown checks to use xlog_is_shutdown() rather than xfs_is_shutdown(), so that we don't start aborting metadata writeback before we shut off journal writes. AFAICT, this race condition is a zero day IO error handling bug in XFS that dates back to the introduction of XLOG_IO_ERROR, XLOG_STATE_IOERROR and XFS_FORCED_SHUTDOWN back in January 1997. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> |
H A D | xfs_trans.h | diff 0dc63c8a Thu Feb 22 01:43:36 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: launder in-memory btree buffers before transaction commit As we've noted in various places, all current users of in-memory btrees are online fsck. Online fsck only stages a btree long enough to rebuild an ondisk data structure, which means that the in-memory btree is ephemeral. Furthermore, if we encounter /any/ errors while updating an in-memory btree, all we do is tear down all the staged data and return an errno to userspace. In-memory btrees need not be transactional, so their buffers should not be committed to the ondisk log, nor should they be checkpointed by the AIL. That's just as well since the ephemeral nature of the btree means that the buftarg and the buffers may disappear quickly anyway. Therefore, we need a way to launder the btree buffers that get attached to the transaction by the generic btree code. Because the buffers are directly mapped to backing file pages, there's no need to bwrite them back to the tmpfs file. All we need to do is clean enough of the buffer log item state so that the bli can be detached from the buffer, remove the bli from the transaction's log item list, and reset the transaction dirty state as if the laundered items had never been there. For simplicity, create xfbtree transaction commit and cancel helpers that launder the in-memory btree buffers for callers. Once laundered, call the write verifier on non-stale buffers to avoid integrity issues, or punch a hole in the backing file for stale buffers. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 0d227466 Tue May 03 19:50:29 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: intent item whiteouts When we log modifications based on intents, we add both intent and intent done items to the modification being made. These get written to the log to ensure that the operation is re-run if the intent done is not found in the log. However, for operations that complete wholly within a single checkpoint, the change in the checkpoint is atomic and will never need replay. In this case, we don't need to actually write the intent and intent done items to the journal because log recovery will never need to manually restart this modification. Log recovery currently handles intent/intent done matching by inserting the intent into the AIL, then removing it when a matching intent done item is found. Hence for all the intent-based operations that complete within a checkpoint, we spend all that time parsing the intent/intent done items just to cancel them and do nothing with them. Hence it follows that the only time we actually need intents in the log is when the modification crosses checkpoint boundaries in the log and so may only be partially complete in the journal. Hence if we commit and intent done item to the CIL and the intent item is in the same checkpoint, we don't actually have to write them to the journal because log recovery will always cancel the intents. We've never really worried about the overhead of logging intents unnecessarily like this because the intents we log are generally very much smaller than the change being made. e.g. freeing an extent involves modifying at lease two freespace btree blocks and the AGF, so the EFI/EFD overhead is only a small increase in space and processing time compared to the overall cost of freeing an extent. However, delayed attributes change this cost equation dramatically, especially for inline attributes. In the case of adding an inline attribute, we only log the inode core and attribute fork at present. With delayed attributes, we now log the attr intent which includes the name and value, the inode core adn attr fork, and finally the attr intent done item. We increase the number of items we log from 1 to 3, and the number of log vectors (regions) goes up from 3 to 7. Hence we tripple the number of objects that the CIL has to process, and more than double the number of log vectors that need to be written to the journal. At scale, this means delayed attributes cause a non-pipelined CIL to become CPU bound processing all the extra items, resulting in a > 40% performance degradation on 16-way file+xattr create worklaods. Pipelining the CIL (as per 5.15) reduces the performance degradation to 20%, but now the limitation is the rate at which the log items can be written to the iclogs and iclogs be dispatched for IO and completed. Even log IO completion is slowed down by these intents, because it now has to process 3x the number of items in the checkpoint. Processing completed intents is especially inefficient here, because we first insert the intent into the AIL, then remove it from the AIL when the intent done is processed. IOWs, we are also doing expensive operations in log IO completion we could completely avoid if we didn't log completed intent/intent done pairs. Enter log item whiteouts. When an intent done is committed, we can check to see if the associated intent is in the same checkpoint as we are currently committing the intent done to. If so, we can mark the intent log item with a whiteout and immediately free the intent done item rather than committing it to the CIL. We can basically skip the entire formatting and CIL insertion steps for the intent done item. However, we cannot remove the intent item from the CIL at this point because the unlocked per-cpu CIL item lists do not permit removal without holding the CIL context lock exclusively. Transaction commit only holds the context lock shared, hence the best we can do is mark the intent item with a whiteout so that the CIL push can release it rather than writing it to the log. This means we never write the intent to the log if the intent done has also been committed to the same checkpoint, but we'll always write the intent if the intent done has not been committed or has been committed to a different checkpoint. This will result in correct log recovery behaviour in all cases, without the overhead of logging unnecessary intents. This intent whiteout concept is generic - we can apply it to all intent/intent done pairs that have a direct 1:1 relationship. The way deferred ops iterate and relog intents mean that all intents currently have a 1:1 relationship with their done intent, and hence we can apply this cancellation to all existing intent/intent done implementations. For delayed attributes with a 16-way 64kB xattr create workload, whiteouts reduce the amount of journalled metadata from ~2.5GB/s down to ~600MB/s and improve the creation rate from 9000/s to 14000/s. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff b9b3fe15 Wed Apr 20 16:44:59 MDT 2022 Dave Chinner <david@fromorbit.com> xfs: convert buffer flags to unsigned. 5.18 w/ std=gnu11 compiled with gcc-5 wants flags stored in unsigned fields to be unsigned. This manifests as a compiler error such as: /kisskb/src/fs/xfs/./xfs_trace.h:432:2: note: in expansion of macro 'TP_printk' TP_printk("dev %d:%d daddr 0x%llx bbcount 0x%x hold %d pincount %d " ^ /kisskb/src/fs/xfs/./xfs_trace.h:440:5: note: in expansion of macro '__print_flags' __print_flags(__entry->flags, "|", XFS_BUF_FLAGS), ^ /kisskb/src/fs/xfs/xfs_buf.h:67:4: note: in expansion of macro 'XBF_UNMAPPED' { XBF_UNMAPPED, "UNMAPPED" } ^ /kisskb/src/fs/xfs/./xfs_trace.h:440:40: note: in expansion of macro 'XFS_BUF_FLAGS' __print_flags(__entry->flags, "|", XFS_BUF_FLAGS), ^ /kisskb/src/fs/xfs/./xfs_trace.h: In function 'trace_raw_output_xfs_buf_flags_class': /kisskb/src/fs/xfs/xfs_buf.h:46:23: error: initializer element is not constant #define XBF_UNMAPPED (1 << 31)/* do not map the buffer */ as __print_flags assigns XFS_BUF_FLAGS to a structure that uses an unsigned long for the flag. Since this results in the value of XBF_UNMAPPED causing a signed integer overflow, the result is technically undefined behavior, which gcc-5 does not accept as an integer constant. This is based on a patch from Arnd Bergman <arnd@arndb.de>. Reported-by: Geert Uytterhoeven <geert@linux-m68k.org> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Chandan Babu R <chandan.babu@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff b9b3fe15 Wed Apr 20 16:44:59 MDT 2022 Dave Chinner <david@fromorbit.com> xfs: convert buffer flags to unsigned. 5.18 w/ std=gnu11 compiled with gcc-5 wants flags stored in unsigned fields to be unsigned. This manifests as a compiler error such as: /kisskb/src/fs/xfs/./xfs_trace.h:432:2: note: in expansion of macro 'TP_printk' TP_printk("dev %d:%d daddr 0x%llx bbcount 0x%x hold %d pincount %d " ^ /kisskb/src/fs/xfs/./xfs_trace.h:440:5: note: in expansion of macro '__print_flags' __print_flags(__entry->flags, "|", XFS_BUF_FLAGS), ^ /kisskb/src/fs/xfs/xfs_buf.h:67:4: note: in expansion of macro 'XBF_UNMAPPED' { XBF_UNMAPPED, "UNMAPPED" } ^ /kisskb/src/fs/xfs/./xfs_trace.h:440:40: note: in expansion of macro 'XFS_BUF_FLAGS' __print_flags(__entry->flags, "|", XFS_BUF_FLAGS), ^ /kisskb/src/fs/xfs/./xfs_trace.h: In function 'trace_raw_output_xfs_buf_flags_class': /kisskb/src/fs/xfs/xfs_buf.h:46:23: error: initializer element is not constant #define XBF_UNMAPPED (1 << 31)/* do not map the buffer */ as __print_flags assigns XFS_BUF_FLAGS to a structure that uses an unsigned long for the flag. Since this results in the value of XBF_UNMAPPED causing a signed integer overflow, the result is technically undefined behavior, which gcc-5 does not accept as an integer constant. This is based on a patch from Arnd Bergman <arnd@arndb.de>. Reported-by: Geert Uytterhoeven <geert@linux-m68k.org> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Chandan Babu R <chandan.babu@oracle.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 1aec7c3d Fri Apr 23 17:02:00 MDT 2021 Darrick J. Wong <djwong@kernel.org> xfs: remove obsolete AGF counter debugging In commit f8f2835a9cf3 we changed the behavior of XFS to use EFIs to remove blocks from an overfilled AGFL because there were complaints about transaction overruns that stemmed from trying to free multiple blocks in a single transaction. Unfortunately, that commit missed a subtlety in the debug-mode transaction accounting when a realtime volume is attached. If a realtime file undergoes a data fork mapping change such that realtime extents are allocated (or freed) in the same transaction that a data device block is also allocated (or freed), we can trip a debugging assertion. This can happen (for example) if a realtime extent is allocated and it is necessary to reshape the bmbt to hold the new mapping. When we go to allocate a bmbt block from an AG, the first thing the data device block allocator does is ensure that the freelist is the proper length. If the freelist is too long, it will trim the freelist to the proper length. In debug mode, trimming the freelist calls xfs_trans_agflist_delta() to record the decrement in the AG free list count. Prior to f8f28 we would put the free block back in the free space btrees in the same transaction, which calls xfs_trans_agblocks_delta() to record the increment in the AG free block count. Since AGFL blocks are included in the global free block count (fdblocks), there is no corresponding fdblocks update, so the AGFL free satisfies the following condition in xfs_trans_apply_sb_deltas: /* * Check that superblock mods match the mods made to AGF counters. */ ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) == (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta + tp->t_ag_btree_delta)); The comparison here used to be: (X + 0) == ((X+1) + -1 + 0), where X is the number blocks that were allocated. After commit f8f28 we defer the block freeing to the next chained transaction, which means that the calls to xfs_trans_agflist_delta and xfs_trans_agblocks_delta occur in separate transactions. The (first) transaction that shortens the free list trips on the comparison, which has now become: (X + 0) == ((X) + -1 + 0) because we haven't freed the AGFL block yet; we've only logged an intention to free it. When the second transaction (the deferred free) commits, it will evaluate the expression as: (0 + 0) == (1 + 0 + 0) and trip over that in turn. At this point, the astute reader may note that the two commits tagged by this patch have been in the kernel for a long time but haven't generated any bug reports. How is it that the author became aware of this bug? This originally surfaced as an intermittent failure when I was testing realtime rmap, but a different bug report by Zorro Lang reveals the same assertion occuring on !lazysbcount filesystems. The common factor to both reports (and why this problem wasn't previously reported) becomes apparent if we consider when xfs_trans_apply_sb_deltas is called by __xfs_trans_commit(): if (tp->t_flags & XFS_TRANS_SB_DIRTY) xfs_trans_apply_sb_deltas(tp); With a modern lazysbcount filesystem, transactions update only the percpu counters, so they don't need to set XFS_TRANS_SB_DIRTY, hence xfs_trans_apply_sb_deltas is rarely called. However, updates to the count of free realtime extents are not part of lazysbcount, so XFS_TRANS_SB_DIRTY will be set on transactions adding or removing data fork mappings to realtime files; similarly, XFS_TRANS_SB_DIRTY is always set on !lazysbcount filesystems. Dave mentioned in response to an earlier version of this patch: "IIUC, what you are saying is that this debug code is simply not exercised in normal testing and hasn't been for the past decade? And it still won't be exercised on anything other than realtime device testing? "...it was debugging code from 1994 that was largely turned into dead code when lazysbcounters were introduced in 2007. Hence I'm not sure it holds any value anymore." This debugging code isn't especially helpful - you can modify the flcount on one AG and the freeblks of another AG, and it won't trigger. Add the fact that nobody noticed for a decade, and let's just get rid of it (and start testing realtime :P). This bug was found by running generic/051 on either a V4 filesystem lacking lazysbcount; or a V5 filesystem with a realtime volume. Cc: bfoster@redhat.com, zlang@redhat.com Fixes: f8f2835a9cf3 ("xfs: defer agfl block frees when dfops is available") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Brian Foster <bfoster@redhat.com> diff 1aec7c3d Fri Apr 23 17:02:00 MDT 2021 Darrick J. Wong <djwong@kernel.org> xfs: remove obsolete AGF counter debugging In commit f8f2835a9cf3 we changed the behavior of XFS to use EFIs to remove blocks from an overfilled AGFL because there were complaints about transaction overruns that stemmed from trying to free multiple blocks in a single transaction. Unfortunately, that commit missed a subtlety in the debug-mode transaction accounting when a realtime volume is attached. If a realtime file undergoes a data fork mapping change such that realtime extents are allocated (or freed) in the same transaction that a data device block is also allocated (or freed), we can trip a debugging assertion. This can happen (for example) if a realtime extent is allocated and it is necessary to reshape the bmbt to hold the new mapping. When we go to allocate a bmbt block from an AG, the first thing the data device block allocator does is ensure that the freelist is the proper length. If the freelist is too long, it will trim the freelist to the proper length. In debug mode, trimming the freelist calls xfs_trans_agflist_delta() to record the decrement in the AG free list count. Prior to f8f28 we would put the free block back in the free space btrees in the same transaction, which calls xfs_trans_agblocks_delta() to record the increment in the AG free block count. Since AGFL blocks are included in the global free block count (fdblocks), there is no corresponding fdblocks update, so the AGFL free satisfies the following condition in xfs_trans_apply_sb_deltas: /* * Check that superblock mods match the mods made to AGF counters. */ ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) == (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta + tp->t_ag_btree_delta)); The comparison here used to be: (X + 0) == ((X+1) + -1 + 0), where X is the number blocks that were allocated. After commit f8f28 we defer the block freeing to the next chained transaction, which means that the calls to xfs_trans_agflist_delta and xfs_trans_agblocks_delta occur in separate transactions. The (first) transaction that shortens the free list trips on the comparison, which has now become: (X + 0) == ((X) + -1 + 0) because we haven't freed the AGFL block yet; we've only logged an intention to free it. When the second transaction (the deferred free) commits, it will evaluate the expression as: (0 + 0) == (1 + 0 + 0) and trip over that in turn. At this point, the astute reader may note that the two commits tagged by this patch have been in the kernel for a long time but haven't generated any bug reports. How is it that the author became aware of this bug? This originally surfaced as an intermittent failure when I was testing realtime rmap, but a different bug report by Zorro Lang reveals the same assertion occuring on !lazysbcount filesystems. The common factor to both reports (and why this problem wasn't previously reported) becomes apparent if we consider when xfs_trans_apply_sb_deltas is called by __xfs_trans_commit(): if (tp->t_flags & XFS_TRANS_SB_DIRTY) xfs_trans_apply_sb_deltas(tp); With a modern lazysbcount filesystem, transactions update only the percpu counters, so they don't need to set XFS_TRANS_SB_DIRTY, hence xfs_trans_apply_sb_deltas is rarely called. However, updates to the count of free realtime extents are not part of lazysbcount, so XFS_TRANS_SB_DIRTY will be set on transactions adding or removing data fork mappings to realtime files; similarly, XFS_TRANS_SB_DIRTY is always set on !lazysbcount filesystems. Dave mentioned in response to an earlier version of this patch: "IIUC, what you are saying is that this debug code is simply not exercised in normal testing and hasn't been for the past decade? And it still won't be exercised on anything other than realtime device testing? "...it was debugging code from 1994 that was largely turned into dead code when lazysbcounters were introduced in 2007. Hence I'm not sure it holds any value anymore." This debugging code isn't especially helpful - you can modify the flcount on one AG and the freeblks of another AG, and it won't trigger. Add the fact that nobody noticed for a decade, and let's just get rid of it (and start testing realtime :P). This bug was found by running generic/051 on either a V4 filesystem lacking lazysbcount; or a V5 filesystem with a realtime volume. Cc: bfoster@redhat.com, zlang@redhat.com Fixes: f8f2835a9cf3 ("xfs: defer agfl block frees when dfops is available") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Brian Foster <bfoster@redhat.com> diff 1aec7c3d Fri Apr 23 17:02:00 MDT 2021 Darrick J. Wong <djwong@kernel.org> xfs: remove obsolete AGF counter debugging In commit f8f2835a9cf3 we changed the behavior of XFS to use EFIs to remove blocks from an overfilled AGFL because there were complaints about transaction overruns that stemmed from trying to free multiple blocks in a single transaction. Unfortunately, that commit missed a subtlety in the debug-mode transaction accounting when a realtime volume is attached. If a realtime file undergoes a data fork mapping change such that realtime extents are allocated (or freed) in the same transaction that a data device block is also allocated (or freed), we can trip a debugging assertion. This can happen (for example) if a realtime extent is allocated and it is necessary to reshape the bmbt to hold the new mapping. When we go to allocate a bmbt block from an AG, the first thing the data device block allocator does is ensure that the freelist is the proper length. If the freelist is too long, it will trim the freelist to the proper length. In debug mode, trimming the freelist calls xfs_trans_agflist_delta() to record the decrement in the AG free list count. Prior to f8f28 we would put the free block back in the free space btrees in the same transaction, which calls xfs_trans_agblocks_delta() to record the increment in the AG free block count. Since AGFL blocks are included in the global free block count (fdblocks), there is no corresponding fdblocks update, so the AGFL free satisfies the following condition in xfs_trans_apply_sb_deltas: /* * Check that superblock mods match the mods made to AGF counters. */ ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) == (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta + tp->t_ag_btree_delta)); The comparison here used to be: (X + 0) == ((X+1) + -1 + 0), where X is the number blocks that were allocated. After commit f8f28 we defer the block freeing to the next chained transaction, which means that the calls to xfs_trans_agflist_delta and xfs_trans_agblocks_delta occur in separate transactions. The (first) transaction that shortens the free list trips on the comparison, which has now become: (X + 0) == ((X) + -1 + 0) because we haven't freed the AGFL block yet; we've only logged an intention to free it. When the second transaction (the deferred free) commits, it will evaluate the expression as: (0 + 0) == (1 + 0 + 0) and trip over that in turn. At this point, the astute reader may note that the two commits tagged by this patch have been in the kernel for a long time but haven't generated any bug reports. How is it that the author became aware of this bug? This originally surfaced as an intermittent failure when I was testing realtime rmap, but a different bug report by Zorro Lang reveals the same assertion occuring on !lazysbcount filesystems. The common factor to both reports (and why this problem wasn't previously reported) becomes apparent if we consider when xfs_trans_apply_sb_deltas is called by __xfs_trans_commit(): if (tp->t_flags & XFS_TRANS_SB_DIRTY) xfs_trans_apply_sb_deltas(tp); With a modern lazysbcount filesystem, transactions update only the percpu counters, so they don't need to set XFS_TRANS_SB_DIRTY, hence xfs_trans_apply_sb_deltas is rarely called. However, updates to the count of free realtime extents are not part of lazysbcount, so XFS_TRANS_SB_DIRTY will be set on transactions adding or removing data fork mappings to realtime files; similarly, XFS_TRANS_SB_DIRTY is always set on !lazysbcount filesystems. Dave mentioned in response to an earlier version of this patch: "IIUC, what you are saying is that this debug code is simply not exercised in normal testing and hasn't been for the past decade? And it still won't be exercised on anything other than realtime device testing? "...it was debugging code from 1994 that was largely turned into dead code when lazysbcounters were introduced in 2007. Hence I'm not sure it holds any value anymore." This debugging code isn't especially helpful - you can modify the flcount on one AG and the freeblks of another AG, and it won't trigger. Add the fact that nobody noticed for a decade, and let's just get rid of it (and start testing realtime :P). This bug was found by running generic/051 on either a V4 filesystem lacking lazysbcount; or a V5 filesystem with a realtime volume. Cc: bfoster@redhat.com, zlang@redhat.com Fixes: f8f2835a9cf3 ("xfs: defer agfl block frees when dfops is available") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Brian Foster <bfoster@redhat.com> diff 1aec7c3d Fri Apr 23 17:02:00 MDT 2021 Darrick J. Wong <djwong@kernel.org> xfs: remove obsolete AGF counter debugging In commit f8f2835a9cf3 we changed the behavior of XFS to use EFIs to remove blocks from an overfilled AGFL because there were complaints about transaction overruns that stemmed from trying to free multiple blocks in a single transaction. Unfortunately, that commit missed a subtlety in the debug-mode transaction accounting when a realtime volume is attached. If a realtime file undergoes a data fork mapping change such that realtime extents are allocated (or freed) in the same transaction that a data device block is also allocated (or freed), we can trip a debugging assertion. This can happen (for example) if a realtime extent is allocated and it is necessary to reshape the bmbt to hold the new mapping. When we go to allocate a bmbt block from an AG, the first thing the data device block allocator does is ensure that the freelist is the proper length. If the freelist is too long, it will trim the freelist to the proper length. In debug mode, trimming the freelist calls xfs_trans_agflist_delta() to record the decrement in the AG free list count. Prior to f8f28 we would put the free block back in the free space btrees in the same transaction, which calls xfs_trans_agblocks_delta() to record the increment in the AG free block count. Since AGFL blocks are included in the global free block count (fdblocks), there is no corresponding fdblocks update, so the AGFL free satisfies the following condition in xfs_trans_apply_sb_deltas: /* * Check that superblock mods match the mods made to AGF counters. */ ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) == (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta + tp->t_ag_btree_delta)); The comparison here used to be: (X + 0) == ((X+1) + -1 + 0), where X is the number blocks that were allocated. After commit f8f28 we defer the block freeing to the next chained transaction, which means that the calls to xfs_trans_agflist_delta and xfs_trans_agblocks_delta occur in separate transactions. The (first) transaction that shortens the free list trips on the comparison, which has now become: (X + 0) == ((X) + -1 + 0) because we haven't freed the AGFL block yet; we've only logged an intention to free it. When the second transaction (the deferred free) commits, it will evaluate the expression as: (0 + 0) == (1 + 0 + 0) and trip over that in turn. At this point, the astute reader may note that the two commits tagged by this patch have been in the kernel for a long time but haven't generated any bug reports. How is it that the author became aware of this bug? This originally surfaced as an intermittent failure when I was testing realtime rmap, but a different bug report by Zorro Lang reveals the same assertion occuring on !lazysbcount filesystems. The common factor to both reports (and why this problem wasn't previously reported) becomes apparent if we consider when xfs_trans_apply_sb_deltas is called by __xfs_trans_commit(): if (tp->t_flags & XFS_TRANS_SB_DIRTY) xfs_trans_apply_sb_deltas(tp); With a modern lazysbcount filesystem, transactions update only the percpu counters, so they don't need to set XFS_TRANS_SB_DIRTY, hence xfs_trans_apply_sb_deltas is rarely called. However, updates to the count of free realtime extents are not part of lazysbcount, so XFS_TRANS_SB_DIRTY will be set on transactions adding or removing data fork mappings to realtime files; similarly, XFS_TRANS_SB_DIRTY is always set on !lazysbcount filesystems. Dave mentioned in response to an earlier version of this patch: "IIUC, what you are saying is that this debug code is simply not exercised in normal testing and hasn't been for the past decade? And it still won't be exercised on anything other than realtime device testing? "...it was debugging code from 1994 that was largely turned into dead code when lazysbcounters were introduced in 2007. Hence I'm not sure it holds any value anymore." This debugging code isn't especially helpful - you can modify the flcount on one AG and the freeblks of another AG, and it won't trigger. Add the fact that nobody noticed for a decade, and let's just get rid of it (and start testing realtime :P). This bug was found by running generic/051 on either a V4 filesystem lacking lazysbcount; or a V5 filesystem with a realtime volume. Cc: bfoster@redhat.com, zlang@redhat.com Fixes: f8f2835a9cf3 ("xfs: defer agfl block frees when dfops is available") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Brian Foster <bfoster@redhat.com> diff 1aec7c3d Fri Apr 23 17:02:00 MDT 2021 Darrick J. Wong <djwong@kernel.org> xfs: remove obsolete AGF counter debugging In commit f8f2835a9cf3 we changed the behavior of XFS to use EFIs to remove blocks from an overfilled AGFL because there were complaints about transaction overruns that stemmed from trying to free multiple blocks in a single transaction. Unfortunately, that commit missed a subtlety in the debug-mode transaction accounting when a realtime volume is attached. If a realtime file undergoes a data fork mapping change such that realtime extents are allocated (or freed) in the same transaction that a data device block is also allocated (or freed), we can trip a debugging assertion. This can happen (for example) if a realtime extent is allocated and it is necessary to reshape the bmbt to hold the new mapping. When we go to allocate a bmbt block from an AG, the first thing the data device block allocator does is ensure that the freelist is the proper length. If the freelist is too long, it will trim the freelist to the proper length. In debug mode, trimming the freelist calls xfs_trans_agflist_delta() to record the decrement in the AG free list count. Prior to f8f28 we would put the free block back in the free space btrees in the same transaction, which calls xfs_trans_agblocks_delta() to record the increment in the AG free block count. Since AGFL blocks are included in the global free block count (fdblocks), there is no corresponding fdblocks update, so the AGFL free satisfies the following condition in xfs_trans_apply_sb_deltas: /* * Check that superblock mods match the mods made to AGF counters. */ ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) == (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta + tp->t_ag_btree_delta)); The comparison here used to be: (X + 0) == ((X+1) + -1 + 0), where X is the number blocks that were allocated. After commit f8f28 we defer the block freeing to the next chained transaction, which means that the calls to xfs_trans_agflist_delta and xfs_trans_agblocks_delta occur in separate transactions. The (first) transaction that shortens the free list trips on the comparison, which has now become: (X + 0) == ((X) + -1 + 0) because we haven't freed the AGFL block yet; we've only logged an intention to free it. When the second transaction (the deferred free) commits, it will evaluate the expression as: (0 + 0) == (1 + 0 + 0) and trip over that in turn. At this point, the astute reader may note that the two commits tagged by this patch have been in the kernel for a long time but haven't generated any bug reports. How is it that the author became aware of this bug? This originally surfaced as an intermittent failure when I was testing realtime rmap, but a different bug report by Zorro Lang reveals the same assertion occuring on !lazysbcount filesystems. The common factor to both reports (and why this problem wasn't previously reported) becomes apparent if we consider when xfs_trans_apply_sb_deltas is called by __xfs_trans_commit(): if (tp->t_flags & XFS_TRANS_SB_DIRTY) xfs_trans_apply_sb_deltas(tp); With a modern lazysbcount filesystem, transactions update only the percpu counters, so they don't need to set XFS_TRANS_SB_DIRTY, hence xfs_trans_apply_sb_deltas is rarely called. However, updates to the count of free realtime extents are not part of lazysbcount, so XFS_TRANS_SB_DIRTY will be set on transactions adding or removing data fork mappings to realtime files; similarly, XFS_TRANS_SB_DIRTY is always set on !lazysbcount filesystems. Dave mentioned in response to an earlier version of this patch: "IIUC, what you are saying is that this debug code is simply not exercised in normal testing and hasn't been for the past decade? And it still won't be exercised on anything other than realtime device testing? "...it was debugging code from 1994 that was largely turned into dead code when lazysbcounters were introduced in 2007. Hence I'm not sure it holds any value anymore." This debugging code isn't especially helpful - you can modify the flcount on one AG and the freeblks of another AG, and it won't trigger. Add the fact that nobody noticed for a decade, and let's just get rid of it (and start testing realtime :P). This bug was found by running generic/051 on either a V4 filesystem lacking lazysbcount; or a V5 filesystem with a realtime volume. Cc: bfoster@redhat.com, zlang@redhat.com Fixes: f8f2835a9cf3 ("xfs: defer agfl block frees when dfops is available") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Brian Foster <bfoster@redhat.com> diff 1aec7c3d Fri Apr 23 17:02:00 MDT 2021 Darrick J. Wong <djwong@kernel.org> xfs: remove obsolete AGF counter debugging In commit f8f2835a9cf3 we changed the behavior of XFS to use EFIs to remove blocks from an overfilled AGFL because there were complaints about transaction overruns that stemmed from trying to free multiple blocks in a single transaction. Unfortunately, that commit missed a subtlety in the debug-mode transaction accounting when a realtime volume is attached. If a realtime file undergoes a data fork mapping change such that realtime extents are allocated (or freed) in the same transaction that a data device block is also allocated (or freed), we can trip a debugging assertion. This can happen (for example) if a realtime extent is allocated and it is necessary to reshape the bmbt to hold the new mapping. When we go to allocate a bmbt block from an AG, the first thing the data device block allocator does is ensure that the freelist is the proper length. If the freelist is too long, it will trim the freelist to the proper length. In debug mode, trimming the freelist calls xfs_trans_agflist_delta() to record the decrement in the AG free list count. Prior to f8f28 we would put the free block back in the free space btrees in the same transaction, which calls xfs_trans_agblocks_delta() to record the increment in the AG free block count. Since AGFL blocks are included in the global free block count (fdblocks), there is no corresponding fdblocks update, so the AGFL free satisfies the following condition in xfs_trans_apply_sb_deltas: /* * Check that superblock mods match the mods made to AGF counters. */ ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) == (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta + tp->t_ag_btree_delta)); The comparison here used to be: (X + 0) == ((X+1) + -1 + 0), where X is the number blocks that were allocated. After commit f8f28 we defer the block freeing to the next chained transaction, which means that the calls to xfs_trans_agflist_delta and xfs_trans_agblocks_delta occur in separate transactions. The (first) transaction that shortens the free list trips on the comparison, which has now become: (X + 0) == ((X) + -1 + 0) because we haven't freed the AGFL block yet; we've only logged an intention to free it. When the second transaction (the deferred free) commits, it will evaluate the expression as: (0 + 0) == (1 + 0 + 0) and trip over that in turn. At this point, the astute reader may note that the two commits tagged by this patch have been in the kernel for a long time but haven't generated any bug reports. How is it that the author became aware of this bug? This originally surfaced as an intermittent failure when I was testing realtime rmap, but a different bug report by Zorro Lang reveals the same assertion occuring on !lazysbcount filesystems. The common factor to both reports (and why this problem wasn't previously reported) becomes apparent if we consider when xfs_trans_apply_sb_deltas is called by __xfs_trans_commit(): if (tp->t_flags & XFS_TRANS_SB_DIRTY) xfs_trans_apply_sb_deltas(tp); With a modern lazysbcount filesystem, transactions update only the percpu counters, so they don't need to set XFS_TRANS_SB_DIRTY, hence xfs_trans_apply_sb_deltas is rarely called. However, updates to the count of free realtime extents are not part of lazysbcount, so XFS_TRANS_SB_DIRTY will be set on transactions adding or removing data fork mappings to realtime files; similarly, XFS_TRANS_SB_DIRTY is always set on !lazysbcount filesystems. Dave mentioned in response to an earlier version of this patch: "IIUC, what you are saying is that this debug code is simply not exercised in normal testing and hasn't been for the past decade? And it still won't be exercised on anything other than realtime device testing? "...it was debugging code from 1994 that was largely turned into dead code when lazysbcounters were introduced in 2007. Hence I'm not sure it holds any value anymore." This debugging code isn't especially helpful - you can modify the flcount on one AG and the freeblks of another AG, and it won't trigger. Add the fact that nobody noticed for a decade, and let's just get rid of it (and start testing realtime :P). This bug was found by running generic/051 on either a V4 filesystem lacking lazysbcount; or a V5 filesystem with a realtime volume. Cc: bfoster@redhat.com, zlang@redhat.com Fixes: f8f2835a9cf3 ("xfs: defer agfl block frees when dfops is available") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Brian Foster <bfoster@redhat.com> diff 1aec7c3d Fri Apr 23 17:02:00 MDT 2021 Darrick J. Wong <djwong@kernel.org> xfs: remove obsolete AGF counter debugging In commit f8f2835a9cf3 we changed the behavior of XFS to use EFIs to remove blocks from an overfilled AGFL because there were complaints about transaction overruns that stemmed from trying to free multiple blocks in a single transaction. Unfortunately, that commit missed a subtlety in the debug-mode transaction accounting when a realtime volume is attached. If a realtime file undergoes a data fork mapping change such that realtime extents are allocated (or freed) in the same transaction that a data device block is also allocated (or freed), we can trip a debugging assertion. This can happen (for example) if a realtime extent is allocated and it is necessary to reshape the bmbt to hold the new mapping. When we go to allocate a bmbt block from an AG, the first thing the data device block allocator does is ensure that the freelist is the proper length. If the freelist is too long, it will trim the freelist to the proper length. In debug mode, trimming the freelist calls xfs_trans_agflist_delta() to record the decrement in the AG free list count. Prior to f8f28 we would put the free block back in the free space btrees in the same transaction, which calls xfs_trans_agblocks_delta() to record the increment in the AG free block count. Since AGFL blocks are included in the global free block count (fdblocks), there is no corresponding fdblocks update, so the AGFL free satisfies the following condition in xfs_trans_apply_sb_deltas: /* * Check that superblock mods match the mods made to AGF counters. */ ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) == (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta + tp->t_ag_btree_delta)); The comparison here used to be: (X + 0) == ((X+1) + -1 + 0), where X is the number blocks that were allocated. After commit f8f28 we defer the block freeing to the next chained transaction, which means that the calls to xfs_trans_agflist_delta and xfs_trans_agblocks_delta occur in separate transactions. The (first) transaction that shortens the free list trips on the comparison, which has now become: (X + 0) == ((X) + -1 + 0) because we haven't freed the AGFL block yet; we've only logged an intention to free it. When the second transaction (the deferred free) commits, it will evaluate the expression as: (0 + 0) == (1 + 0 + 0) and trip over that in turn. At this point, the astute reader may note that the two commits tagged by this patch have been in the kernel for a long time but haven't generated any bug reports. How is it that the author became aware of this bug? This originally surfaced as an intermittent failure when I was testing realtime rmap, but a different bug report by Zorro Lang reveals the same assertion occuring on !lazysbcount filesystems. The common factor to both reports (and why this problem wasn't previously reported) becomes apparent if we consider when xfs_trans_apply_sb_deltas is called by __xfs_trans_commit(): if (tp->t_flags & XFS_TRANS_SB_DIRTY) xfs_trans_apply_sb_deltas(tp); With a modern lazysbcount filesystem, transactions update only the percpu counters, so they don't need to set XFS_TRANS_SB_DIRTY, hence xfs_trans_apply_sb_deltas is rarely called. However, updates to the count of free realtime extents are not part of lazysbcount, so XFS_TRANS_SB_DIRTY will be set on transactions adding or removing data fork mappings to realtime files; similarly, XFS_TRANS_SB_DIRTY is always set on !lazysbcount filesystems. Dave mentioned in response to an earlier version of this patch: "IIUC, what you are saying is that this debug code is simply not exercised in normal testing and hasn't been for the past decade? And it still won't be exercised on anything other than realtime device testing? "...it was debugging code from 1994 that was largely turned into dead code when lazysbcounters were introduced in 2007. Hence I'm not sure it holds any value anymore." This debugging code isn't especially helpful - you can modify the flcount on one AG and the freeblks of another AG, and it won't trigger. Add the fact that nobody noticed for a decade, and let's just get rid of it (and start testing realtime :P). This bug was found by running generic/051 on either a V4 filesystem lacking lazysbcount; or a V5 filesystem with a realtime volume. Cc: bfoster@redhat.com, zlang@redhat.com Fixes: f8f2835a9cf3 ("xfs: defer agfl block frees when dfops is available") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Brian Foster <bfoster@redhat.com> diff 1aec7c3d Fri Apr 23 17:02:00 MDT 2021 Darrick J. Wong <djwong@kernel.org> xfs: remove obsolete AGF counter debugging In commit f8f2835a9cf3 we changed the behavior of XFS to use EFIs to remove blocks from an overfilled AGFL because there were complaints about transaction overruns that stemmed from trying to free multiple blocks in a single transaction. Unfortunately, that commit missed a subtlety in the debug-mode transaction accounting when a realtime volume is attached. If a realtime file undergoes a data fork mapping change such that realtime extents are allocated (or freed) in the same transaction that a data device block is also allocated (or freed), we can trip a debugging assertion. This can happen (for example) if a realtime extent is allocated and it is necessary to reshape the bmbt to hold the new mapping. When we go to allocate a bmbt block from an AG, the first thing the data device block allocator does is ensure that the freelist is the proper length. If the freelist is too long, it will trim the freelist to the proper length. In debug mode, trimming the freelist calls xfs_trans_agflist_delta() to record the decrement in the AG free list count. Prior to f8f28 we would put the free block back in the free space btrees in the same transaction, which calls xfs_trans_agblocks_delta() to record the increment in the AG free block count. Since AGFL blocks are included in the global free block count (fdblocks), there is no corresponding fdblocks update, so the AGFL free satisfies the following condition in xfs_trans_apply_sb_deltas: /* * Check that superblock mods match the mods made to AGF counters. */ ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) == (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta + tp->t_ag_btree_delta)); The comparison here used to be: (X + 0) == ((X+1) + -1 + 0), where X is the number blocks that were allocated. After commit f8f28 we defer the block freeing to the next chained transaction, which means that the calls to xfs_trans_agflist_delta and xfs_trans_agblocks_delta occur in separate transactions. The (first) transaction that shortens the free list trips on the comparison, which has now become: (X + 0) == ((X) + -1 + 0) because we haven't freed the AGFL block yet; we've only logged an intention to free it. When the second transaction (the deferred free) commits, it will evaluate the expression as: (0 + 0) == (1 + 0 + 0) and trip over that in turn. At this point, the astute reader may note that the two commits tagged by this patch have been in the kernel for a long time but haven't generated any bug reports. How is it that the author became aware of this bug? This originally surfaced as an intermittent failure when I was testing realtime rmap, but a different bug report by Zorro Lang reveals the same assertion occuring on !lazysbcount filesystems. The common factor to both reports (and why this problem wasn't previously reported) becomes apparent if we consider when xfs_trans_apply_sb_deltas is called by __xfs_trans_commit(): if (tp->t_flags & XFS_TRANS_SB_DIRTY) xfs_trans_apply_sb_deltas(tp); With a modern lazysbcount filesystem, transactions update only the percpu counters, so they don't need to set XFS_TRANS_SB_DIRTY, hence xfs_trans_apply_sb_deltas is rarely called. However, updates to the count of free realtime extents are not part of lazysbcount, so XFS_TRANS_SB_DIRTY will be set on transactions adding or removing data fork mappings to realtime files; similarly, XFS_TRANS_SB_DIRTY is always set on !lazysbcount filesystems. Dave mentioned in response to an earlier version of this patch: "IIUC, what you are saying is that this debug code is simply not exercised in normal testing and hasn't been for the past decade? And it still won't be exercised on anything other than realtime device testing? "...it was debugging code from 1994 that was largely turned into dead code when lazysbcounters were introduced in 2007. Hence I'm not sure it holds any value anymore." This debugging code isn't especially helpful - you can modify the flcount on one AG and the freeblks of another AG, and it won't trigger. Add the fact that nobody noticed for a decade, and let's just get rid of it (and start testing realtime :P). This bug was found by running generic/051 on either a V4 filesystem lacking lazysbcount; or a V5 filesystem with a realtime volume. Cc: bfoster@redhat.com, zlang@redhat.com Fixes: f8f2835a9cf3 ("xfs: defer agfl block frees when dfops is available") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Brian Foster <bfoster@redhat.com> |
H A D | xfs_icache.c | diff f12b9668 Mon Sep 11 09:39:07 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: use i_prev_unlinked to distinguish inodes that are not on the unlinked list Alter the definition of i_prev_unlinked slightly to make it more obvious when an inode with 0 link count is not part of the iunlink bucket lists rooted in the AGI. This distinction is necessary because it is not sufficient to check inode.i_nlink to decide if an inode is on the unlinked list. Updates to i_nlink can happen while holding only ILOCK_EXCL, but updates to an inode's position in the AGI unlinked list (which happen after the nlink update) requires both ILOCK_EXCL and the AGI buffer lock. The next few patches will make it possible to reload an entire unlinked bucket list when we're walking the inode table or performing handle operations and need more than the ability to iget the last inode in the chain. The upcoming directory repair code also needs to be able to make this distinction to decide if a zero link count directory should be moved to the orphanage or allowed to inactivate. An upcoming enhancement to the online AGI fsck code will need this distinction to check and rebuild the AGI unlinked buckets. Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0d296634 Thu Aug 10 08:48:12 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: hide xfs_inode_is_allocated in scrub common code This function is only used by online fsck, so let's move it there. In the next patch, we'll fix it to work properly and to require that the caller hold the AGI buffer locked. No major changes aside from adjusting the signature a bit. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 2254a739 Mon May 01 17:16:14 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: fix xfs_inodegc_stop racing with mod_delayed_work syzbot reported this warning from the faux inodegc shrinker that tries to kick off inodegc work: ------------[ cut here ]------------ WARNING: CPU: 1 PID: 102 at kernel/workqueue.c:1445 __queue_work+0xd44/0x1120 kernel/workqueue.c:1444 RIP: 0010:__queue_work+0xd44/0x1120 kernel/workqueue.c:1444 Call Trace: __queue_delayed_work+0x1c8/0x270 kernel/workqueue.c:1672 mod_delayed_work_on+0xe1/0x220 kernel/workqueue.c:1746 xfs_inodegc_shrinker_scan fs/xfs/xfs_icache.c:2212 [inline] xfs_inodegc_shrinker_scan+0x250/0x4f0 fs/xfs/xfs_icache.c:2191 do_shrink_slab+0x428/0xaa0 mm/vmscan.c:853 shrink_slab+0x175/0x660 mm/vmscan.c:1013 shrink_one+0x502/0x810 mm/vmscan.c:5343 shrink_many mm/vmscan.c:5394 [inline] lru_gen_shrink_node mm/vmscan.c:5511 [inline] shrink_node+0x2064/0x35f0 mm/vmscan.c:6459 kswapd_shrink_node mm/vmscan.c:7262 [inline] balance_pgdat+0xa02/0x1ac0 mm/vmscan.c:7452 kswapd+0x677/0xd60 mm/vmscan.c:7712 kthread+0x2e8/0x3a0 kernel/kthread.c:376 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308 This warning corresponds to this code in __queue_work: /* * For a draining wq, only works from the same workqueue are * allowed. The __WQ_DESTROYING helps to spot the issue that * queues a new work item to a wq after destroy_workqueue(wq). */ if (unlikely(wq->flags & (__WQ_DESTROYING | __WQ_DRAINING) && WARN_ON_ONCE(!is_chained_work(wq)))) return; For this to trip, we must have a thread draining the inodedgc workqueue and a second thread trying to queue inodegc work to that workqueue. This can happen if freezing or a ro remount race with reclaim poking our faux inodegc shrinker and another thread dropping an unlinked O_RDONLY file: Thread 0 Thread 1 Thread 2 xfs_inodegc_stop xfs_inodegc_shrinker_scan xfs_is_inodegc_enabled <yes, will continue> xfs_clear_inodegc_enabled xfs_inodegc_queue_all <list empty, do not queue inodegc worker> xfs_inodegc_queue <add to list> xfs_is_inodegc_enabled <no, returns> drain_workqueue <set WQ_DRAINING> llist_empty <no, will queue list> mod_delayed_work_on(..., 0) __queue_work <sees WQ_DRAINING, kaboom> In other words, everything between the access to inodegc_enabled state and the decision to poke the inodegc workqueue requires some kind of coordination to avoid the WQ_DRAINING state. We could perhaps introduce a lock here, but we could also try to eliminate WQ_DRAINING from the picture. We could replace the drain_workqueue call with a loop that flushes the workqueue and queues workers as long as there is at least one inode present in the per-cpu inodegc llists. We've disabled inodegc at this point, so we know that the number of queued inodes will eventually hit zero as long as xfs_inodegc_start cannot reactivate the workers. There are four callers of xfs_inodegc_start. Three of them come from the VFS with s_umount held: filesystem thawing, failed filesystem freezing, and the rw remount transition. The fourth caller is mounting rw (no remount or freezing possible). There are three callers ofs xfs_inodegc_stop. One is unmounting (no remount or thaw possible). Two of them come from the VFS with s_umount held: fs freezing and ro remount transition. Hence, it is correct to replace the drain_workqueue call with a loop that drains the inodegc llists. Fixes: 6191cf3ad59f ("xfs: flush inodegc workqueue tasks before cancel") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com> |
H A D | xfs_aops.c | diff 4ce02c67 Mon Jul 10 15:12:43 MDT 2023 Ritesh Harjani (IBM) <ritesh.list@gmail.com> iomap: Add per-block dirty state tracking to improve performance When filesystem blocksize is less than folio size (either with mapping_large_folio_support() or with blocksize < pagesize) and when the folio is uptodate in pagecache, then even a byte write can cause an entire folio to be written to disk during writeback. This happens because we currently don't have a mechanism to track per-block dirty state within struct iomap_folio_state. We currently only track uptodate state. This patch implements support for tracking per-block dirty state in iomap_folio_state->state bitmap. This should help improve the filesystem write performance and help reduce write amplification. Performance testing of below fio workload reveals ~16x performance improvement using nvme with XFS (4k blocksize) on Power (64K pagesize) FIO reported write bw scores improved from around ~28 MBps to ~452 MBps. 1. <test_randwrite.fio> [global] ioengine=psync rw=randwrite overwrite=1 pre_read=1 direct=0 bs=4k size=1G dir=./ numjobs=8 fdatasync=1 runtime=60 iodepth=64 group_reporting=1 [fio-run] 2. Also our internal performance team reported that this patch improves their database workload performance by around ~83% (with XFS on Power) Reported-by: Aravinda Herle <araherle@in.ibm.com> Reported-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> diff 8ac5b996 Sun Mar 05 16:13:23 MST 2023 Dave Chinner <dchinner@redhat.com> xfs: fix off-by-one-block in xfs_discard_folio() The recent writeback corruption fixes changed the code in xfs_discard_folio() to calculate a byte range to for punching delalloc extents. A mistake was made in using round_up(pos) for the end offset, because when pos points at the first byte of a block, it does not get rounded up to point to the end byte of the block. hence the punch range is short, and this leads to unexpected behaviour in certain cases in xfs_bmap_punch_delalloc_range. e.g. pos = 0 means we call xfs_bmap_punch_delalloc_range(0,0), so there is no previous extent and it rounds up the punch to the end of the delalloc extent it found at offset 0, not the end of the range given to xfs_bmap_punch_delalloc_range(). Fix this by handling the zero block offset case correctly. Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=217030 Link: https://lore.kernel.org/linux-xfs/Y+vOfaxIWX1c%2Fyy9@bfoster/ Fixes: 7348b322332d ("xfs: xfs_bmap_punch_delalloc_range() should take a byte range") Reported-by: Pengfei Xu <pengfei.xu@intel.com> Found-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 8ac5b996 Sun Mar 05 16:13:23 MST 2023 Dave Chinner <dchinner@redhat.com> xfs: fix off-by-one-block in xfs_discard_folio() The recent writeback corruption fixes changed the code in xfs_discard_folio() to calculate a byte range to for punching delalloc extents. A mistake was made in using round_up(pos) for the end offset, because when pos points at the first byte of a block, it does not get rounded up to point to the end byte of the block. hence the punch range is short, and this leads to unexpected behaviour in certain cases in xfs_bmap_punch_delalloc_range. e.g. pos = 0 means we call xfs_bmap_punch_delalloc_range(0,0), so there is no previous extent and it rounds up the punch to the end of the delalloc extent it found at offset 0, not the end of the range given to xfs_bmap_punch_delalloc_range(). Fix this by handling the zero block offset case correctly. Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=217030 Link: https://lore.kernel.org/linux-xfs/Y+vOfaxIWX1c%2Fyy9@bfoster/ Fixes: 7348b322332d ("xfs: xfs_bmap_punch_delalloc_range() should take a byte range") Reported-by: Pengfei Xu <pengfei.xu@intel.com> Found-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 8ac5b996 Sun Mar 05 16:13:23 MST 2023 Dave Chinner <dchinner@redhat.com> xfs: fix off-by-one-block in xfs_discard_folio() The recent writeback corruption fixes changed the code in xfs_discard_folio() to calculate a byte range to for punching delalloc extents. A mistake was made in using round_up(pos) for the end offset, because when pos points at the first byte of a block, it does not get rounded up to point to the end byte of the block. hence the punch range is short, and this leads to unexpected behaviour in certain cases in xfs_bmap_punch_delalloc_range. e.g. pos = 0 means we call xfs_bmap_punch_delalloc_range(0,0), so there is no previous extent and it rounds up the punch to the end of the delalloc extent it found at offset 0, not the end of the range given to xfs_bmap_punch_delalloc_range(). Fix this by handling the zero block offset case correctly. Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=217030 Link: https://lore.kernel.org/linux-xfs/Y+vOfaxIWX1c%2Fyy9@bfoster/ Fixes: 7348b322332d ("xfs: xfs_bmap_punch_delalloc_range() should take a byte range") Reported-by: Pengfei Xu <pengfei.xu@intel.com> Found-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 8ac5b996 Sun Mar 05 16:13:23 MST 2023 Dave Chinner <dchinner@redhat.com> xfs: fix off-by-one-block in xfs_discard_folio() The recent writeback corruption fixes changed the code in xfs_discard_folio() to calculate a byte range to for punching delalloc extents. A mistake was made in using round_up(pos) for the end offset, because when pos points at the first byte of a block, it does not get rounded up to point to the end byte of the block. hence the punch range is short, and this leads to unexpected behaviour in certain cases in xfs_bmap_punch_delalloc_range. e.g. pos = 0 means we call xfs_bmap_punch_delalloc_range(0,0), so there is no previous extent and it rounds up the punch to the end of the delalloc extent it found at offset 0, not the end of the range given to xfs_bmap_punch_delalloc_range(). Fix this by handling the zero block offset case correctly. Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=217030 Link: https://lore.kernel.org/linux-xfs/Y+vOfaxIWX1c%2Fyy9@bfoster/ Fixes: 7348b322332d ("xfs: xfs_bmap_punch_delalloc_range() should take a byte range") Reported-by: Pengfei Xu <pengfei.xu@intel.com> Found-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff ebb7fb15 Wed Jan 26 10:19:20 MST 2022 Dave Chinner <dchinner@redhat.com> xfs, iomap: limit individual ioend chain lengths in writeback Trond Myklebust reported soft lockups in XFS IO completion such as this: watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [kworker/12:1:3106] CPU: 12 PID: 3106 Comm: kworker/12:1 Not tainted 4.18.0-305.10.2.el8_4.x86_64 #1 Workqueue: xfs-conv/md127 xfs_end_io [xfs] RIP: 0010:_raw_spin_unlock_irqrestore+0x11/0x20 Call Trace: wake_up_page_bit+0x8a/0x110 iomap_finish_ioend+0xd7/0x1c0 iomap_finish_ioends+0x7f/0xb0 xfs_end_ioend+0x6b/0x100 [xfs] xfs_end_io+0xb9/0xe0 [xfs] process_one_work+0x1a7/0x360 worker_thread+0x1fa/0x390 kthread+0x116/0x130 ret_from_fork+0x35/0x40 Ioends are processed as an atomic completion unit when all the chained bios in the ioend have completed their IO. Logically contiguous ioends can also be merged and completed as a single, larger unit. Both of these things can be problematic as both the bio chains per ioend and the size of the merged ioends processed as a single completion are both unbound. If we have a large sequential dirty region in the page cache, write_cache_pages() will keep feeding us sequential pages and we will keep mapping them into ioends and bios until we get a dirty page at a non-sequential file offset. These large sequential runs can will result in bio and ioend chaining to optimise the io patterns. The pages iunder writeback are pinned within these chains until the submission chaining is broken, allowing the entire chain to be completed. This can result in huge chains being processed in IO completion context. We get deep bio chaining if we have large contiguous physical extents. We will keep adding pages to the current bio until it is full, then we'll chain a new bio to keep adding pages for writeback. Hence we can build bio chains that map millions of pages and tens of gigabytes of RAM if the page cache contains big enough contiguous dirty file regions. This long bio chain pins those pages until the final bio in the chain completes and the ioend can iterate all the chained bios and complete them. OTOH, if we have a physically fragmented file, we end up submitting one ioend per physical fragment that each have a small bio or bio chain attached to them. We do not chain these at IO submission time, but instead we chain them at completion time based on file offset via iomap_ioend_try_merge(). Hence we can end up with unbound ioend chains being built via completion merging. XFS can then do COW remapping or unwritten extent conversion on that merged chain, which involves walking an extent fragment at a time and running a transaction to modify the physical extent information. IOWs, we merge all the discontiguous ioends together into a contiguous file range, only to then process them individually as discontiguous extents. This extent manipulation is computationally expensive and can run in a tight loop, so merging logically contiguous but physically discontigous ioends gains us nothing except for hiding the fact the fact we broke the ioends up into individual physical extents at submission and then need to loop over those individual physical extents at completion. Hence we need to have mechanisms to limit ioend sizes and to break up completion processing of large merged ioend chains: 1. bio chains per ioend need to be bound in length. Pure overwrites go straight to iomap_finish_ioend() in softirq context with the exact bio chain attached to the ioend by submission. Hence the only way to prevent long holdoffs here is to bound ioend submission sizes because we can't reschedule in softirq context. 2. iomap_finish_ioends() has to handle unbound merged ioend chains correctly. This relies on any one call to iomap_finish_ioend() being bound in runtime so that cond_resched() can be issued regularly as the long ioend chain is processed. i.e. this relies on mechanism #1 to limit individual ioend sizes to work correctly. 3. filesystems have to loop over the merged ioends to process physical extent manipulations. This means they can loop internally, and so we break merging at physical extent boundaries so the filesystem can easily insert reschedule points between individual extent manipulations. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reported-and-tested-by: Trond Myklebust <trondmy@hammerspace.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> |
H A D | xfs_trace.h | diff 0dc63c8a Thu Feb 22 01:43:36 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: launder in-memory btree buffers before transaction commit As we've noted in various places, all current users of in-memory btrees are online fsck. Online fsck only stages a btree long enough to rebuild an ondisk data structure, which means that the in-memory btree is ephemeral. Furthermore, if we encounter /any/ errors while updating an in-memory btree, all we do is tear down all the staged data and return an errno to userspace. In-memory btrees need not be transactional, so their buffers should not be committed to the ondisk log, nor should they be checkpointed by the AIL. That's just as well since the ephemeral nature of the btree means that the buftarg and the buffers may disappear quickly anyway. Therefore, we need a way to launder the btree buffers that get attached to the transaction by the generic btree code. Because the buffers are directly mapped to backing file pages, there's no need to bwrite them back to the tmpfs file. All we need to do is clean enough of the buffer log item state so that the bli can be detached from the buffer, remove the bli from the transaction's log item list, and reset the transaction dirty state as if the laundered items had never been there. For simplicity, create xfbtree transaction commit and cancel helpers that launder the in-memory btree buffers for callers. Once laundered, call the write verifier on non-stale buffers to avoid integrity issues, or punch a hole in the backing file for stale buffers. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 0e24ec3c Thu Feb 22 01:33:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: remember sick inodes that get inactivated If an unhealthy inode gets inactivated, remember this fact in the per-fs health summary. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 0b8686f1 Thu Feb 22 01:31:01 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: separate the marking of sick and checked metadata Split the setting of the sick and checked masks into separate functions as part of preparing to add the ability for regular runtime fs code (i.e. not scrub) to mark metadata structures sick when corruptions are found. Improve the documentation of libxfs' requirements for helper behavior. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 68b957f6 Mon Sep 11 09:39:06 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: load uncached unlinked inodes into memory on demand shrikanth hegde reports that filesystems fail shortly after mount with the following failure: WARNING: CPU: 56 PID: 12450 at fs/xfs/xfs_inode.c:1839 xfs_iunlink_lookup+0x58/0x80 [xfs] This of course is the WARN_ON_ONCE in xfs_iunlink_lookup: ip = radix_tree_lookup(&pag->pag_ici_root, agino); if (WARN_ON_ONCE(!ip || !ip->i_ino)) { ... } From diagnostic data collected by the bug reporters, it would appear that we cleanly mounted a filesystem that contained unlinked inodes. Unlinked inodes are only processed as a final step of log recovery, which means that clean mounts do not process the unlinked list at all. Prior to the introduction of the incore unlinked lists, this wasn't a problem because the unlink code would (very expensively) traverse the entire ondisk metadata iunlink chain to keep things up to date. However, the incore unlinked list code complains when it realizes that it is out of sync with the ondisk metadata and shuts down the fs, which is bad. Ritesh proposed to solve this problem by unconditionally parsing the unlinked lists at mount time, but this imposes a mount time cost for every filesystem to catch something that should be very infrequent. Instead, let's target the places where we can encounter a next_unlinked pointer that refers to an inode that is not in cache, and load it into cache. Note: This patch does not address the problem of iget loading an inode from the middle of the iunlink list and needing to set i_prev_unlinked correctly. Reported-by: shrikanth hegde <sshegde@linux.vnet.ibm.com> Triaged-by: Ritesh Harjani <ritesh.list@gmail.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> diff 68b957f6 Mon Sep 11 09:39:06 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: load uncached unlinked inodes into memory on demand shrikanth hegde reports that filesystems fail shortly after mount with the following failure: WARNING: CPU: 56 PID: 12450 at fs/xfs/xfs_inode.c:1839 xfs_iunlink_lookup+0x58/0x80 [xfs] This of course is the WARN_ON_ONCE in xfs_iunlink_lookup: ip = radix_tree_lookup(&pag->pag_ici_root, agino); if (WARN_ON_ONCE(!ip || !ip->i_ino)) { ... } From diagnostic data collected by the bug reporters, it would appear that we cleanly mounted a filesystem that contained unlinked inodes. Unlinked inodes are only processed as a final step of log recovery, which means that clean mounts do not process the unlinked list at all. Prior to the introduction of the incore unlinked lists, this wasn't a problem because the unlink code would (very expensively) traverse the entire ondisk metadata iunlink chain to keep things up to date. However, the incore unlinked list code complains when it realizes that it is out of sync with the ondisk metadata and shuts down the fs, which is bad. Ritesh proposed to solve this problem by unconditionally parsing the unlinked lists at mount time, but this imposes a mount time cost for every filesystem to catch something that should be very infrequent. Instead, let's target the places where we can encounter a next_unlinked pointer that refers to an inode that is not in cache, and load it into cache. Note: This patch does not address the problem of iget loading an inode from the middle of the iunlink list and needing to set i_prev_unlinked correctly. Reported-by: shrikanth hegde <sshegde@linux.vnet.ibm.com> Triaged-by: Ritesh Harjani <ritesh.list@gmail.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> diff f045dd00 Thu Jun 29 18:39:44 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: clean up the rtbitmap fsmap backend The rtbitmap fsmap backend doesn't query the rmapbt, so it's wasteful to spend time initializing the rmap_irec objects. Worse yet, the logic to query the rtbitmap is spread across three separate functions, which is unnecessarily difficult to follow. Compute the start rtextent that we want from keys[0] directly and combine the functions to avoid passing parameters around everywhere, and consolidate all the logic into a single function. At one point many years ago I intended to use __xfs_getfsmap_rtdev as the launching point for realtime rmapbt queries, but this hasn't been the case for a long time. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> diff 0b11553e Wed Feb 01 11:16:04 MST 2023 Darrick J. Wong <djwong@kernel.org> xfs: pass refcount intent directly through the log intent code Pass the incore refcount intent through the CUI logging code instead of repeatedly boxing and unboxing parameters. Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 8838dafe Sun Sep 18 14:51:14 MDT 2022 Zeng Heng <zengheng4@huawei.com> xfs: missing space in xfs trace log Add space between arguments would help someone to locate the key words they want, so break quoted strings at a space character. Such as below: [Before] kworker/1:0-280 [001] ..... 600.782135: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffffflags ATTRFORK ... [After] kworker/1:2-564 [001] ..... 23817.906160: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffff flags ATTRFORK ... Signed-off-by: Zeng Heng <zengheng4@huawei.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 8838dafe Sun Sep 18 14:51:14 MDT 2022 Zeng Heng <zengheng4@huawei.com> xfs: missing space in xfs trace log Add space between arguments would help someone to locate the key words they want, so break quoted strings at a space character. Such as below: [Before] kworker/1:0-280 [001] ..... 600.782135: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffffflags ATTRFORK ... [After] kworker/1:2-564 [001] ..... 23817.906160: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffff flags ATTRFORK ... Signed-off-by: Zeng Heng <zengheng4@huawei.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 8838dafe Sun Sep 18 14:51:14 MDT 2022 Zeng Heng <zengheng4@huawei.com> xfs: missing space in xfs trace log Add space between arguments would help someone to locate the key words they want, so break quoted strings at a space character. Such as below: [Before] kworker/1:0-280 [001] ..... 600.782135: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffffflags ATTRFORK ... [After] kworker/1:2-564 [001] ..... 23817.906160: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffff flags ATTRFORK ... Signed-off-by: Zeng Heng <zengheng4@huawei.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 8838dafe Sun Sep 18 14:51:14 MDT 2022 Zeng Heng <zengheng4@huawei.com> xfs: missing space in xfs trace log Add space between arguments would help someone to locate the key words they want, so break quoted strings at a space character. Such as below: [Before] kworker/1:0-280 [001] ..... 600.782135: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffffflags ATTRFORK ... [After] kworker/1:2-564 [001] ..... 23817.906160: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffff flags ATTRFORK ... Signed-off-by: Zeng Heng <zengheng4@huawei.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 8838dafe Sun Sep 18 14:51:14 MDT 2022 Zeng Heng <zengheng4@huawei.com> xfs: missing space in xfs trace log Add space between arguments would help someone to locate the key words they want, so break quoted strings at a space character. Such as below: [Before] kworker/1:0-280 [001] ..... 600.782135: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffffflags ATTRFORK ... [After] kworker/1:2-564 [001] ..... 23817.906160: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffff flags ATTRFORK ... Signed-off-by: Zeng Heng <zengheng4@huawei.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 8838dafe Sun Sep 18 14:51:14 MDT 2022 Zeng Heng <zengheng4@huawei.com> xfs: missing space in xfs trace log Add space between arguments would help someone to locate the key words they want, so break quoted strings at a space character. Such as below: [Before] kworker/1:0-280 [001] ..... 600.782135: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffffflags ATTRFORK ... [After] kworker/1:2-564 [001] ..... 23817.906160: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffff flags ATTRFORK ... Signed-off-by: Zeng Heng <zengheng4@huawei.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 8838dafe Sun Sep 18 14:51:14 MDT 2022 Zeng Heng <zengheng4@huawei.com> xfs: missing space in xfs trace log Add space between arguments would help someone to locate the key words they want, so break quoted strings at a space character. Such as below: [Before] kworker/1:0-280 [001] ..... 600.782135: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffffflags ATTRFORK ... [After] kworker/1:2-564 [001] ..... 23817.906160: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffff flags ATTRFORK ... Signed-off-by: Zeng Heng <zengheng4@huawei.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 8838dafe Sun Sep 18 14:51:14 MDT 2022 Zeng Heng <zengheng4@huawei.com> xfs: missing space in xfs trace log Add space between arguments would help someone to locate the key words they want, so break quoted strings at a space character. Such as below: [Before] kworker/1:0-280 [001] ..... 600.782135: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffffflags ATTRFORK ... [After] kworker/1:2-564 [001] ..... 23817.906160: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffff flags ATTRFORK ... Signed-off-by: Zeng Heng <zengheng4@huawei.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 8838dafe Sun Sep 18 14:51:14 MDT 2022 Zeng Heng <zengheng4@huawei.com> xfs: missing space in xfs trace log Add space between arguments would help someone to locate the key words they want, so break quoted strings at a space character. Such as below: [Before] kworker/1:0-280 [001] ..... 600.782135: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffffflags ATTRFORK ... [After] kworker/1:2-564 [001] ..... 23817.906160: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffff flags ATTRFORK ... Signed-off-by: Zeng Heng <zengheng4@huawei.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 8838dafe Sun Sep 18 14:51:14 MDT 2022 Zeng Heng <zengheng4@huawei.com> xfs: missing space in xfs trace log Add space between arguments would help someone to locate the key words they want, so break quoted strings at a space character. Such as below: [Before] kworker/1:0-280 [001] ..... 600.782135: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffffflags ATTRFORK ... [After] kworker/1:2-564 [001] ..... 23817.906160: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffff flags ATTRFORK ... Signed-off-by: Zeng Heng <zengheng4@huawei.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com> diff 8838dafe Sun Sep 18 14:51:14 MDT 2022 Zeng Heng <zengheng4@huawei.com> xfs: missing space in xfs trace log Add space between arguments would help someone to locate the key words they want, so break quoted strings at a space character. Such as below: [Before] kworker/1:0-280 [001] ..... 600.782135: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffffflags ATTRFORK ... [After] kworker/1:2-564 [001] ..... 23817.906160: xfs_bunmap: dev 7:0 ino 0x85 disize 0x0 fileoff 0x0 fsbcount 0x400000001fffff flags ATTRFORK ... Signed-off-by: Zeng Heng <zengheng4@huawei.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com> |
H A D | xfs_dquot.c | diff 0c6ca06a Thu Mar 14 07:16:39 MDT 2024 Dave Chinner <dchinner@redhat.com> xfs: quota radix tree allocations need to be NOFS on insert In converting the XFS code from GFP_NOFS to scoped contexts, we converted the quota radix tree to GFP_KERNEL. Unfortunately, it was not clearly documented that this set was because there is a dependency on the quotainfo->qi_tree_lock being taken in memory reclaim to remove dquots from the radix tree. In hindsight this is obvious, but the radix tree allocations on insert are not immediately obvious, and we avoid this for the inode cache radix trees by using preloading and hence completely avoiding the radix tree node allocation under tree lock constraints. Hence there are a few solutions here. The first is to reinstate GFP_NOFS for the radix tree and add a comment explaining why GFP_NOFS is used. The second is to use memalloc_nofs_save() on the radix tree insert context, which makes it obvious that the radix tree insert runs under GFP_NOFS constraints. The third option is to simply replace the radix tree and it's lock with an xarray which can do memory allocation safely in an insert context. The first is OK, but not really the direction we want to head. The second is my preferred short term solution. The third - converting XFS radix trees to xarray - is the longer term solution. Hence to fix the regression here, we take option 2 as it moves us in the direction we want to head with memory allocation and GFP_NOFS removal. Reported-by: syzbot+8fdff861a781522bda4d@syzkaller.appspotmail.com Reported-by: syzbot+d247769793ec169e4bf9@syzkaller.appspotmail.com Fixes: 94a69db2367e ("xfs: use __GFP_NOLOCKDEP instead of GFP_NOFS") Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: "Darrick J. Wong" <djwong@kernel.org> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff 0a97c01c Thu Nov 30 12:40:18 MST 2023 Nhat Pham <nphamcs@gmail.com> list_lru: allow explicit memcg and NUMA node selection Patch series "workload-specific and memory pressure-driven zswap writeback", v8. There are currently several issues with zswap writeback: 1. There is only a single global LRU for zswap, making it impossible to perform worload-specific shrinking - an memcg under memory pressure cannot determine which pages in the pool it owns, and often ends up writing pages from other memcgs. This issue has been previously observed in practice and mitigated by simply disabling memcg-initiated shrinking: https://lore.kernel.org/all/20230530232435.3097106-1-nphamcs@gmail.com/T/#u But this solution leaves a lot to be desired, as we still do not have an avenue for an memcg to free up its own memory locked up in the zswap pool. 2. We only shrink the zswap pool when the user-defined limit is hit. This means that if we set the limit too high, cold data that are unlikely to be used again will reside in the pool, wasting precious memory. It is hard to predict how much zswap space will be needed ahead of time, as this depends on the workload (specifically, on factors such as memory access patterns and compressibility of the memory pages). This patch series solves these issues by separating the global zswap LRU into per-memcg and per-NUMA LRUs, and performs workload-specific (i.e memcg- and NUMA-aware) zswap writeback under memory pressure. The new shrinker does not have any parameter that must be tuned by the user, and can be opted in or out on a per-memcg basis. As a proof of concept, we ran the following synthetic benchmark: build the linux kernel in a memory-limited cgroup, and allocate some cold data in tmpfs to see if the shrinker could write them out and improved the overall performance. Depending on the amount of cold data generated, we observe from 14% to 35% reduction in kernel CPU time used in the kernel builds. This patch (of 6): The interface of list_lru is based on the assumption that the list node and the data it represents belong to the same allocated on the correct node/memcg. While this assumption is valid for existing slab objects LRU such as dentries and inodes, it is undocumented, and rather inflexible for certain potential list_lru users (such as the upcoming zswap shrinker and the THP shrinker). It has caused us a lot of issues during our development. This patch changes list_lru interface so that the caller must explicitly specify numa node and memcg when adding and removing objects. The old list_lru_add() and list_lru_del() are renamed to list_lru_add_obj() and list_lru_del_obj(), respectively. It also extends the list_lru API with a new function, list_lru_putback, which undoes a previous list_lru_isolate call. Unlike list_lru_add, it does not increment the LRU node count (as list_lru_isolate does not decrement the node count). list_lru_putback also allows for explicit memcg and NUMA node selection. Link: https://lkml.kernel.org/r/20231130194023.4102148-1-nphamcs@gmail.com Link: https://lkml.kernel.org/r/20231130194023.4102148-2-nphamcs@gmail.com Signed-off-by: Nhat Pham <nphamcs@gmail.com> Suggested-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Tested-by: Bagas Sanjaya <bagasdotme@gmail.com> Cc: Chris Li <chrisl@kernel.org> Cc: Dan Streetman <ddstreet@ieee.org> Cc: Domenico Cerasuolo <cerasuolodomenico@gmail.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Muchun Song <muchun.song@linux.dev> Cc: Roman Gushchin <roman.gushchin@linux.dev> Cc: Seth Jennings <sjenning@redhat.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Shuah Khan <shuah@kernel.org> Cc: Vitaly Wool <vitaly.wool@konsulko.com> Cc: Yosry Ahmed <yosryahmed@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> diff 0b04dd5d Wed Jul 15 18:51:47 MDT 2020 Darrick J. Wong <darrick.wong@oracle.com> xfs: always use xfs_dquot_type when extracting type from a dquot Always use the xfs_dquot_type helper to extract the quota type from an incore dquot. This moves responsibility for filtering internal state information and whatnot to anybody passing around a struct xfs_dquot. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 0b0fa1d1 Tue Jul 14 11:37:22 MDT 2020 Darrick J. Wong <darrick.wong@oracle.com> xfs: stop using q_core.d_flags in the quota code Use the incore dq_flags to figure out the dquot type. This is the first step towards removing xfs_disk_dquot from the incore dquot. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Chandan Babu R <chandanrlinux@gmail.com> diff 0c7e5afb Mon Jun 29 15:48:46 MDT 2020 Dave Chinner <david@fromorbit.com> xfs: mark dquot buffers in cache dquot buffers always have write IO callbacks, so by marking them directly we can avoid needing to attach ->b_iodone functions to them. This avoids an indirect call, and makes future modifications much simpler. This is largely a rearrangement of the code at this point - no IO completion functionality changes at this point, just how the code is run is modified. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 78bba5c8 Wed May 13 16:33:27 MDT 2020 Darrick J. Wong <darrick.wong@oracle.com> xfs: use ordered buffers to initialize dquot buffers during quotacheck While QAing the new xfs_repair quotacheck code, I uncovered a quota corruption bug resulting from a bad interaction between dquot buffer initialization and quotacheck. The bug can be reproduced with the following sequence: # mkfs.xfs -f /dev/sdf # mount /dev/sdf /opt -o usrquota # su nobody -s /bin/bash -c 'touch /opt/barf' # sync # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 3 0 0 00 [------] nobody 1 0 0 00 [------] # xfs_io -x -c 'shutdown' /opt # umount /opt # mount /dev/sdf /opt -o usrquota # touch /opt/man2 # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 1 0 0 00 [------] nobody 1 0 0 00 [------] # umount /opt Notice how the initial quotacheck set the root dquot icount to 3 (rootino, rbmino, rsumino), but after shutdown -> remount -> recovery, xfs_quota reports that the root dquot has only 1 icount. We haven't deleted anything from the filesystem, which means that quota is now under-counting. This behavior is not limited to icount or the root dquot, but this is the shortest reproducer. I traced the cause of this discrepancy to the way that we handle ondisk dquot updates during quotacheck vs. regular fs activity. Normally, when we allocate a disk block for a dquot, we log the buffer as a regular (dquot) buffer. Subsequent updates to the dquots backed by that block are done via separate dquot log item updates, which means that they depend on the logged buffer update being written to disk before the dquot items. Because individual dquots have their own LSN fields, that initial dquot buffer must always be recovered. However, the story changes for quotacheck, which can cause dquot block allocations but persists the final dquot counter values via a delwri list. Because recovery doesn't gate dquot buffer replay on an LSN, this means that the initial dquot buffer can be replayed over the (newer) contents that were delwritten at the end of quotacheck. In effect, this re-initializes the dquot counters after they've been updated. If the log does not contain any other dquot items to recover, the obsolete dquot contents will not be corrected by log recovery. Because quotacheck uses a transaction to log the setting of the CHKD flags in the superblock, we skip quotacheck during the second mount call, which allows the incorrect icount to remain. Fix this by changing the ondisk dquot initialization function to use ordered buffers to write out fresh dquot blocks if it detects that we're running quotacheck. If the system goes down before quotacheck can complete, the CHKD flags will not be set in the superblock and the next mount will run quotacheck again, which can fix uninitialized dquot buffers. This requires amending the defer code to maintaine ordered buffer state across defer rolls for the sake of the dquot allocation code. For regular operations we preserve the current behavior since the dquot items require properly initialized ondisk dquot records. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 78bba5c8 Wed May 13 16:33:27 MDT 2020 Darrick J. Wong <darrick.wong@oracle.com> xfs: use ordered buffers to initialize dquot buffers during quotacheck While QAing the new xfs_repair quotacheck code, I uncovered a quota corruption bug resulting from a bad interaction between dquot buffer initialization and quotacheck. The bug can be reproduced with the following sequence: # mkfs.xfs -f /dev/sdf # mount /dev/sdf /opt -o usrquota # su nobody -s /bin/bash -c 'touch /opt/barf' # sync # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 3 0 0 00 [------] nobody 1 0 0 00 [------] # xfs_io -x -c 'shutdown' /opt # umount /opt # mount /dev/sdf /opt -o usrquota # touch /opt/man2 # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 1 0 0 00 [------] nobody 1 0 0 00 [------] # umount /opt Notice how the initial quotacheck set the root dquot icount to 3 (rootino, rbmino, rsumino), but after shutdown -> remount -> recovery, xfs_quota reports that the root dquot has only 1 icount. We haven't deleted anything from the filesystem, which means that quota is now under-counting. This behavior is not limited to icount or the root dquot, but this is the shortest reproducer. I traced the cause of this discrepancy to the way that we handle ondisk dquot updates during quotacheck vs. regular fs activity. Normally, when we allocate a disk block for a dquot, we log the buffer as a regular (dquot) buffer. Subsequent updates to the dquots backed by that block are done via separate dquot log item updates, which means that they depend on the logged buffer update being written to disk before the dquot items. Because individual dquots have their own LSN fields, that initial dquot buffer must always be recovered. However, the story changes for quotacheck, which can cause dquot block allocations but persists the final dquot counter values via a delwri list. Because recovery doesn't gate dquot buffer replay on an LSN, this means that the initial dquot buffer can be replayed over the (newer) contents that were delwritten at the end of quotacheck. In effect, this re-initializes the dquot counters after they've been updated. If the log does not contain any other dquot items to recover, the obsolete dquot contents will not be corrected by log recovery. Because quotacheck uses a transaction to log the setting of the CHKD flags in the superblock, we skip quotacheck during the second mount call, which allows the incorrect icount to remain. Fix this by changing the ondisk dquot initialization function to use ordered buffers to write out fresh dquot blocks if it detects that we're running quotacheck. If the system goes down before quotacheck can complete, the CHKD flags will not be set in the superblock and the next mount will run quotacheck again, which can fix uninitialized dquot buffers. This requires amending the defer code to maintaine ordered buffer state across defer rolls for the sake of the dquot allocation code. For regular operations we preserve the current behavior since the dquot items require properly initialized ondisk dquot records. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 78bba5c8 Wed May 13 16:33:27 MDT 2020 Darrick J. Wong <darrick.wong@oracle.com> xfs: use ordered buffers to initialize dquot buffers during quotacheck While QAing the new xfs_repair quotacheck code, I uncovered a quota corruption bug resulting from a bad interaction between dquot buffer initialization and quotacheck. The bug can be reproduced with the following sequence: # mkfs.xfs -f /dev/sdf # mount /dev/sdf /opt -o usrquota # su nobody -s /bin/bash -c 'touch /opt/barf' # sync # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 3 0 0 00 [------] nobody 1 0 0 00 [------] # xfs_io -x -c 'shutdown' /opt # umount /opt # mount /dev/sdf /opt -o usrquota # touch /opt/man2 # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 1 0 0 00 [------] nobody 1 0 0 00 [------] # umount /opt Notice how the initial quotacheck set the root dquot icount to 3 (rootino, rbmino, rsumino), but after shutdown -> remount -> recovery, xfs_quota reports that the root dquot has only 1 icount. We haven't deleted anything from the filesystem, which means that quota is now under-counting. This behavior is not limited to icount or the root dquot, but this is the shortest reproducer. I traced the cause of this discrepancy to the way that we handle ondisk dquot updates during quotacheck vs. regular fs activity. Normally, when we allocate a disk block for a dquot, we log the buffer as a regular (dquot) buffer. Subsequent updates to the dquots backed by that block are done via separate dquot log item updates, which means that they depend on the logged buffer update being written to disk before the dquot items. Because individual dquots have their own LSN fields, that initial dquot buffer must always be recovered. However, the story changes for quotacheck, which can cause dquot block allocations but persists the final dquot counter values via a delwri list. Because recovery doesn't gate dquot buffer replay on an LSN, this means that the initial dquot buffer can be replayed over the (newer) contents that were delwritten at the end of quotacheck. In effect, this re-initializes the dquot counters after they've been updated. If the log does not contain any other dquot items to recover, the obsolete dquot contents will not be corrected by log recovery. Because quotacheck uses a transaction to log the setting of the CHKD flags in the superblock, we skip quotacheck during the second mount call, which allows the incorrect icount to remain. Fix this by changing the ondisk dquot initialization function to use ordered buffers to write out fresh dquot blocks if it detects that we're running quotacheck. If the system goes down before quotacheck can complete, the CHKD flags will not be set in the superblock and the next mount will run quotacheck again, which can fix uninitialized dquot buffers. This requires amending the defer code to maintaine ordered buffer state across defer rolls for the sake of the dquot allocation code. For regular operations we preserve the current behavior since the dquot items require properly initialized ondisk dquot records. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 78bba5c8 Wed May 13 16:33:27 MDT 2020 Darrick J. Wong <darrick.wong@oracle.com> xfs: use ordered buffers to initialize dquot buffers during quotacheck While QAing the new xfs_repair quotacheck code, I uncovered a quota corruption bug resulting from a bad interaction between dquot buffer initialization and quotacheck. The bug can be reproduced with the following sequence: # mkfs.xfs -f /dev/sdf # mount /dev/sdf /opt -o usrquota # su nobody -s /bin/bash -c 'touch /opt/barf' # sync # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 3 0 0 00 [------] nobody 1 0 0 00 [------] # xfs_io -x -c 'shutdown' /opt # umount /opt # mount /dev/sdf /opt -o usrquota # touch /opt/man2 # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 1 0 0 00 [------] nobody 1 0 0 00 [------] # umount /opt Notice how the initial quotacheck set the root dquot icount to 3 (rootino, rbmino, rsumino), but after shutdown -> remount -> recovery, xfs_quota reports that the root dquot has only 1 icount. We haven't deleted anything from the filesystem, which means that quota is now under-counting. This behavior is not limited to icount or the root dquot, but this is the shortest reproducer. I traced the cause of this discrepancy to the way that we handle ondisk dquot updates during quotacheck vs. regular fs activity. Normally, when we allocate a disk block for a dquot, we log the buffer as a regular (dquot) buffer. Subsequent updates to the dquots backed by that block are done via separate dquot log item updates, which means that they depend on the logged buffer update being written to disk before the dquot items. Because individual dquots have their own LSN fields, that initial dquot buffer must always be recovered. However, the story changes for quotacheck, which can cause dquot block allocations but persists the final dquot counter values via a delwri list. Because recovery doesn't gate dquot buffer replay on an LSN, this means that the initial dquot buffer can be replayed over the (newer) contents that were delwritten at the end of quotacheck. In effect, this re-initializes the dquot counters after they've been updated. If the log does not contain any other dquot items to recover, the obsolete dquot contents will not be corrected by log recovery. Because quotacheck uses a transaction to log the setting of the CHKD flags in the superblock, we skip quotacheck during the second mount call, which allows the incorrect icount to remain. Fix this by changing the ondisk dquot initialization function to use ordered buffers to write out fresh dquot blocks if it detects that we're running quotacheck. If the system goes down before quotacheck can complete, the CHKD flags will not be set in the superblock and the next mount will run quotacheck again, which can fix uninitialized dquot buffers. This requires amending the defer code to maintaine ordered buffer state across defer rolls for the sake of the dquot allocation code. For regular operations we preserve the current behavior since the dquot items require properly initialized ondisk dquot records. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 78bba5c8 Wed May 13 16:33:27 MDT 2020 Darrick J. Wong <darrick.wong@oracle.com> xfs: use ordered buffers to initialize dquot buffers during quotacheck While QAing the new xfs_repair quotacheck code, I uncovered a quota corruption bug resulting from a bad interaction between dquot buffer initialization and quotacheck. The bug can be reproduced with the following sequence: # mkfs.xfs -f /dev/sdf # mount /dev/sdf /opt -o usrquota # su nobody -s /bin/bash -c 'touch /opt/barf' # sync # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 3 0 0 00 [------] nobody 1 0 0 00 [------] # xfs_io -x -c 'shutdown' /opt # umount /opt # mount /dev/sdf /opt -o usrquota # touch /opt/man2 # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 1 0 0 00 [------] nobody 1 0 0 00 [------] # umount /opt Notice how the initial quotacheck set the root dquot icount to 3 (rootino, rbmino, rsumino), but after shutdown -> remount -> recovery, xfs_quota reports that the root dquot has only 1 icount. We haven't deleted anything from the filesystem, which means that quota is now under-counting. This behavior is not limited to icount or the root dquot, but this is the shortest reproducer. I traced the cause of this discrepancy to the way that we handle ondisk dquot updates during quotacheck vs. regular fs activity. Normally, when we allocate a disk block for a dquot, we log the buffer as a regular (dquot) buffer. Subsequent updates to the dquots backed by that block are done via separate dquot log item updates, which means that they depend on the logged buffer update being written to disk before the dquot items. Because individual dquots have their own LSN fields, that initial dquot buffer must always be recovered. However, the story changes for quotacheck, which can cause dquot block allocations but persists the final dquot counter values via a delwri list. Because recovery doesn't gate dquot buffer replay on an LSN, this means that the initial dquot buffer can be replayed over the (newer) contents that were delwritten at the end of quotacheck. In effect, this re-initializes the dquot counters after they've been updated. If the log does not contain any other dquot items to recover, the obsolete dquot contents will not be corrected by log recovery. Because quotacheck uses a transaction to log the setting of the CHKD flags in the superblock, we skip quotacheck during the second mount call, which allows the incorrect icount to remain. Fix this by changing the ondisk dquot initialization function to use ordered buffers to write out fresh dquot blocks if it detects that we're running quotacheck. If the system goes down before quotacheck can complete, the CHKD flags will not be set in the superblock and the next mount will run quotacheck again, which can fix uninitialized dquot buffers. This requires amending the defer code to maintaine ordered buffer state across defer rolls for the sake of the dquot allocation code. For regular operations we preserve the current behavior since the dquot items require properly initialized ondisk dquot records. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 78bba5c8 Wed May 13 16:33:27 MDT 2020 Darrick J. Wong <darrick.wong@oracle.com> xfs: use ordered buffers to initialize dquot buffers during quotacheck While QAing the new xfs_repair quotacheck code, I uncovered a quota corruption bug resulting from a bad interaction between dquot buffer initialization and quotacheck. The bug can be reproduced with the following sequence: # mkfs.xfs -f /dev/sdf # mount /dev/sdf /opt -o usrquota # su nobody -s /bin/bash -c 'touch /opt/barf' # sync # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 3 0 0 00 [------] nobody 1 0 0 00 [------] # xfs_io -x -c 'shutdown' /opt # umount /opt # mount /dev/sdf /opt -o usrquota # touch /opt/man2 # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 1 0 0 00 [------] nobody 1 0 0 00 [------] # umount /opt Notice how the initial quotacheck set the root dquot icount to 3 (rootino, rbmino, rsumino), but after shutdown -> remount -> recovery, xfs_quota reports that the root dquot has only 1 icount. We haven't deleted anything from the filesystem, which means that quota is now under-counting. This behavior is not limited to icount or the root dquot, but this is the shortest reproducer. I traced the cause of this discrepancy to the way that we handle ondisk dquot updates during quotacheck vs. regular fs activity. Normally, when we allocate a disk block for a dquot, we log the buffer as a regular (dquot) buffer. Subsequent updates to the dquots backed by that block are done via separate dquot log item updates, which means that they depend on the logged buffer update being written to disk before the dquot items. Because individual dquots have their own LSN fields, that initial dquot buffer must always be recovered. However, the story changes for quotacheck, which can cause dquot block allocations but persists the final dquot counter values via a delwri list. Because recovery doesn't gate dquot buffer replay on an LSN, this means that the initial dquot buffer can be replayed over the (newer) contents that were delwritten at the end of quotacheck. In effect, this re-initializes the dquot counters after they've been updated. If the log does not contain any other dquot items to recover, the obsolete dquot contents will not be corrected by log recovery. Because quotacheck uses a transaction to log the setting of the CHKD flags in the superblock, we skip quotacheck during the second mount call, which allows the incorrect icount to remain. Fix this by changing the ondisk dquot initialization function to use ordered buffers to write out fresh dquot blocks if it detects that we're running quotacheck. If the system goes down before quotacheck can complete, the CHKD flags will not be set in the superblock and the next mount will run quotacheck again, which can fix uninitialized dquot buffers. This requires amending the defer code to maintaine ordered buffer state across defer rolls for the sake of the dquot allocation code. For regular operations we preserve the current behavior since the dquot items require properly initialized ondisk dquot records. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 78bba5c8 Wed May 13 16:33:27 MDT 2020 Darrick J. Wong <darrick.wong@oracle.com> xfs: use ordered buffers to initialize dquot buffers during quotacheck While QAing the new xfs_repair quotacheck code, I uncovered a quota corruption bug resulting from a bad interaction between dquot buffer initialization and quotacheck. The bug can be reproduced with the following sequence: # mkfs.xfs -f /dev/sdf # mount /dev/sdf /opt -o usrquota # su nobody -s /bin/bash -c 'touch /opt/barf' # sync # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 3 0 0 00 [------] nobody 1 0 0 00 [------] # xfs_io -x -c 'shutdown' /opt # umount /opt # mount /dev/sdf /opt -o usrquota # touch /opt/man2 # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 1 0 0 00 [------] nobody 1 0 0 00 [------] # umount /opt Notice how the initial quotacheck set the root dquot icount to 3 (rootino, rbmino, rsumino), but after shutdown -> remount -> recovery, xfs_quota reports that the root dquot has only 1 icount. We haven't deleted anything from the filesystem, which means that quota is now under-counting. This behavior is not limited to icount or the root dquot, but this is the shortest reproducer. I traced the cause of this discrepancy to the way that we handle ondisk dquot updates during quotacheck vs. regular fs activity. Normally, when we allocate a disk block for a dquot, we log the buffer as a regular (dquot) buffer. Subsequent updates to the dquots backed by that block are done via separate dquot log item updates, which means that they depend on the logged buffer update being written to disk before the dquot items. Because individual dquots have their own LSN fields, that initial dquot buffer must always be recovered. However, the story changes for quotacheck, which can cause dquot block allocations but persists the final dquot counter values via a delwri list. Because recovery doesn't gate dquot buffer replay on an LSN, this means that the initial dquot buffer can be replayed over the (newer) contents that were delwritten at the end of quotacheck. In effect, this re-initializes the dquot counters after they've been updated. If the log does not contain any other dquot items to recover, the obsolete dquot contents will not be corrected by log recovery. Because quotacheck uses a transaction to log the setting of the CHKD flags in the superblock, we skip quotacheck during the second mount call, which allows the incorrect icount to remain. Fix this by changing the ondisk dquot initialization function to use ordered buffers to write out fresh dquot blocks if it detects that we're running quotacheck. If the system goes down before quotacheck can complete, the CHKD flags will not be set in the superblock and the next mount will run quotacheck again, which can fix uninitialized dquot buffers. This requires amending the defer code to maintaine ordered buffer state across defer rolls for the sake of the dquot allocation code. For regular operations we preserve the current behavior since the dquot items require properly initialized ondisk dquot records. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 78bba5c8 Wed May 13 16:33:27 MDT 2020 Darrick J. Wong <darrick.wong@oracle.com> xfs: use ordered buffers to initialize dquot buffers during quotacheck While QAing the new xfs_repair quotacheck code, I uncovered a quota corruption bug resulting from a bad interaction between dquot buffer initialization and quotacheck. The bug can be reproduced with the following sequence: # mkfs.xfs -f /dev/sdf # mount /dev/sdf /opt -o usrquota # su nobody -s /bin/bash -c 'touch /opt/barf' # sync # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 3 0 0 00 [------] nobody 1 0 0 00 [------] # xfs_io -x -c 'shutdown' /opt # umount /opt # mount /dev/sdf /opt -o usrquota # touch /opt/man2 # xfs_quota -x -c 'report -ahi' /opt User quota on /opt (/dev/sdf) Inodes User ID Used Soft Hard Warn/Grace ---------- --------------------------------- root 1 0 0 00 [------] nobody 1 0 0 00 [------] # umount /opt Notice how the initial quotacheck set the root dquot icount to 3 (rootino, rbmino, rsumino), but after shutdown -> remount -> recovery, xfs_quota reports that the root dquot has only 1 icount. We haven't deleted anything from the filesystem, which means that quota is now under-counting. This behavior is not limited to icount or the root dquot, but this is the shortest reproducer. I traced the cause of this discrepancy to the way that we handle ondisk dquot updates during quotacheck vs. regular fs activity. Normally, when we allocate a disk block for a dquot, we log the buffer as a regular (dquot) buffer. Subsequent updates to the dquots backed by that block are done via separate dquot log item updates, which means that they depend on the logged buffer update being written to disk before the dquot items. Because individual dquots have their own LSN fields, that initial dquot buffer must always be recovered. However, the story changes for quotacheck, which can cause dquot block allocations but persists the final dquot counter values via a delwri list. Because recovery doesn't gate dquot buffer replay on an LSN, this means that the initial dquot buffer can be replayed over the (newer) contents that were delwritten at the end of quotacheck. In effect, this re-initializes the dquot counters after they've been updated. If the log does not contain any other dquot items to recover, the obsolete dquot contents will not be corrected by log recovery. Because quotacheck uses a transaction to log the setting of the CHKD flags in the superblock, we skip quotacheck during the second mount call, which allows the incorrect icount to remain. Fix this by changing the ondisk dquot initialization function to use ordered buffers to write out fresh dquot blocks if it detects that we're running quotacheck. If the system goes down before quotacheck can complete, the CHKD flags will not be set in the superblock and the next mount will run quotacheck again, which can fix uninitialized dquot buffers. This requires amending the defer code to maintaine ordered buffer state across defer rolls for the sake of the dquot allocation code. For regular operations we preserve the current behavior since the dquot items require properly initialized ondisk dquot records. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> |
H A D | xfs_trans_dquot.c | diff 707e0dda Mon Aug 26 01:06:22 MDT 2019 Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> fs: xfs: Remove KM_NOSLEEP and KM_SLEEP. Since no caller is using KM_NOSLEEP and no callee branches on KM_SLEEP, we can remove KM_NOSLEEP and replace KM_SLEEP with 0. Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> |
H A D | xfs_trans_ail.c | diff 939bd50d Wed Jun 28 12:04:31 MDT 2023 Dave Chinner <dchinner@redhat.com> xfs: don't reverse order of items in bulk AIL insertion XFS has strict metadata ordering requirements. One of the things it does is maintain the commit order of items from transaction commit through the CIL and into the AIL. That is, if a transaction logs item A before item B in a modification, then they will be inserted into the CIL in the order {A, B}. These items are then written into the iclog during checkpointing in the order {A, B}. When the checkpoint commits, they are supposed to be inserted into the AIL in the order {A, B}, and when they are pushed from the AIL, they are pushed in the order {A, B}. If we crash, log recovery then replays the two items from the checkpoint in the order {A, B}, resulting in the objects the items apply to being queued for writeback at the end of the checkpoint in the order {A, B}. This means recovery behaves the same way as the runtime code. In places, we have subtle dependencies on this ordering being maintained. One of this place is performing intent recovery from the log. It assumes that recovering an intent will result in a non-intent object being the first thing that is modified in the recovery transaction, and so when the transaction commits and the journal flushes, the first object inserted into the AIL beyond the intent recovery range will be a non-intent item. It uses the transistion from intent items to non-intent items to stop the recovery pass. A recent log recovery issue indicated that an intent was appearing as the first item in the AIL beyond the recovery range, hence breaking the end of recovery detection that exists. Tracing indicated insertion of the items into the AIL was apparently occurring in the right order (the intent was last in the commit item list), but the intent was appearing first in the AIL. IOWs, the order of items in the AIL was {D,C,B,A}, not {A,B,C,D}, and bulk insertion was reversing the order of the items in the batch of items being inserted. Lucky for us, all the items fed to bulk insertion have the same LSN, so the reversal of order does not affect the log head/tail tracking that is based on the contents of the AIL. It only impacts on code that has implicit, subtle dependencies on object order, and AFAICT only the intent recovery loop is impacted by it. Make sure bulk AIL insertion does not reorder items incorrectly. Fixes: 0e57f6a36f9b ("xfs: bulk AIL insertion during transaction commit") Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Chandan Babu R <chandan.babu@oracle.com> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> |
H A D | xfs_trans_buf.c | diff 0dc63c8a Thu Feb 22 01:43:36 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: launder in-memory btree buffers before transaction commit As we've noted in various places, all current users of in-memory btrees are online fsck. Online fsck only stages a btree long enough to rebuild an ondisk data structure, which means that the in-memory btree is ephemeral. Furthermore, if we encounter /any/ errors while updating an in-memory btree, all we do is tear down all the staged data and return an errno to userspace. In-memory btrees need not be transactional, so their buffers should not be committed to the ondisk log, nor should they be checkpointed by the AIL. That's just as well since the ephemeral nature of the btree means that the buftarg and the buffers may disappear quickly anyway. Therefore, we need a way to launder the btree buffers that get attached to the transaction by the generic btree code. Because the buffers are directly mapped to backing file pages, there's no need to bwrite them back to the tmpfs file. All we need to do is clean enough of the buffer log item state so that the bli can be detached from the buffer, remove the bli from the transaction's log item list, and reset the transaction dirty state as if the laundered items had never been there. For simplicity, create xfbtree transaction commit and cancel helpers that launder the in-memory btree buffers for callers. Once laundered, call the write verifier on non-stale buffers to avoid integrity issues, or punch a hole in the backing file for stale buffers. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 04fcad80 Wed Aug 18 19:46:57 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: introduce xfs_buf_daddr() Introduce a helper function xfs_buf_daddr() to extract the disk address of the buffer from the struct xfs_buf. This will replace direct accesses to bp->b_bn and bp->b_maps[0].bm_bn, as well as the XFS_BUF_ADDR() macro. This patch introduces the helper function and replaces all uses of XFS_BUF_ADDR() as this is just a simple sed replacement. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0c7e5afb Mon Jun 29 15:48:46 MDT 2020 Dave Chinner <david@fromorbit.com> xfs: mark dquot buffers in cache dquot buffers always have write IO callbacks, so by marking them directly we can avoid needing to attach ->b_iodone functions to them. This avoids an indirect call, and makes future modifications much simpler. This is largely a rearrangement of the code at this point - no IO completion functionality changes at this point, just how the code is run is modified. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 8c9ce2f7 Wed Jun 12 09:59:58 MDT 2019 Eric Sandeen <sandeen@sandeen.net> xfs: remove unused flags arg from getsb interfaces The flags value is always passed as 0 so remove the argument. Signed-off-by: Eric Sandeen <sandeen@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> |
H A D | xfs_trans.c | diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 3c4cb76b Tue Mar 29 19:22:01 MDT 2022 Dave Chinner <dchinner@redhat.com> xfs: xfs_trans_commit() path must check for log shutdown If a shut races with xfs_trans_commit() and we have shut down the filesystem but not the log, we will still cancel the transaction. This can result in aborting dirty log items instead of committing and pinning them whilst the log is still running. Hence we can end up with dirty, unlogged metadata that isn't in the AIL in memory that can be flushed to disk via writeback clustering. This was discovered from a g/388 trace where an inode log item was having IO completed on it and it wasn't in the AIL, hence tripping asserts xfs_ail_check(). Inode cluster writeback started long after the filesystem shutdown started, and long after the transaction containing the dirty inode was aborted and the log item marked XFS_LI_ABORTED. The inode was seen as dirty and unpinned, so it was flushed. IO completion tried to remove the inode from the AIL, at which point stuff went bad: XFS (pmem1): Log I/O Error (0x6) detected at xfs_fs_goingdown+0xa3/0xf0 (fs/xfs/xfs_fsops.c:500). Shutting down filesystem. XFS: Assertion failed: in_ail, file: fs/xfs/xfs_trans_ail.c, line: 67 XFS (pmem1): Please unmount the filesystem and rectify the problem(s) Workqueue: xfs-buf/pmem1 xfs_buf_ioend_work RIP: 0010:assfail+0x27/0x2d Call Trace: <TASK> xfs_ail_check+0xa8/0x180 xfs_ail_delete_one+0x3b/0xf0 xfs_buf_inode_iodone+0x329/0x3f0 xfs_buf_ioend+0x1f8/0x530 xfs_buf_ioend_work+0x15/0x20 process_one_work+0x1ac/0x390 worker_thread+0x56/0x3c0 kthread+0xf6/0x120 ret_from_fork+0x1f/0x30 </TASK> xfs_trans_commit() needs to check log state for shutdown, not mount state. It cannot abort dirty log items while the log is still running as dirty items must remained pinned in memory until they are either committed to the journal or the log has shut down and they can be safely tossed away. Hence if the log has not shut down, the xfs_trans_commit() path must allow completed transactions to commit to the CIL and pin the dirty items even if a mount shutdown has started. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> |
H A D | xfs_symlink.c | diff e6a688c3 Mon Mar 22 10:52:03 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: initialise attr fork on inode create When we allocate a new inode, we often need to add an attribute to the inode as part of the create. This can happen as a result of needing to add default ACLs or security labels before the inode is made visible to userspace. This is highly inefficient right now. We do the create transaction to allocate the inode, then we do an "add attr fork" transaction to modify the just created empty inode to set the inode fork offset to allow attributes to be stored, then we go and do the attribute creation. This means 3 transactions instead of 1 to allocate an inode, and this greatly increases the load on the CIL commit code, resulting in excessive contention on the CIL spin locks and performance degradation: 18.99% [kernel] [k] __pv_queued_spin_lock_slowpath 3.57% [kernel] [k] do_raw_spin_lock 2.51% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock 2.48% [kernel] [k] memcpy 2.34% [kernel] [k] xfs_log_commit_cil The typical profile resulting from running fsmark on a selinux enabled filesytem is adds this overhead to the create path: - 15.30% xfs_init_security - 15.23% security_inode_init_security - 13.05% xfs_initxattrs - 12.94% xfs_attr_set - 6.75% xfs_bmap_add_attrfork - 5.51% xfs_trans_commit - 5.48% __xfs_trans_commit - 5.35% xfs_log_commit_cil - 3.86% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.70% xfs_trans_alloc 0.52% xfs_trans_reserve - 5.41% xfs_attr_set_args - 5.39% xfs_attr_set_shortform.constprop.0 - 4.46% xfs_trans_commit - 4.46% __xfs_trans_commit - 4.33% xfs_log_commit_cil - 2.74% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath 0.60% xfs_inode_item_format 0.90% xfs_attr_try_sf_addname - 1.99% selinux_inode_init_security - 1.02% security_sid_to_context_force - 1.00% security_sid_to_context_core - 0.92% sidtab_entry_to_string - 0.90% sidtab_sid2str_get 0.59% sidtab_sid2str_put.part.0 - 0.82% selinux_determine_inode_label - 0.77% security_transition_sid 0.70% security_compute_sid.part.0 And fsmark creation rate performance drops by ~25%. The key point to note here is that half the additional overhead comes from adding the attribute fork to the newly created inode. That's crazy, considering we can do this same thing at inode create time with a couple of lines of code and no extra overhead. So, if we know we are going to add an attribute immediately after creating the inode, let's just initialise the attribute fork inside the create transaction and chop that whole chunk of code out of the create fast path. This completely removes the performance drop caused by enabling SELinux, and the profile looks like: - 8.99% xfs_init_security - 9.00% security_inode_init_security - 6.43% xfs_initxattrs - 6.37% xfs_attr_set - 5.45% xfs_attr_set_args - 5.42% xfs_attr_set_shortform.constprop.0 - 4.51% xfs_trans_commit - 4.54% __xfs_trans_commit - 4.59% xfs_log_commit_cil - 2.67% _raw_spin_lock - 3.28% do_raw_spin_lock 3.08% __pv_queued_spin_lock_slowpath 0.66% xfs_inode_item_format - 0.90% xfs_attr_try_sf_addname - 0.60% xfs_trans_alloc - 2.35% selinux_inode_init_security - 1.25% security_sid_to_context_force - 1.21% security_sid_to_context_core - 1.19% sidtab_entry_to_string - 1.20% sidtab_sid2str_get - 0.86% sidtab_sid2str_put.part.0 - 0.62% _raw_spin_lock_irqsave - 0.77% do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.84% selinux_determine_inode_label - 0.83% security_transition_sid 0.86% security_compute_sid.part.0 Which indicates the XFS overhead of creating the selinux xattr has been halved. This doesn't fix the CIL lock contention problem, just means it's not a limiting factor for this workload. Lock contention in the security subsystems is going to be an issue soon, though... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> [djwong: fix compilation error when CONFIG_SECURITY=n] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Gao Xiang <hsiangkao@redhat.com> diff e6a688c3 Mon Mar 22 10:52:03 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: initialise attr fork on inode create When we allocate a new inode, we often need to add an attribute to the inode as part of the create. This can happen as a result of needing to add default ACLs or security labels before the inode is made visible to userspace. This is highly inefficient right now. We do the create transaction to allocate the inode, then we do an "add attr fork" transaction to modify the just created empty inode to set the inode fork offset to allow attributes to be stored, then we go and do the attribute creation. This means 3 transactions instead of 1 to allocate an inode, and this greatly increases the load on the CIL commit code, resulting in excessive contention on the CIL spin locks and performance degradation: 18.99% [kernel] [k] __pv_queued_spin_lock_slowpath 3.57% [kernel] [k] do_raw_spin_lock 2.51% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock 2.48% [kernel] [k] memcpy 2.34% [kernel] [k] xfs_log_commit_cil The typical profile resulting from running fsmark on a selinux enabled filesytem is adds this overhead to the create path: - 15.30% xfs_init_security - 15.23% security_inode_init_security - 13.05% xfs_initxattrs - 12.94% xfs_attr_set - 6.75% xfs_bmap_add_attrfork - 5.51% xfs_trans_commit - 5.48% __xfs_trans_commit - 5.35% xfs_log_commit_cil - 3.86% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.70% xfs_trans_alloc 0.52% xfs_trans_reserve - 5.41% xfs_attr_set_args - 5.39% xfs_attr_set_shortform.constprop.0 - 4.46% xfs_trans_commit - 4.46% __xfs_trans_commit - 4.33% xfs_log_commit_cil - 2.74% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath 0.60% xfs_inode_item_format 0.90% xfs_attr_try_sf_addname - 1.99% selinux_inode_init_security - 1.02% security_sid_to_context_force - 1.00% security_sid_to_context_core - 0.92% sidtab_entry_to_string - 0.90% sidtab_sid2str_get 0.59% sidtab_sid2str_put.part.0 - 0.82% selinux_determine_inode_label - 0.77% security_transition_sid 0.70% security_compute_sid.part.0 And fsmark creation rate performance drops by ~25%. The key point to note here is that half the additional overhead comes from adding the attribute fork to the newly created inode. That's crazy, considering we can do this same thing at inode create time with a couple of lines of code and no extra overhead. So, if we know we are going to add an attribute immediately after creating the inode, let's just initialise the attribute fork inside the create transaction and chop that whole chunk of code out of the create fast path. This completely removes the performance drop caused by enabling SELinux, and the profile looks like: - 8.99% xfs_init_security - 9.00% security_inode_init_security - 6.43% xfs_initxattrs - 6.37% xfs_attr_set - 5.45% xfs_attr_set_args - 5.42% xfs_attr_set_shortform.constprop.0 - 4.51% xfs_trans_commit - 4.54% __xfs_trans_commit - 4.59% xfs_log_commit_cil - 2.67% _raw_spin_lock - 3.28% do_raw_spin_lock 3.08% __pv_queued_spin_lock_slowpath 0.66% xfs_inode_item_format - 0.90% xfs_attr_try_sf_addname - 0.60% xfs_trans_alloc - 2.35% selinux_inode_init_security - 1.25% security_sid_to_context_force - 1.21% security_sid_to_context_core - 1.19% sidtab_entry_to_string - 1.20% sidtab_sid2str_get - 0.86% sidtab_sid2str_put.part.0 - 0.62% _raw_spin_lock_irqsave - 0.77% do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.84% selinux_determine_inode_label - 0.83% security_transition_sid 0.86% security_compute_sid.part.0 Which indicates the XFS overhead of creating the selinux xattr has been halved. This doesn't fix the CIL lock contention problem, just means it's not a limiting factor for this workload. Lock contention in the security subsystems is going to be an issue soon, though... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> [djwong: fix compilation error when CONFIG_SECURITY=n] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Gao Xiang <hsiangkao@redhat.com> diff e6a688c3 Mon Mar 22 10:52:03 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: initialise attr fork on inode create When we allocate a new inode, we often need to add an attribute to the inode as part of the create. This can happen as a result of needing to add default ACLs or security labels before the inode is made visible to userspace. This is highly inefficient right now. We do the create transaction to allocate the inode, then we do an "add attr fork" transaction to modify the just created empty inode to set the inode fork offset to allow attributes to be stored, then we go and do the attribute creation. This means 3 transactions instead of 1 to allocate an inode, and this greatly increases the load on the CIL commit code, resulting in excessive contention on the CIL spin locks and performance degradation: 18.99% [kernel] [k] __pv_queued_spin_lock_slowpath 3.57% [kernel] [k] do_raw_spin_lock 2.51% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock 2.48% [kernel] [k] memcpy 2.34% [kernel] [k] xfs_log_commit_cil The typical profile resulting from running fsmark on a selinux enabled filesytem is adds this overhead to the create path: - 15.30% xfs_init_security - 15.23% security_inode_init_security - 13.05% xfs_initxattrs - 12.94% xfs_attr_set - 6.75% xfs_bmap_add_attrfork - 5.51% xfs_trans_commit - 5.48% __xfs_trans_commit - 5.35% xfs_log_commit_cil - 3.86% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.70% xfs_trans_alloc 0.52% xfs_trans_reserve - 5.41% xfs_attr_set_args - 5.39% xfs_attr_set_shortform.constprop.0 - 4.46% xfs_trans_commit - 4.46% __xfs_trans_commit - 4.33% xfs_log_commit_cil - 2.74% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath 0.60% xfs_inode_item_format 0.90% xfs_attr_try_sf_addname - 1.99% selinux_inode_init_security - 1.02% security_sid_to_context_force - 1.00% security_sid_to_context_core - 0.92% sidtab_entry_to_string - 0.90% sidtab_sid2str_get 0.59% sidtab_sid2str_put.part.0 - 0.82% selinux_determine_inode_label - 0.77% security_transition_sid 0.70% security_compute_sid.part.0 And fsmark creation rate performance drops by ~25%. The key point to note here is that half the additional overhead comes from adding the attribute fork to the newly created inode. That's crazy, considering we can do this same thing at inode create time with a couple of lines of code and no extra overhead. So, if we know we are going to add an attribute immediately after creating the inode, let's just initialise the attribute fork inside the create transaction and chop that whole chunk of code out of the create fast path. This completely removes the performance drop caused by enabling SELinux, and the profile looks like: - 8.99% xfs_init_security - 9.00% security_inode_init_security - 6.43% xfs_initxattrs - 6.37% xfs_attr_set - 5.45% xfs_attr_set_args - 5.42% xfs_attr_set_shortform.constprop.0 - 4.51% xfs_trans_commit - 4.54% __xfs_trans_commit - 4.59% xfs_log_commit_cil - 2.67% _raw_spin_lock - 3.28% do_raw_spin_lock 3.08% __pv_queued_spin_lock_slowpath 0.66% xfs_inode_item_format - 0.90% xfs_attr_try_sf_addname - 0.60% xfs_trans_alloc - 2.35% selinux_inode_init_security - 1.25% security_sid_to_context_force - 1.21% security_sid_to_context_core - 1.19% sidtab_entry_to_string - 1.20% sidtab_sid2str_get - 0.86% sidtab_sid2str_put.part.0 - 0.62% _raw_spin_lock_irqsave - 0.77% do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.84% selinux_determine_inode_label - 0.83% security_transition_sid 0.86% security_compute_sid.part.0 Which indicates the XFS overhead of creating the selinux xattr has been halved. This doesn't fix the CIL lock contention problem, just means it's not a limiting factor for this workload. Lock contention in the security subsystems is going to be an issue soon, though... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> [djwong: fix compilation error when CONFIG_SECURITY=n] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Gao Xiang <hsiangkao@redhat.com> diff e6a688c3 Mon Mar 22 10:52:03 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: initialise attr fork on inode create When we allocate a new inode, we often need to add an attribute to the inode as part of the create. This can happen as a result of needing to add default ACLs or security labels before the inode is made visible to userspace. This is highly inefficient right now. We do the create transaction to allocate the inode, then we do an "add attr fork" transaction to modify the just created empty inode to set the inode fork offset to allow attributes to be stored, then we go and do the attribute creation. This means 3 transactions instead of 1 to allocate an inode, and this greatly increases the load on the CIL commit code, resulting in excessive contention on the CIL spin locks and performance degradation: 18.99% [kernel] [k] __pv_queued_spin_lock_slowpath 3.57% [kernel] [k] do_raw_spin_lock 2.51% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock 2.48% [kernel] [k] memcpy 2.34% [kernel] [k] xfs_log_commit_cil The typical profile resulting from running fsmark on a selinux enabled filesytem is adds this overhead to the create path: - 15.30% xfs_init_security - 15.23% security_inode_init_security - 13.05% xfs_initxattrs - 12.94% xfs_attr_set - 6.75% xfs_bmap_add_attrfork - 5.51% xfs_trans_commit - 5.48% __xfs_trans_commit - 5.35% xfs_log_commit_cil - 3.86% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.70% xfs_trans_alloc 0.52% xfs_trans_reserve - 5.41% xfs_attr_set_args - 5.39% xfs_attr_set_shortform.constprop.0 - 4.46% xfs_trans_commit - 4.46% __xfs_trans_commit - 4.33% xfs_log_commit_cil - 2.74% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath 0.60% xfs_inode_item_format 0.90% xfs_attr_try_sf_addname - 1.99% selinux_inode_init_security - 1.02% security_sid_to_context_force - 1.00% security_sid_to_context_core - 0.92% sidtab_entry_to_string - 0.90% sidtab_sid2str_get 0.59% sidtab_sid2str_put.part.0 - 0.82% selinux_determine_inode_label - 0.77% security_transition_sid 0.70% security_compute_sid.part.0 And fsmark creation rate performance drops by ~25%. The key point to note here is that half the additional overhead comes from adding the attribute fork to the newly created inode. That's crazy, considering we can do this same thing at inode create time with a couple of lines of code and no extra overhead. So, if we know we are going to add an attribute immediately after creating the inode, let's just initialise the attribute fork inside the create transaction and chop that whole chunk of code out of the create fast path. This completely removes the performance drop caused by enabling SELinux, and the profile looks like: - 8.99% xfs_init_security - 9.00% security_inode_init_security - 6.43% xfs_initxattrs - 6.37% xfs_attr_set - 5.45% xfs_attr_set_args - 5.42% xfs_attr_set_shortform.constprop.0 - 4.51% xfs_trans_commit - 4.54% __xfs_trans_commit - 4.59% xfs_log_commit_cil - 2.67% _raw_spin_lock - 3.28% do_raw_spin_lock 3.08% __pv_queued_spin_lock_slowpath 0.66% xfs_inode_item_format - 0.90% xfs_attr_try_sf_addname - 0.60% xfs_trans_alloc - 2.35% selinux_inode_init_security - 1.25% security_sid_to_context_force - 1.21% security_sid_to_context_core - 1.19% sidtab_entry_to_string - 1.20% sidtab_sid2str_get - 0.86% sidtab_sid2str_put.part.0 - 0.62% _raw_spin_lock_irqsave - 0.77% do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.84% selinux_determine_inode_label - 0.83% security_transition_sid 0.86% security_compute_sid.part.0 Which indicates the XFS overhead of creating the selinux xattr has been halved. This doesn't fix the CIL lock contention problem, just means it's not a limiting factor for this workload. Lock contention in the security subsystems is going to be an issue soon, though... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> [djwong: fix compilation error when CONFIG_SECURITY=n] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Gao Xiang <hsiangkao@redhat.com> diff e6a688c3 Mon Mar 22 10:52:03 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: initialise attr fork on inode create When we allocate a new inode, we often need to add an attribute to the inode as part of the create. This can happen as a result of needing to add default ACLs or security labels before the inode is made visible to userspace. This is highly inefficient right now. We do the create transaction to allocate the inode, then we do an "add attr fork" transaction to modify the just created empty inode to set the inode fork offset to allow attributes to be stored, then we go and do the attribute creation. This means 3 transactions instead of 1 to allocate an inode, and this greatly increases the load on the CIL commit code, resulting in excessive contention on the CIL spin locks and performance degradation: 18.99% [kernel] [k] __pv_queued_spin_lock_slowpath 3.57% [kernel] [k] do_raw_spin_lock 2.51% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock 2.48% [kernel] [k] memcpy 2.34% [kernel] [k] xfs_log_commit_cil The typical profile resulting from running fsmark on a selinux enabled filesytem is adds this overhead to the create path: - 15.30% xfs_init_security - 15.23% security_inode_init_security - 13.05% xfs_initxattrs - 12.94% xfs_attr_set - 6.75% xfs_bmap_add_attrfork - 5.51% xfs_trans_commit - 5.48% __xfs_trans_commit - 5.35% xfs_log_commit_cil - 3.86% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.70% xfs_trans_alloc 0.52% xfs_trans_reserve - 5.41% xfs_attr_set_args - 5.39% xfs_attr_set_shortform.constprop.0 - 4.46% xfs_trans_commit - 4.46% __xfs_trans_commit - 4.33% xfs_log_commit_cil - 2.74% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath 0.60% xfs_inode_item_format 0.90% xfs_attr_try_sf_addname - 1.99% selinux_inode_init_security - 1.02% security_sid_to_context_force - 1.00% security_sid_to_context_core - 0.92% sidtab_entry_to_string - 0.90% sidtab_sid2str_get 0.59% sidtab_sid2str_put.part.0 - 0.82% selinux_determine_inode_label - 0.77% security_transition_sid 0.70% security_compute_sid.part.0 And fsmark creation rate performance drops by ~25%. The key point to note here is that half the additional overhead comes from adding the attribute fork to the newly created inode. That's crazy, considering we can do this same thing at inode create time with a couple of lines of code and no extra overhead. So, if we know we are going to add an attribute immediately after creating the inode, let's just initialise the attribute fork inside the create transaction and chop that whole chunk of code out of the create fast path. This completely removes the performance drop caused by enabling SELinux, and the profile looks like: - 8.99% xfs_init_security - 9.00% security_inode_init_security - 6.43% xfs_initxattrs - 6.37% xfs_attr_set - 5.45% xfs_attr_set_args - 5.42% xfs_attr_set_shortform.constprop.0 - 4.51% xfs_trans_commit - 4.54% __xfs_trans_commit - 4.59% xfs_log_commit_cil - 2.67% _raw_spin_lock - 3.28% do_raw_spin_lock 3.08% __pv_queued_spin_lock_slowpath 0.66% xfs_inode_item_format - 0.90% xfs_attr_try_sf_addname - 0.60% xfs_trans_alloc - 2.35% selinux_inode_init_security - 1.25% security_sid_to_context_force - 1.21% security_sid_to_context_core - 1.19% sidtab_entry_to_string - 1.20% sidtab_sid2str_get - 0.86% sidtab_sid2str_put.part.0 - 0.62% _raw_spin_lock_irqsave - 0.77% do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.84% selinux_determine_inode_label - 0.83% security_transition_sid 0.86% security_compute_sid.part.0 Which indicates the XFS overhead of creating the selinux xattr has been halved. This doesn't fix the CIL lock contention problem, just means it's not a limiting factor for this workload. Lock contention in the security subsystems is going to be an issue soon, though... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> [djwong: fix compilation error when CONFIG_SECURITY=n] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Gao Xiang <hsiangkao@redhat.com> diff e6a688c3 Mon Mar 22 10:52:03 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: initialise attr fork on inode create When we allocate a new inode, we often need to add an attribute to the inode as part of the create. This can happen as a result of needing to add default ACLs or security labels before the inode is made visible to userspace. This is highly inefficient right now. We do the create transaction to allocate the inode, then we do an "add attr fork" transaction to modify the just created empty inode to set the inode fork offset to allow attributes to be stored, then we go and do the attribute creation. This means 3 transactions instead of 1 to allocate an inode, and this greatly increases the load on the CIL commit code, resulting in excessive contention on the CIL spin locks and performance degradation: 18.99% [kernel] [k] __pv_queued_spin_lock_slowpath 3.57% [kernel] [k] do_raw_spin_lock 2.51% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock 2.48% [kernel] [k] memcpy 2.34% [kernel] [k] xfs_log_commit_cil The typical profile resulting from running fsmark on a selinux enabled filesytem is adds this overhead to the create path: - 15.30% xfs_init_security - 15.23% security_inode_init_security - 13.05% xfs_initxattrs - 12.94% xfs_attr_set - 6.75% xfs_bmap_add_attrfork - 5.51% xfs_trans_commit - 5.48% __xfs_trans_commit - 5.35% xfs_log_commit_cil - 3.86% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.70% xfs_trans_alloc 0.52% xfs_trans_reserve - 5.41% xfs_attr_set_args - 5.39% xfs_attr_set_shortform.constprop.0 - 4.46% xfs_trans_commit - 4.46% __xfs_trans_commit - 4.33% xfs_log_commit_cil - 2.74% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath 0.60% xfs_inode_item_format 0.90% xfs_attr_try_sf_addname - 1.99% selinux_inode_init_security - 1.02% security_sid_to_context_force - 1.00% security_sid_to_context_core - 0.92% sidtab_entry_to_string - 0.90% sidtab_sid2str_get 0.59% sidtab_sid2str_put.part.0 - 0.82% selinux_determine_inode_label - 0.77% security_transition_sid 0.70% security_compute_sid.part.0 And fsmark creation rate performance drops by ~25%. The key point to note here is that half the additional overhead comes from adding the attribute fork to the newly created inode. That's crazy, considering we can do this same thing at inode create time with a couple of lines of code and no extra overhead. So, if we know we are going to add an attribute immediately after creating the inode, let's just initialise the attribute fork inside the create transaction and chop that whole chunk of code out of the create fast path. This completely removes the performance drop caused by enabling SELinux, and the profile looks like: - 8.99% xfs_init_security - 9.00% security_inode_init_security - 6.43% xfs_initxattrs - 6.37% xfs_attr_set - 5.45% xfs_attr_set_args - 5.42% xfs_attr_set_shortform.constprop.0 - 4.51% xfs_trans_commit - 4.54% __xfs_trans_commit - 4.59% xfs_log_commit_cil - 2.67% _raw_spin_lock - 3.28% do_raw_spin_lock 3.08% __pv_queued_spin_lock_slowpath 0.66% xfs_inode_item_format - 0.90% xfs_attr_try_sf_addname - 0.60% xfs_trans_alloc - 2.35% selinux_inode_init_security - 1.25% security_sid_to_context_force - 1.21% security_sid_to_context_core - 1.19% sidtab_entry_to_string - 1.20% sidtab_sid2str_get - 0.86% sidtab_sid2str_put.part.0 - 0.62% _raw_spin_lock_irqsave - 0.77% do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.84% selinux_determine_inode_label - 0.83% security_transition_sid 0.86% security_compute_sid.part.0 Which indicates the XFS overhead of creating the selinux xattr has been halved. This doesn't fix the CIL lock contention problem, just means it's not a limiting factor for this workload. Lock contention in the security subsystems is going to be an issue soon, though... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> [djwong: fix compilation error when CONFIG_SECURITY=n] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Gao Xiang <hsiangkao@redhat.com> diff 8aa921a9 Fri Jan 22 17:48:20 MST 2021 Jeffrey Mitchell <jeffrey.mitchell@starlab.io> xfs: set inode size after creating symlink When XFS creates a new symlink, it writes its size to disk but not to the VFS inode. This causes i_size_read() to return 0 for that symlink until it is re-read from disk, for example when the system is rebooted. I found this inconsistency while protecting directories with eCryptFS. The command "stat path/to/symlink/in/ecryptfs" will report "Size: 0" if the symlink was created after the last reboot on an XFS root. Call i_size_write() in xfs_symlink() Signed-off-by: Jeffrey Mitchell <jeffrey.mitchell@starlab.io> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Brian Foster <bfoster@redhat.com> diff 8aa921a9 Fri Jan 22 17:48:20 MST 2021 Jeffrey Mitchell <jeffrey.mitchell@starlab.io> xfs: set inode size after creating symlink When XFS creates a new symlink, it writes its size to disk but not to the VFS inode. This causes i_size_read() to return 0 for that symlink until it is re-read from disk, for example when the system is rebooted. I found this inconsistency while protecting directories with eCryptFS. The command "stat path/to/symlink/in/ecryptfs" will report "Size: 0" if the symlink was created after the last reboot on an XFS root. Call i_size_write() in xfs_symlink() Signed-off-by: Jeffrey Mitchell <jeffrey.mitchell@starlab.io> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Brian Foster <bfoster@redhat.com> diff 0e3eccce Thu Jan 23 18:01:17 MST 2020 Darrick J. Wong <darrick.wong@oracle.com> xfs: make xfs_buf_read return an error code Convert xfs_buf_read() to return numeric error codes like most everywhere else in xfs. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Dave Chinner <dchinner@redhat.com> diff a8198666 Wed Aug 01 08:20:32 MDT 2018 Brian Foster <bfoster@redhat.com> xfs: automatic dfops inode relogging Inodes that are held across deferred operations are explicitly joined to the dfops structure to ensure appropriate relogging. While inodes are currently joined explicitly, we can detect the conditions that require relogging at dfops finish time by inspecting the transaction item list for inodes with ili_lock_flags == 0. Replace the xfs_defer_ijoin() infrastructure with such detection and automatic relogging of held inodes. This eliminates the need for the per-dfops inode list, replaced by an on-stack variant in xfs_defer_trans_roll(). Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> |
H A D | xfs_symlink.h | diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> |
H A D | xfs_sysfs.c | diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> |
H A D | xfs_trace.c | diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0020a190 Tue Aug 10 19:00:44 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: AIL needs asynchronous CIL forcing The AIL pushing is stalling on log forces when it comes across pinned items. This is happening on removal workloads where the AIL is dominated by stale items that are removed from AIL when the checkpoint that marks the items stale is committed to the journal. This results is relatively few items in the AIL, but those that are are often pinned as directories items are being removed from are still being logged. As a result, many push cycles through the CIL will first issue a blocking log force to unpin the items. This can take some time to complete, with tracing regularly showing push delays of half a second and sometimes up into the range of several seconds. Sequences like this aren't uncommon: .... 399.829437: xfsaild: last lsn 0x11002dd000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 270ms delay> 400.099622: xfsaild: target 0x11002f3600, prev 0x11002f3600, last lsn 0x0 400.099623: xfsaild: first lsn 0x11002f3600 400.099679: xfsaild: last lsn 0x1100305000 count 16 stuck 11 flushing 0 tout 50 <wanted 50ms, got 500ms delay> 400.589348: xfsaild: target 0x110032e600, prev 0x11002f3600, last lsn 0x0 400.589349: xfsaild: first lsn 0x1100305000 400.589595: xfsaild: last lsn 0x110032e600 count 156 stuck 101 flushing 30 tout 50 <wanted 50ms, got 460ms delay> 400.950341: xfsaild: target 0x1100353000, prev 0x110032e600, last lsn 0x0 400.950343: xfsaild: first lsn 0x1100317c00 400.950436: xfsaild: last lsn 0x110033d200 count 105 stuck 101 flushing 0 tout 20 <wanted 20ms, got 200ms delay> 401.142333: xfsaild: target 0x1100361600, prev 0x1100353000, last lsn 0x0 401.142334: xfsaild: first lsn 0x110032e600 401.142535: xfsaild: last lsn 0x1100353000 count 122 stuck 101 flushing 8 tout 10 <wanted 10ms, got 10ms delay> 401.154323: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x1100353000 401.154328: xfsaild: first lsn 0x1100353000 401.154389: xfsaild: last lsn 0x1100353000 count 101 stuck 101 flushing 0 tout 20 <wanted 20ms, got 300ms delay> 401.451525: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 401.451526: xfsaild: first lsn 0x1100353000 401.451804: xfsaild: last lsn 0x1100377200 count 170 stuck 22 flushing 122 tout 50 <wanted 50ms, got 500ms delay> 401.933581: xfsaild: target 0x1100361600, prev 0x1100361600, last lsn 0x0 .... In each of these cases, every AIL pass saw 101 log items stuck on the AIL (pinned) with very few other items being found. Each pass, a log force was issued, and delay between last/first is the sleep time + the sync log force time. Some of these 101 items pinned the tail of the log. The tail of the log does slowly creep forward (first lsn), but the problem is that the log is actually out of reservation space because it's been running so many transactions that stale items that never reach the AIL but consume log space. Hence we have a largely empty AIL, with long term pins on items that pin the tail of the log that don't get pushed frequently enough to keep log space available. The problem is the hundreds of milliseconds that we block in the log force pushing the CIL out to disk. The AIL should not be stalled like this - it needs to run and flush items that are at the tail of the log with minimal latency. What we really need to do is trigger a log flush, but then not wait for it at all - we've already done our waiting for stuff to complete when we backed off prior to the log force being issued. Even if we remove the XFS_LOG_SYNC from the xfs_log_force() call, we still do a blocking flush of the CIL and that is what is causing the issue. Hence we need a new interface for the CIL to trigger an immediate background push of the CIL to get it moving faster but not to wait on that to occur. While the CIL is pushing, the AIL can also be pushing. We already have an internal interface to do this - xlog_cil_push_now() - but we need a wrapper for it to be used externally. xlog_cil_force_seq() can easily be extended to do what we need as it already implements the synchronous CIL push via xlog_cil_push_now(). Add the necessary flags and "push current sequence" semantics to xlog_cil_force_seq() and convert the AIL pushing to use it. One of the complexities here is that the CIL push does not guarantee that the commit record for the CIL checkpoint is written to disk. The current log force ensures this by submitting the current ACTIVE iclog that the commit record was written to. We need the CIL to actually write this commit record to disk for an async push to ensure that the checkpoint actually makes it to disk and unpins the pinned items in the checkpoint on completion. Hence we need to pass down to the CIL push that we are doing an async flush so that it can switch out the commit_iclog if necessary to get written to disk when the commit iclog is finally released. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org> |
H A D | xfs_rtalloc.c | diff ce42b5d3 Sun Dec 17 21:57:22 MST 2023 Christoph Hellwig <hch@lst.de> xfs: return -ENOSPC from xfs_rtallocate_* Just return -ENOSPC instead of returning 0 and setting the return rt extent number to NULLRTEXTNO. This is turn removes all users of NULLRTEXTNO, so remove that as well. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: "Darrick J. Wong" <djwong@kernel.org> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff a6a38f30 Fri Dec 01 10:17:40 MST 2023 Darrick J. Wong <djwong@kernel.org> xfs: make rextslog computation consistent with mkfs There's a weird discrepancy in xfsprogs dating back to the creation of the Linux port -- if there are zero rt extents, mkfs will set sb_rextents and sb_rextslog both to zero: sbp->sb_rextslog = (uint8_t)(rtextents ? libxfs_highbit32((unsigned int)rtextents) : 0); However, that's not the check that xfs_repair uses for nonzero rtblocks: if (sb->sb_rextslog != libxfs_highbit32((unsigned int)sb->sb_rextents)) The difference here is that xfs_highbit32 returns -1 if its argument is zero. Unfortunately, this means that in the weird corner case of a realtime volume shorter than 1 rt extent, xfs_repair will immediately flag a freshly formatted filesystem as corrupt. Because mkfs has been writing ondisk artifacts like this for decades, we have to accept that as "correct". TBH, zero rextslog for zero rtextents makes more sense to me anyway. Regrettably, the superblock verifier checks created in commit copied xfs_repair even though mkfs has been writing out such filesystems for ages. Fix the superblock verifier to accept what mkfs spits out; the userspace version of this patch will have to fix xfs_repair as well. Note that the new helper leaves the zeroday bug where the upper 32 bits of sb_rextents is ripped off and fed to highbit32. This leads to a seriously undersized rt summary file, which immediately breaks mkfs: $ hugedisk.sh foo /dev/sdc $(( 0x100000080 * 4096))B $ /sbin/mkfs.xfs -f /dev/sda -m rmapbt=0,reflink=0 -r rtdev=/dev/mapper/foo meta-data=/dev/sda isize=512 agcount=4, agsize=1298176 blks = sectsz=512 attr=2, projid32bit=1 = crc=1 finobt=1, sparse=1, rmapbt=0 = reflink=0 bigtime=1 inobtcount=1 nrext64=1 data = bsize=4096 blocks=5192704, imaxpct=25 = sunit=0 swidth=0 blks naming =version 2 bsize=4096 ascii-ci=0, ftype=1 log =internal log bsize=4096 blocks=16384, version=2 = sectsz=512 sunit=0 blks, lazy-count=1 realtime =/dev/mapper/foo extsz=4096 blocks=4294967424, rtextents=4294967424 Discarding blocks...Done. mkfs.xfs: Error initializing the realtime space [117 - Structure needs cleaning] The next patch will drop support for rt volumes with fewer than 1 or more than 2^32-1 rt extents, since they've clearly been broken forever. Fixes: f8e566c0f5e1f ("xfs: validate the realtime geometry in xfs_validate_sb_common") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff a6a38f30 Fri Dec 01 10:17:40 MST 2023 Darrick J. Wong <djwong@kernel.org> xfs: make rextslog computation consistent with mkfs There's a weird discrepancy in xfsprogs dating back to the creation of the Linux port -- if there are zero rt extents, mkfs will set sb_rextents and sb_rextslog both to zero: sbp->sb_rextslog = (uint8_t)(rtextents ? libxfs_highbit32((unsigned int)rtextents) : 0); However, that's not the check that xfs_repair uses for nonzero rtblocks: if (sb->sb_rextslog != libxfs_highbit32((unsigned int)sb->sb_rextents)) The difference here is that xfs_highbit32 returns -1 if its argument is zero. Unfortunately, this means that in the weird corner case of a realtime volume shorter than 1 rt extent, xfs_repair will immediately flag a freshly formatted filesystem as corrupt. Because mkfs has been writing ondisk artifacts like this for decades, we have to accept that as "correct". TBH, zero rextslog for zero rtextents makes more sense to me anyway. Regrettably, the superblock verifier checks created in commit copied xfs_repair even though mkfs has been writing out such filesystems for ages. Fix the superblock verifier to accept what mkfs spits out; the userspace version of this patch will have to fix xfs_repair as well. Note that the new helper leaves the zeroday bug where the upper 32 bits of sb_rextents is ripped off and fed to highbit32. This leads to a seriously undersized rt summary file, which immediately breaks mkfs: $ hugedisk.sh foo /dev/sdc $(( 0x100000080 * 4096))B $ /sbin/mkfs.xfs -f /dev/sda -m rmapbt=0,reflink=0 -r rtdev=/dev/mapper/foo meta-data=/dev/sda isize=512 agcount=4, agsize=1298176 blks = sectsz=512 attr=2, projid32bit=1 = crc=1 finobt=1, sparse=1, rmapbt=0 = reflink=0 bigtime=1 inobtcount=1 nrext64=1 data = bsize=4096 blocks=5192704, imaxpct=25 = sunit=0 swidth=0 blks naming =version 2 bsize=4096 ascii-ci=0, ftype=1 log =internal log bsize=4096 blocks=16384, version=2 = sectsz=512 sunit=0 blks, lazy-count=1 realtime =/dev/mapper/foo extsz=4096 blocks=4294967424, rtextents=4294967424 Discarding blocks...Done. mkfs.xfs: Error initializing the realtime space [117 - Structure needs cleaning] The next patch will drop support for rt volumes with fewer than 1 or more than 2^32-1 rt extents, since they've clearly been broken forever. Fixes: f8e566c0f5e1f ("xfs: validate the realtime geometry in xfs_validate_sb_common") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff a6a38f30 Fri Dec 01 10:17:40 MST 2023 Darrick J. Wong <djwong@kernel.org> xfs: make rextslog computation consistent with mkfs There's a weird discrepancy in xfsprogs dating back to the creation of the Linux port -- if there are zero rt extents, mkfs will set sb_rextents and sb_rextslog both to zero: sbp->sb_rextslog = (uint8_t)(rtextents ? libxfs_highbit32((unsigned int)rtextents) : 0); However, that's not the check that xfs_repair uses for nonzero rtblocks: if (sb->sb_rextslog != libxfs_highbit32((unsigned int)sb->sb_rextents)) The difference here is that xfs_highbit32 returns -1 if its argument is zero. Unfortunately, this means that in the weird corner case of a realtime volume shorter than 1 rt extent, xfs_repair will immediately flag a freshly formatted filesystem as corrupt. Because mkfs has been writing ondisk artifacts like this for decades, we have to accept that as "correct". TBH, zero rextslog for zero rtextents makes more sense to me anyway. Regrettably, the superblock verifier checks created in commit copied xfs_repair even though mkfs has been writing out such filesystems for ages. Fix the superblock verifier to accept what mkfs spits out; the userspace version of this patch will have to fix xfs_repair as well. Note that the new helper leaves the zeroday bug where the upper 32 bits of sb_rextents is ripped off and fed to highbit32. This leads to a seriously undersized rt summary file, which immediately breaks mkfs: $ hugedisk.sh foo /dev/sdc $(( 0x100000080 * 4096))B $ /sbin/mkfs.xfs -f /dev/sda -m rmapbt=0,reflink=0 -r rtdev=/dev/mapper/foo meta-data=/dev/sda isize=512 agcount=4, agsize=1298176 blks = sectsz=512 attr=2, projid32bit=1 = crc=1 finobt=1, sparse=1, rmapbt=0 = reflink=0 bigtime=1 inobtcount=1 nrext64=1 data = bsize=4096 blocks=5192704, imaxpct=25 = sunit=0 swidth=0 blks naming =version 2 bsize=4096 ascii-ci=0, ftype=1 log =internal log bsize=4096 blocks=16384, version=2 = sectsz=512 sunit=0 blks, lazy-count=1 realtime =/dev/mapper/foo extsz=4096 blocks=4294967424, rtextents=4294967424 Discarding blocks...Done. mkfs.xfs: Error initializing the realtime space [117 - Structure needs cleaning] The next patch will drop support for rt volumes with fewer than 1 or more than 2^32-1 rt extents, since they've clearly been broken forever. Fixes: f8e566c0f5e1f ("xfs: validate the realtime geometry in xfs_validate_sb_common") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff a6a38f30 Fri Dec 01 10:17:40 MST 2023 Darrick J. Wong <djwong@kernel.org> xfs: make rextslog computation consistent with mkfs There's a weird discrepancy in xfsprogs dating back to the creation of the Linux port -- if there are zero rt extents, mkfs will set sb_rextents and sb_rextslog both to zero: sbp->sb_rextslog = (uint8_t)(rtextents ? libxfs_highbit32((unsigned int)rtextents) : 0); However, that's not the check that xfs_repair uses for nonzero rtblocks: if (sb->sb_rextslog != libxfs_highbit32((unsigned int)sb->sb_rextents)) The difference here is that xfs_highbit32 returns -1 if its argument is zero. Unfortunately, this means that in the weird corner case of a realtime volume shorter than 1 rt extent, xfs_repair will immediately flag a freshly formatted filesystem as corrupt. Because mkfs has been writing ondisk artifacts like this for decades, we have to accept that as "correct". TBH, zero rextslog for zero rtextents makes more sense to me anyway. Regrettably, the superblock verifier checks created in commit copied xfs_repair even though mkfs has been writing out such filesystems for ages. Fix the superblock verifier to accept what mkfs spits out; the userspace version of this patch will have to fix xfs_repair as well. Note that the new helper leaves the zeroday bug where the upper 32 bits of sb_rextents is ripped off and fed to highbit32. This leads to a seriously undersized rt summary file, which immediately breaks mkfs: $ hugedisk.sh foo /dev/sdc $(( 0x100000080 * 4096))B $ /sbin/mkfs.xfs -f /dev/sda -m rmapbt=0,reflink=0 -r rtdev=/dev/mapper/foo meta-data=/dev/sda isize=512 agcount=4, agsize=1298176 blks = sectsz=512 attr=2, projid32bit=1 = crc=1 finobt=1, sparse=1, rmapbt=0 = reflink=0 bigtime=1 inobtcount=1 nrext64=1 data = bsize=4096 blocks=5192704, imaxpct=25 = sunit=0 swidth=0 blks naming =version 2 bsize=4096 ascii-ci=0, ftype=1 log =internal log bsize=4096 blocks=16384, version=2 = sectsz=512 sunit=0 blks, lazy-count=1 realtime =/dev/mapper/foo extsz=4096 blocks=4294967424, rtextents=4294967424 Discarding blocks...Done. mkfs.xfs: Error initializing the realtime space [117 - Structure needs cleaning] The next patch will drop support for rt volumes with fewer than 1 or more than 2^32-1 rt extents, since they've clearly been broken forever. Fixes: f8e566c0f5e1f ("xfs: validate the realtime geometry in xfs_validate_sb_common") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff a6a38f30 Fri Dec 01 10:17:40 MST 2023 Darrick J. Wong <djwong@kernel.org> xfs: make rextslog computation consistent with mkfs There's a weird discrepancy in xfsprogs dating back to the creation of the Linux port -- if there are zero rt extents, mkfs will set sb_rextents and sb_rextslog both to zero: sbp->sb_rextslog = (uint8_t)(rtextents ? libxfs_highbit32((unsigned int)rtextents) : 0); However, that's not the check that xfs_repair uses for nonzero rtblocks: if (sb->sb_rextslog != libxfs_highbit32((unsigned int)sb->sb_rextents)) The difference here is that xfs_highbit32 returns -1 if its argument is zero. Unfortunately, this means that in the weird corner case of a realtime volume shorter than 1 rt extent, xfs_repair will immediately flag a freshly formatted filesystem as corrupt. Because mkfs has been writing ondisk artifacts like this for decades, we have to accept that as "correct". TBH, zero rextslog for zero rtextents makes more sense to me anyway. Regrettably, the superblock verifier checks created in commit copied xfs_repair even though mkfs has been writing out such filesystems for ages. Fix the superblock verifier to accept what mkfs spits out; the userspace version of this patch will have to fix xfs_repair as well. Note that the new helper leaves the zeroday bug where the upper 32 bits of sb_rextents is ripped off and fed to highbit32. This leads to a seriously undersized rt summary file, which immediately breaks mkfs: $ hugedisk.sh foo /dev/sdc $(( 0x100000080 * 4096))B $ /sbin/mkfs.xfs -f /dev/sda -m rmapbt=0,reflink=0 -r rtdev=/dev/mapper/foo meta-data=/dev/sda isize=512 agcount=4, agsize=1298176 blks = sectsz=512 attr=2, projid32bit=1 = crc=1 finobt=1, sparse=1, rmapbt=0 = reflink=0 bigtime=1 inobtcount=1 nrext64=1 data = bsize=4096 blocks=5192704, imaxpct=25 = sunit=0 swidth=0 blks naming =version 2 bsize=4096 ascii-ci=0, ftype=1 log =internal log bsize=4096 blocks=16384, version=2 = sectsz=512 sunit=0 blks, lazy-count=1 realtime =/dev/mapper/foo extsz=4096 blocks=4294967424, rtextents=4294967424 Discarding blocks...Done. mkfs.xfs: Error initializing the realtime space [117 - Structure needs cleaning] The next patch will drop support for rt volumes with fewer than 1 or more than 2^32-1 rt extents, since they've clearly been broken forever. Fixes: f8e566c0f5e1f ("xfs: validate the realtime geometry in xfs_validate_sb_common") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff a6a38f30 Fri Dec 01 10:17:40 MST 2023 Darrick J. Wong <djwong@kernel.org> xfs: make rextslog computation consistent with mkfs There's a weird discrepancy in xfsprogs dating back to the creation of the Linux port -- if there are zero rt extents, mkfs will set sb_rextents and sb_rextslog both to zero: sbp->sb_rextslog = (uint8_t)(rtextents ? libxfs_highbit32((unsigned int)rtextents) : 0); However, that's not the check that xfs_repair uses for nonzero rtblocks: if (sb->sb_rextslog != libxfs_highbit32((unsigned int)sb->sb_rextents)) The difference here is that xfs_highbit32 returns -1 if its argument is zero. Unfortunately, this means that in the weird corner case of a realtime volume shorter than 1 rt extent, xfs_repair will immediately flag a freshly formatted filesystem as corrupt. Because mkfs has been writing ondisk artifacts like this for decades, we have to accept that as "correct". TBH, zero rextslog for zero rtextents makes more sense to me anyway. Regrettably, the superblock verifier checks created in commit copied xfs_repair even though mkfs has been writing out such filesystems for ages. Fix the superblock verifier to accept what mkfs spits out; the userspace version of this patch will have to fix xfs_repair as well. Note that the new helper leaves the zeroday bug where the upper 32 bits of sb_rextents is ripped off and fed to highbit32. This leads to a seriously undersized rt summary file, which immediately breaks mkfs: $ hugedisk.sh foo /dev/sdc $(( 0x100000080 * 4096))B $ /sbin/mkfs.xfs -f /dev/sda -m rmapbt=0,reflink=0 -r rtdev=/dev/mapper/foo meta-data=/dev/sda isize=512 agcount=4, agsize=1298176 blks = sectsz=512 attr=2, projid32bit=1 = crc=1 finobt=1, sparse=1, rmapbt=0 = reflink=0 bigtime=1 inobtcount=1 nrext64=1 data = bsize=4096 blocks=5192704, imaxpct=25 = sunit=0 swidth=0 blks naming =version 2 bsize=4096 ascii-ci=0, ftype=1 log =internal log bsize=4096 blocks=16384, version=2 = sectsz=512 sunit=0 blks, lazy-count=1 realtime =/dev/mapper/foo extsz=4096 blocks=4294967424, rtextents=4294967424 Discarding blocks...Done. mkfs.xfs: Error initializing the realtime space [117 - Structure needs cleaning] The next patch will drop support for rt volumes with fewer than 1 or more than 2^32-1 rt extents, since they've clearly been broken forever. Fixes: f8e566c0f5e1f ("xfs: validate the realtime geometry in xfs_validate_sb_common") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff a6a38f30 Fri Dec 01 10:17:40 MST 2023 Darrick J. Wong <djwong@kernel.org> xfs: make rextslog computation consistent with mkfs There's a weird discrepancy in xfsprogs dating back to the creation of the Linux port -- if there are zero rt extents, mkfs will set sb_rextents and sb_rextslog both to zero: sbp->sb_rextslog = (uint8_t)(rtextents ? libxfs_highbit32((unsigned int)rtextents) : 0); However, that's not the check that xfs_repair uses for nonzero rtblocks: if (sb->sb_rextslog != libxfs_highbit32((unsigned int)sb->sb_rextents)) The difference here is that xfs_highbit32 returns -1 if its argument is zero. Unfortunately, this means that in the weird corner case of a realtime volume shorter than 1 rt extent, xfs_repair will immediately flag a freshly formatted filesystem as corrupt. Because mkfs has been writing ondisk artifacts like this for decades, we have to accept that as "correct". TBH, zero rextslog for zero rtextents makes more sense to me anyway. Regrettably, the superblock verifier checks created in commit copied xfs_repair even though mkfs has been writing out such filesystems for ages. Fix the superblock verifier to accept what mkfs spits out; the userspace version of this patch will have to fix xfs_repair as well. Note that the new helper leaves the zeroday bug where the upper 32 bits of sb_rextents is ripped off and fed to highbit32. This leads to a seriously undersized rt summary file, which immediately breaks mkfs: $ hugedisk.sh foo /dev/sdc $(( 0x100000080 * 4096))B $ /sbin/mkfs.xfs -f /dev/sda -m rmapbt=0,reflink=0 -r rtdev=/dev/mapper/foo meta-data=/dev/sda isize=512 agcount=4, agsize=1298176 blks = sectsz=512 attr=2, projid32bit=1 = crc=1 finobt=1, sparse=1, rmapbt=0 = reflink=0 bigtime=1 inobtcount=1 nrext64=1 data = bsize=4096 blocks=5192704, imaxpct=25 = sunit=0 swidth=0 blks naming =version 2 bsize=4096 ascii-ci=0, ftype=1 log =internal log bsize=4096 blocks=16384, version=2 = sectsz=512 sunit=0 blks, lazy-count=1 realtime =/dev/mapper/foo extsz=4096 blocks=4294967424, rtextents=4294967424 Discarding blocks...Done. mkfs.xfs: Error initializing the realtime space [117 - Structure needs cleaning] The next patch will drop support for rt volumes with fewer than 1 or more than 2^32-1 rt extents, since they've clearly been broken forever. Fixes: f8e566c0f5e1f ("xfs: validate the realtime geometry in xfs_validate_sb_common") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff a6a38f30 Fri Dec 01 10:17:40 MST 2023 Darrick J. Wong <djwong@kernel.org> xfs: make rextslog computation consistent with mkfs There's a weird discrepancy in xfsprogs dating back to the creation of the Linux port -- if there are zero rt extents, mkfs will set sb_rextents and sb_rextslog both to zero: sbp->sb_rextslog = (uint8_t)(rtextents ? libxfs_highbit32((unsigned int)rtextents) : 0); However, that's not the check that xfs_repair uses for nonzero rtblocks: if (sb->sb_rextslog != libxfs_highbit32((unsigned int)sb->sb_rextents)) The difference here is that xfs_highbit32 returns -1 if its argument is zero. Unfortunately, this means that in the weird corner case of a realtime volume shorter than 1 rt extent, xfs_repair will immediately flag a freshly formatted filesystem as corrupt. Because mkfs has been writing ondisk artifacts like this for decades, we have to accept that as "correct". TBH, zero rextslog for zero rtextents makes more sense to me anyway. Regrettably, the superblock verifier checks created in commit copied xfs_repair even though mkfs has been writing out such filesystems for ages. Fix the superblock verifier to accept what mkfs spits out; the userspace version of this patch will have to fix xfs_repair as well. Note that the new helper leaves the zeroday bug where the upper 32 bits of sb_rextents is ripped off and fed to highbit32. This leads to a seriously undersized rt summary file, which immediately breaks mkfs: $ hugedisk.sh foo /dev/sdc $(( 0x100000080 * 4096))B $ /sbin/mkfs.xfs -f /dev/sda -m rmapbt=0,reflink=0 -r rtdev=/dev/mapper/foo meta-data=/dev/sda isize=512 agcount=4, agsize=1298176 blks = sectsz=512 attr=2, projid32bit=1 = crc=1 finobt=1, sparse=1, rmapbt=0 = reflink=0 bigtime=1 inobtcount=1 nrext64=1 data = bsize=4096 blocks=5192704, imaxpct=25 = sunit=0 swidth=0 blks naming =version 2 bsize=4096 ascii-ci=0, ftype=1 log =internal log bsize=4096 blocks=16384, version=2 = sectsz=512 sunit=0 blks, lazy-count=1 realtime =/dev/mapper/foo extsz=4096 blocks=4294967424, rtextents=4294967424 Discarding blocks...Done. mkfs.xfs: Error initializing the realtime space [117 - Structure needs cleaning] The next patch will drop support for rt volumes with fewer than 1 or more than 2^32-1 rt extents, since they've clearly been broken forever. Fixes: f8e566c0f5e1f ("xfs: validate the realtime geometry in xfs_validate_sb_common") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff a6a38f30 Fri Dec 01 10:17:40 MST 2023 Darrick J. Wong <djwong@kernel.org> xfs: make rextslog computation consistent with mkfs There's a weird discrepancy in xfsprogs dating back to the creation of the Linux port -- if there are zero rt extents, mkfs will set sb_rextents and sb_rextslog both to zero: sbp->sb_rextslog = (uint8_t)(rtextents ? libxfs_highbit32((unsigned int)rtextents) : 0); However, that's not the check that xfs_repair uses for nonzero rtblocks: if (sb->sb_rextslog != libxfs_highbit32((unsigned int)sb->sb_rextents)) The difference here is that xfs_highbit32 returns -1 if its argument is zero. Unfortunately, this means that in the weird corner case of a realtime volume shorter than 1 rt extent, xfs_repair will immediately flag a freshly formatted filesystem as corrupt. Because mkfs has been writing ondisk artifacts like this for decades, we have to accept that as "correct". TBH, zero rextslog for zero rtextents makes more sense to me anyway. Regrettably, the superblock verifier checks created in commit copied xfs_repair even though mkfs has been writing out such filesystems for ages. Fix the superblock verifier to accept what mkfs spits out; the userspace version of this patch will have to fix xfs_repair as well. Note that the new helper leaves the zeroday bug where the upper 32 bits of sb_rextents is ripped off and fed to highbit32. This leads to a seriously undersized rt summary file, which immediately breaks mkfs: $ hugedisk.sh foo /dev/sdc $(( 0x100000080 * 4096))B $ /sbin/mkfs.xfs -f /dev/sda -m rmapbt=0,reflink=0 -r rtdev=/dev/mapper/foo meta-data=/dev/sda isize=512 agcount=4, agsize=1298176 blks = sectsz=512 attr=2, projid32bit=1 = crc=1 finobt=1, sparse=1, rmapbt=0 = reflink=0 bigtime=1 inobtcount=1 nrext64=1 data = bsize=4096 blocks=5192704, imaxpct=25 = sunit=0 swidth=0 blks naming =version 2 bsize=4096 ascii-ci=0, ftype=1 log =internal log bsize=4096 blocks=16384, version=2 = sectsz=512 sunit=0 blks, lazy-count=1 realtime =/dev/mapper/foo extsz=4096 blocks=4294967424, rtextents=4294967424 Discarding blocks...Done. mkfs.xfs: Error initializing the realtime space [117 - Structure needs cleaning] The next patch will drop support for rt volumes with fewer than 1 or more than 2^32-1 rt extents, since they've clearly been broken forever. Fixes: f8e566c0f5e1f ("xfs: validate the realtime geometry in xfs_validate_sb_common") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff a6a38f30 Fri Dec 01 10:17:40 MST 2023 Darrick J. Wong <djwong@kernel.org> xfs: make rextslog computation consistent with mkfs There's a weird discrepancy in xfsprogs dating back to the creation of the Linux port -- if there are zero rt extents, mkfs will set sb_rextents and sb_rextslog both to zero: sbp->sb_rextslog = (uint8_t)(rtextents ? libxfs_highbit32((unsigned int)rtextents) : 0); However, that's not the check that xfs_repair uses for nonzero rtblocks: if (sb->sb_rextslog != libxfs_highbit32((unsigned int)sb->sb_rextents)) The difference here is that xfs_highbit32 returns -1 if its argument is zero. Unfortunately, this means that in the weird corner case of a realtime volume shorter than 1 rt extent, xfs_repair will immediately flag a freshly formatted filesystem as corrupt. Because mkfs has been writing ondisk artifacts like this for decades, we have to accept that as "correct". TBH, zero rextslog for zero rtextents makes more sense to me anyway. Regrettably, the superblock verifier checks created in commit copied xfs_repair even though mkfs has been writing out such filesystems for ages. Fix the superblock verifier to accept what mkfs spits out; the userspace version of this patch will have to fix xfs_repair as well. Note that the new helper leaves the zeroday bug where the upper 32 bits of sb_rextents is ripped off and fed to highbit32. This leads to a seriously undersized rt summary file, which immediately breaks mkfs: $ hugedisk.sh foo /dev/sdc $(( 0x100000080 * 4096))B $ /sbin/mkfs.xfs -f /dev/sda -m rmapbt=0,reflink=0 -r rtdev=/dev/mapper/foo meta-data=/dev/sda isize=512 agcount=4, agsize=1298176 blks = sectsz=512 attr=2, projid32bit=1 = crc=1 finobt=1, sparse=1, rmapbt=0 = reflink=0 bigtime=1 inobtcount=1 nrext64=1 data = bsize=4096 blocks=5192704, imaxpct=25 = sunit=0 swidth=0 blks naming =version 2 bsize=4096 ascii-ci=0, ftype=1 log =internal log bsize=4096 blocks=16384, version=2 = sectsz=512 sunit=0 blks, lazy-count=1 realtime =/dev/mapper/foo extsz=4096 blocks=4294967424, rtextents=4294967424 Discarding blocks...Done. mkfs.xfs: Error initializing the realtime space [117 - Structure needs cleaning] The next patch will drop support for rt volumes with fewer than 1 or more than 2^32-1 rt extents, since they've clearly been broken forever. Fixes: f8e566c0f5e1f ("xfs: validate the realtime geometry in xfs_validate_sb_common") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> |
H A D | xfs_stats.h | diff 18a1e644 Thu Feb 22 01:43:40 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: define an in-memory btree for storing refcount bag info during repairs Create a new in-memory btree type so that we can store refcount bag info in a much more memory-efficient and performant format. Recall that the refcount recordset regenerator computes the new recordset from browsing the rmap records. Let's say that the rmap records are: {agbno: 10, length: 40, ...} {agbno: 11, length: 3, ...} {agbno: 12, length: 20, ...} {agbno: 15, length: 1, ...} It is convenient to have a data structure that could quickly tell us the refcount for an arbitrary agbno without wasting memory. An array or a list could do that pretty easily. List suck because of the pointer overhead. xfarrays are a lot more compact, but we want to minimize sparse holes in the xfarray to constrain memory usage. Maintaining any kind of record order isn't needed for correctness, so I created the "rcbag", which is shorthand for an unordered list of (excerpted) reverse mappings. So we add the first rmap to the rcbag, and it looks like: 0: {agbno: 10, length: 40} The refcount for agbno 10 is 1. Then we move on to block 11, so we add the second rmap: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} The refcount for agbno 11 is 2. We move on to block 12, so we add the third: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} 2: {agbno: 12, length: 20} The refcount for agbno 12 and 13 is 3. We move on to block 14, and remove the second rmap: 0: {agbno: 10, length: 40} 1: NULL 2: {agbno: 12, length: 20} The refcount for agbno 14 is 2. We move on to block 15, and add the last rmap. But we don't care where it is and we don't want to expand the array so we put it in slot 1: 0: {agbno: 10, length: 40} 1: {agbno: 15, length: 1} 2: {agbno: 12, length: 20} The refcount for block 15 is 3. Notice how order doesn't matter in this list? That's why repair uses an unordered list, or "bag". The data structure is not a set because it does not guarantee uniqueness. That said, adding and removing specific items is now an O(n) operation because we have no idea where that item might be in the list. Overall, the runtime is O(n^2) which is bad. I realized that I could easily refactor the btree code and reimplement the refcount bag with an xfbtree. Adding and removing is now O(log2 n), so the runtime is at least O(n log2 n), which is much faster. In the end, the rcbag becomes a sorted list, but that's merely a detail of the implementation. The repair code doesn't care. (Note: That horrible xfs_db bmap_inflate command can be used to exercise this sort of rcbag insanity by cranking up refcounts quickly.) Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 18a1e644 Thu Feb 22 01:43:40 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: define an in-memory btree for storing refcount bag info during repairs Create a new in-memory btree type so that we can store refcount bag info in a much more memory-efficient and performant format. Recall that the refcount recordset regenerator computes the new recordset from browsing the rmap records. Let's say that the rmap records are: {agbno: 10, length: 40, ...} {agbno: 11, length: 3, ...} {agbno: 12, length: 20, ...} {agbno: 15, length: 1, ...} It is convenient to have a data structure that could quickly tell us the refcount for an arbitrary agbno without wasting memory. An array or a list could do that pretty easily. List suck because of the pointer overhead. xfarrays are a lot more compact, but we want to minimize sparse holes in the xfarray to constrain memory usage. Maintaining any kind of record order isn't needed for correctness, so I created the "rcbag", which is shorthand for an unordered list of (excerpted) reverse mappings. So we add the first rmap to the rcbag, and it looks like: 0: {agbno: 10, length: 40} The refcount for agbno 10 is 1. Then we move on to block 11, so we add the second rmap: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} The refcount for agbno 11 is 2. We move on to block 12, so we add the third: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} 2: {agbno: 12, length: 20} The refcount for agbno 12 and 13 is 3. We move on to block 14, and remove the second rmap: 0: {agbno: 10, length: 40} 1: NULL 2: {agbno: 12, length: 20} The refcount for agbno 14 is 2. We move on to block 15, and add the last rmap. But we don't care where it is and we don't want to expand the array so we put it in slot 1: 0: {agbno: 10, length: 40} 1: {agbno: 15, length: 1} 2: {agbno: 12, length: 20} The refcount for block 15 is 3. Notice how order doesn't matter in this list? That's why repair uses an unordered list, or "bag". The data structure is not a set because it does not guarantee uniqueness. That said, adding and removing specific items is now an O(n) operation because we have no idea where that item might be in the list. Overall, the runtime is O(n^2) which is bad. I realized that I could easily refactor the btree code and reimplement the refcount bag with an xfbtree. Adding and removing is now O(log2 n), so the runtime is at least O(n log2 n), which is much faster. In the end, the rcbag becomes a sorted list, but that's merely a detail of the implementation. The repair code doesn't care. (Note: That horrible xfs_db bmap_inflate command can be used to exercise this sort of rcbag insanity by cranking up refcounts quickly.) Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 18a1e644 Thu Feb 22 01:43:40 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: define an in-memory btree for storing refcount bag info during repairs Create a new in-memory btree type so that we can store refcount bag info in a much more memory-efficient and performant format. Recall that the refcount recordset regenerator computes the new recordset from browsing the rmap records. Let's say that the rmap records are: {agbno: 10, length: 40, ...} {agbno: 11, length: 3, ...} {agbno: 12, length: 20, ...} {agbno: 15, length: 1, ...} It is convenient to have a data structure that could quickly tell us the refcount for an arbitrary agbno without wasting memory. An array or a list could do that pretty easily. List suck because of the pointer overhead. xfarrays are a lot more compact, but we want to minimize sparse holes in the xfarray to constrain memory usage. Maintaining any kind of record order isn't needed for correctness, so I created the "rcbag", which is shorthand for an unordered list of (excerpted) reverse mappings. So we add the first rmap to the rcbag, and it looks like: 0: {agbno: 10, length: 40} The refcount for agbno 10 is 1. Then we move on to block 11, so we add the second rmap: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} The refcount for agbno 11 is 2. We move on to block 12, so we add the third: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} 2: {agbno: 12, length: 20} The refcount for agbno 12 and 13 is 3. We move on to block 14, and remove the second rmap: 0: {agbno: 10, length: 40} 1: NULL 2: {agbno: 12, length: 20} The refcount for agbno 14 is 2. We move on to block 15, and add the last rmap. But we don't care where it is and we don't want to expand the array so we put it in slot 1: 0: {agbno: 10, length: 40} 1: {agbno: 15, length: 1} 2: {agbno: 12, length: 20} The refcount for block 15 is 3. Notice how order doesn't matter in this list? That's why repair uses an unordered list, or "bag". The data structure is not a set because it does not guarantee uniqueness. That said, adding and removing specific items is now an O(n) operation because we have no idea where that item might be in the list. Overall, the runtime is O(n^2) which is bad. I realized that I could easily refactor the btree code and reimplement the refcount bag with an xfbtree. Adding and removing is now O(log2 n), so the runtime is at least O(n log2 n), which is much faster. In the end, the rcbag becomes a sorted list, but that's merely a detail of the implementation. The repair code doesn't care. (Note: That horrible xfs_db bmap_inflate command can be used to exercise this sort of rcbag insanity by cranking up refcounts quickly.) Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 18a1e644 Thu Feb 22 01:43:40 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: define an in-memory btree for storing refcount bag info during repairs Create a new in-memory btree type so that we can store refcount bag info in a much more memory-efficient and performant format. Recall that the refcount recordset regenerator computes the new recordset from browsing the rmap records. Let's say that the rmap records are: {agbno: 10, length: 40, ...} {agbno: 11, length: 3, ...} {agbno: 12, length: 20, ...} {agbno: 15, length: 1, ...} It is convenient to have a data structure that could quickly tell us the refcount for an arbitrary agbno without wasting memory. An array or a list could do that pretty easily. List suck because of the pointer overhead. xfarrays are a lot more compact, but we want to minimize sparse holes in the xfarray to constrain memory usage. Maintaining any kind of record order isn't needed for correctness, so I created the "rcbag", which is shorthand for an unordered list of (excerpted) reverse mappings. So we add the first rmap to the rcbag, and it looks like: 0: {agbno: 10, length: 40} The refcount for agbno 10 is 1. Then we move on to block 11, so we add the second rmap: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} The refcount for agbno 11 is 2. We move on to block 12, so we add the third: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} 2: {agbno: 12, length: 20} The refcount for agbno 12 and 13 is 3. We move on to block 14, and remove the second rmap: 0: {agbno: 10, length: 40} 1: NULL 2: {agbno: 12, length: 20} The refcount for agbno 14 is 2. We move on to block 15, and add the last rmap. But we don't care where it is and we don't want to expand the array so we put it in slot 1: 0: {agbno: 10, length: 40} 1: {agbno: 15, length: 1} 2: {agbno: 12, length: 20} The refcount for block 15 is 3. Notice how order doesn't matter in this list? That's why repair uses an unordered list, or "bag". The data structure is not a set because it does not guarantee uniqueness. That said, adding and removing specific items is now an O(n) operation because we have no idea where that item might be in the list. Overall, the runtime is O(n^2) which is bad. I realized that I could easily refactor the btree code and reimplement the refcount bag with an xfbtree. Adding and removing is now O(log2 n), so the runtime is at least O(n log2 n), which is much faster. In the end, the rcbag becomes a sorted list, but that's merely a detail of the implementation. The repair code doesn't care. (Note: That horrible xfs_db bmap_inflate command can be used to exercise this sort of rcbag insanity by cranking up refcounts quickly.) Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 18a1e644 Thu Feb 22 01:43:40 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: define an in-memory btree for storing refcount bag info during repairs Create a new in-memory btree type so that we can store refcount bag info in a much more memory-efficient and performant format. Recall that the refcount recordset regenerator computes the new recordset from browsing the rmap records. Let's say that the rmap records are: {agbno: 10, length: 40, ...} {agbno: 11, length: 3, ...} {agbno: 12, length: 20, ...} {agbno: 15, length: 1, ...} It is convenient to have a data structure that could quickly tell us the refcount for an arbitrary agbno without wasting memory. An array or a list could do that pretty easily. List suck because of the pointer overhead. xfarrays are a lot more compact, but we want to minimize sparse holes in the xfarray to constrain memory usage. Maintaining any kind of record order isn't needed for correctness, so I created the "rcbag", which is shorthand for an unordered list of (excerpted) reverse mappings. So we add the first rmap to the rcbag, and it looks like: 0: {agbno: 10, length: 40} The refcount for agbno 10 is 1. Then we move on to block 11, so we add the second rmap: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} The refcount for agbno 11 is 2. We move on to block 12, so we add the third: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} 2: {agbno: 12, length: 20} The refcount for agbno 12 and 13 is 3. We move on to block 14, and remove the second rmap: 0: {agbno: 10, length: 40} 1: NULL 2: {agbno: 12, length: 20} The refcount for agbno 14 is 2. We move on to block 15, and add the last rmap. But we don't care where it is and we don't want to expand the array so we put it in slot 1: 0: {agbno: 10, length: 40} 1: {agbno: 15, length: 1} 2: {agbno: 12, length: 20} The refcount for block 15 is 3. Notice how order doesn't matter in this list? That's why repair uses an unordered list, or "bag". The data structure is not a set because it does not guarantee uniqueness. That said, adding and removing specific items is now an O(n) operation because we have no idea where that item might be in the list. Overall, the runtime is O(n^2) which is bad. I realized that I could easily refactor the btree code and reimplement the refcount bag with an xfbtree. Adding and removing is now O(log2 n), so the runtime is at least O(n log2 n), which is much faster. In the end, the rcbag becomes a sorted list, but that's merely a detail of the implementation. The repair code doesn't care. (Note: That horrible xfs_db bmap_inflate command can be used to exercise this sort of rcbag insanity by cranking up refcounts quickly.) Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> |
H A D | xfs_stats.c | diff 18a1e644 Thu Feb 22 01:43:40 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: define an in-memory btree for storing refcount bag info during repairs Create a new in-memory btree type so that we can store refcount bag info in a much more memory-efficient and performant format. Recall that the refcount recordset regenerator computes the new recordset from browsing the rmap records. Let's say that the rmap records are: {agbno: 10, length: 40, ...} {agbno: 11, length: 3, ...} {agbno: 12, length: 20, ...} {agbno: 15, length: 1, ...} It is convenient to have a data structure that could quickly tell us the refcount for an arbitrary agbno without wasting memory. An array or a list could do that pretty easily. List suck because of the pointer overhead. xfarrays are a lot more compact, but we want to minimize sparse holes in the xfarray to constrain memory usage. Maintaining any kind of record order isn't needed for correctness, so I created the "rcbag", which is shorthand for an unordered list of (excerpted) reverse mappings. So we add the first rmap to the rcbag, and it looks like: 0: {agbno: 10, length: 40} The refcount for agbno 10 is 1. Then we move on to block 11, so we add the second rmap: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} The refcount for agbno 11 is 2. We move on to block 12, so we add the third: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} 2: {agbno: 12, length: 20} The refcount for agbno 12 and 13 is 3. We move on to block 14, and remove the second rmap: 0: {agbno: 10, length: 40} 1: NULL 2: {agbno: 12, length: 20} The refcount for agbno 14 is 2. We move on to block 15, and add the last rmap. But we don't care where it is and we don't want to expand the array so we put it in slot 1: 0: {agbno: 10, length: 40} 1: {agbno: 15, length: 1} 2: {agbno: 12, length: 20} The refcount for block 15 is 3. Notice how order doesn't matter in this list? That's why repair uses an unordered list, or "bag". The data structure is not a set because it does not guarantee uniqueness. That said, adding and removing specific items is now an O(n) operation because we have no idea where that item might be in the list. Overall, the runtime is O(n^2) which is bad. I realized that I could easily refactor the btree code and reimplement the refcount bag with an xfbtree. Adding and removing is now O(log2 n), so the runtime is at least O(n log2 n), which is much faster. In the end, the rcbag becomes a sorted list, but that's merely a detail of the implementation. The repair code doesn't care. (Note: That horrible xfs_db bmap_inflate command can be used to exercise this sort of rcbag insanity by cranking up refcounts quickly.) Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 18a1e644 Thu Feb 22 01:43:40 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: define an in-memory btree for storing refcount bag info during repairs Create a new in-memory btree type so that we can store refcount bag info in a much more memory-efficient and performant format. Recall that the refcount recordset regenerator computes the new recordset from browsing the rmap records. Let's say that the rmap records are: {agbno: 10, length: 40, ...} {agbno: 11, length: 3, ...} {agbno: 12, length: 20, ...} {agbno: 15, length: 1, ...} It is convenient to have a data structure that could quickly tell us the refcount for an arbitrary agbno without wasting memory. An array or a list could do that pretty easily. List suck because of the pointer overhead. xfarrays are a lot more compact, but we want to minimize sparse holes in the xfarray to constrain memory usage. Maintaining any kind of record order isn't needed for correctness, so I created the "rcbag", which is shorthand for an unordered list of (excerpted) reverse mappings. So we add the first rmap to the rcbag, and it looks like: 0: {agbno: 10, length: 40} The refcount for agbno 10 is 1. Then we move on to block 11, so we add the second rmap: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} The refcount for agbno 11 is 2. We move on to block 12, so we add the third: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} 2: {agbno: 12, length: 20} The refcount for agbno 12 and 13 is 3. We move on to block 14, and remove the second rmap: 0: {agbno: 10, length: 40} 1: NULL 2: {agbno: 12, length: 20} The refcount for agbno 14 is 2. We move on to block 15, and add the last rmap. But we don't care where it is and we don't want to expand the array so we put it in slot 1: 0: {agbno: 10, length: 40} 1: {agbno: 15, length: 1} 2: {agbno: 12, length: 20} The refcount for block 15 is 3. Notice how order doesn't matter in this list? That's why repair uses an unordered list, or "bag". The data structure is not a set because it does not guarantee uniqueness. That said, adding and removing specific items is now an O(n) operation because we have no idea where that item might be in the list. Overall, the runtime is O(n^2) which is bad. I realized that I could easily refactor the btree code and reimplement the refcount bag with an xfbtree. Adding and removing is now O(log2 n), so the runtime is at least O(n log2 n), which is much faster. In the end, the rcbag becomes a sorted list, but that's merely a detail of the implementation. The repair code doesn't care. (Note: That horrible xfs_db bmap_inflate command can be used to exercise this sort of rcbag insanity by cranking up refcounts quickly.) Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 18a1e644 Thu Feb 22 01:43:40 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: define an in-memory btree for storing refcount bag info during repairs Create a new in-memory btree type so that we can store refcount bag info in a much more memory-efficient and performant format. Recall that the refcount recordset regenerator computes the new recordset from browsing the rmap records. Let's say that the rmap records are: {agbno: 10, length: 40, ...} {agbno: 11, length: 3, ...} {agbno: 12, length: 20, ...} {agbno: 15, length: 1, ...} It is convenient to have a data structure that could quickly tell us the refcount for an arbitrary agbno without wasting memory. An array or a list could do that pretty easily. List suck because of the pointer overhead. xfarrays are a lot more compact, but we want to minimize sparse holes in the xfarray to constrain memory usage. Maintaining any kind of record order isn't needed for correctness, so I created the "rcbag", which is shorthand for an unordered list of (excerpted) reverse mappings. So we add the first rmap to the rcbag, and it looks like: 0: {agbno: 10, length: 40} The refcount for agbno 10 is 1. Then we move on to block 11, so we add the second rmap: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} The refcount for agbno 11 is 2. We move on to block 12, so we add the third: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} 2: {agbno: 12, length: 20} The refcount for agbno 12 and 13 is 3. We move on to block 14, and remove the second rmap: 0: {agbno: 10, length: 40} 1: NULL 2: {agbno: 12, length: 20} The refcount for agbno 14 is 2. We move on to block 15, and add the last rmap. But we don't care where it is and we don't want to expand the array so we put it in slot 1: 0: {agbno: 10, length: 40} 1: {agbno: 15, length: 1} 2: {agbno: 12, length: 20} The refcount for block 15 is 3. Notice how order doesn't matter in this list? That's why repair uses an unordered list, or "bag". The data structure is not a set because it does not guarantee uniqueness. That said, adding and removing specific items is now an O(n) operation because we have no idea where that item might be in the list. Overall, the runtime is O(n^2) which is bad. I realized that I could easily refactor the btree code and reimplement the refcount bag with an xfbtree. Adding and removing is now O(log2 n), so the runtime is at least O(n log2 n), which is much faster. In the end, the rcbag becomes a sorted list, but that's merely a detail of the implementation. The repair code doesn't care. (Note: That horrible xfs_db bmap_inflate command can be used to exercise this sort of rcbag insanity by cranking up refcounts quickly.) Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 18a1e644 Thu Feb 22 01:43:40 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: define an in-memory btree for storing refcount bag info during repairs Create a new in-memory btree type so that we can store refcount bag info in a much more memory-efficient and performant format. Recall that the refcount recordset regenerator computes the new recordset from browsing the rmap records. Let's say that the rmap records are: {agbno: 10, length: 40, ...} {agbno: 11, length: 3, ...} {agbno: 12, length: 20, ...} {agbno: 15, length: 1, ...} It is convenient to have a data structure that could quickly tell us the refcount for an arbitrary agbno without wasting memory. An array or a list could do that pretty easily. List suck because of the pointer overhead. xfarrays are a lot more compact, but we want to minimize sparse holes in the xfarray to constrain memory usage. Maintaining any kind of record order isn't needed for correctness, so I created the "rcbag", which is shorthand for an unordered list of (excerpted) reverse mappings. So we add the first rmap to the rcbag, and it looks like: 0: {agbno: 10, length: 40} The refcount for agbno 10 is 1. Then we move on to block 11, so we add the second rmap: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} The refcount for agbno 11 is 2. We move on to block 12, so we add the third: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} 2: {agbno: 12, length: 20} The refcount for agbno 12 and 13 is 3. We move on to block 14, and remove the second rmap: 0: {agbno: 10, length: 40} 1: NULL 2: {agbno: 12, length: 20} The refcount for agbno 14 is 2. We move on to block 15, and add the last rmap. But we don't care where it is and we don't want to expand the array so we put it in slot 1: 0: {agbno: 10, length: 40} 1: {agbno: 15, length: 1} 2: {agbno: 12, length: 20} The refcount for block 15 is 3. Notice how order doesn't matter in this list? That's why repair uses an unordered list, or "bag". The data structure is not a set because it does not guarantee uniqueness. That said, adding and removing specific items is now an O(n) operation because we have no idea where that item might be in the list. Overall, the runtime is O(n^2) which is bad. I realized that I could easily refactor the btree code and reimplement the refcount bag with an xfbtree. Adding and removing is now O(log2 n), so the runtime is at least O(n log2 n), which is much faster. In the end, the rcbag becomes a sorted list, but that's merely a detail of the implementation. The repair code doesn't care. (Note: That horrible xfs_db bmap_inflate command can be used to exercise this sort of rcbag insanity by cranking up refcounts quickly.) Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 18a1e644 Thu Feb 22 01:43:40 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: define an in-memory btree for storing refcount bag info during repairs Create a new in-memory btree type so that we can store refcount bag info in a much more memory-efficient and performant format. Recall that the refcount recordset regenerator computes the new recordset from browsing the rmap records. Let's say that the rmap records are: {agbno: 10, length: 40, ...} {agbno: 11, length: 3, ...} {agbno: 12, length: 20, ...} {agbno: 15, length: 1, ...} It is convenient to have a data structure that could quickly tell us the refcount for an arbitrary agbno without wasting memory. An array or a list could do that pretty easily. List suck because of the pointer overhead. xfarrays are a lot more compact, but we want to minimize sparse holes in the xfarray to constrain memory usage. Maintaining any kind of record order isn't needed for correctness, so I created the "rcbag", which is shorthand for an unordered list of (excerpted) reverse mappings. So we add the first rmap to the rcbag, and it looks like: 0: {agbno: 10, length: 40} The refcount for agbno 10 is 1. Then we move on to block 11, so we add the second rmap: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} The refcount for agbno 11 is 2. We move on to block 12, so we add the third: 0: {agbno: 10, length: 40} 1: {agbno: 11, length: 3} 2: {agbno: 12, length: 20} The refcount for agbno 12 and 13 is 3. We move on to block 14, and remove the second rmap: 0: {agbno: 10, length: 40} 1: NULL 2: {agbno: 12, length: 20} The refcount for agbno 14 is 2. We move on to block 15, and add the last rmap. But we don't care where it is and we don't want to expand the array so we put it in slot 1: 0: {agbno: 10, length: 40} 1: {agbno: 15, length: 1} 2: {agbno: 12, length: 20} The refcount for block 15 is 3. Notice how order doesn't matter in this list? That's why repair uses an unordered list, or "bag". The data structure is not a set because it does not guarantee uniqueness. That said, adding and removing specific items is now an O(n) operation because we have no idea where that item might be in the list. Overall, the runtime is O(n^2) which is bad. I realized that I could easily refactor the btree code and reimplement the refcount bag with an xfbtree. Adding and removing is now O(log2 n), so the runtime is at least O(n log2 n), which is much faster. In the end, the rcbag becomes a sorted list, but that's merely a detail of the implementation. The repair code doesn't care. (Note: That horrible xfs_db bmap_inflate command can be used to exercise this sort of rcbag insanity by cranking up refcounts quickly.) Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> |
H A D | xfs_rmap_item.c | diff 3c919b09 Mon Sep 11 09:39:05 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: reserve less log space when recovering log intent items Wengang Wang reports that a customer's system was running a number of truncate operations on a filesystem with a very small log. Contention on the reserve heads lead to other threads stalling on smaller updates (e.g. mtime updates) long enough to result in the node being rebooted on account of the lack of responsivenes. The node failed to recover because log recovery of an EFI became stuck waiting for a grant of reserve space. From Wengang's report: "For the file deletion, log bytes are reserved basing on xfs_mount->tr_itruncate which is: tr_logres = 175488, tr_logcount = 2, tr_logflags = XFS_TRANS_PERM_LOG_RES, "You see it's a permanent log reservation with two log operations (two transactions in rolling mode). After calculation (xlog_calc_unit_res() adds space for various log headers), the final log space needed per transaction changes from 175488 to 180208 bytes. So the total log space needed is 360416 bytes (180208 * 2). [That quantity] of log space (360416 bytes) needs to be reserved for both run time inode removing (xfs_inactive_truncate()) and EFI recover (xfs_efi_item_recover())." In other words, runtime pre-reserves 360K of space in anticipation of running a chain of two transactions in which each transaction gets a 180K reservation. Now that we've allocated the transaction, we delete the bmap mapping, log an EFI to free the space, and roll the transaction as part of finishing the deferops chain. Rolling creates a new xfs_trans which shares its ticket with the old transaction. Next, xfs_trans_roll calls __xfs_trans_commit with regrant == true, which calls xlog_cil_commit with the same regrant parameter. xlog_cil_commit calls xfs_log_ticket_regrant, which decrements t_cnt and subtracts t_curr_res from the reservation and write heads. If the filesystem is fresh and the first transaction only used (say) 20K, then t_curr_res will be 160K, and we give that much reservation back to the reservation head. Or if the file is really fragmented and the first transaction actually uses 170K, then t_curr_res will be 10K, and that's what we give back to the reservation. Having done that, we're now headed into the second transaction with an EFI and 180K of reservation. Other threads apparently consumed all the reservation for smaller transactions, such as timestamp updates. Now let's say the first transaction gets written to disk and we crash without ever completing the second transaction. Now we remount the fs, log recovery finds the unfinished EFI, and calls xfs_efi_recover to finish the EFI. However, xfs_efi_recover starts a new tr_itruncate tranasction, which asks for 360K log reservation. This is a lot more than the 180K that we had reserved at the time of the crash. If the first EFI to be recovered is also pinning the tail of the log, we will be unable to free any space in the log, and recovery livelocks. Wengang confirmed this: "Now we have the second transaction which has 180208 log bytes reserved too. The second transaction is supposed to process intents including extent freeing. With my hacking patch, I blocked the extent freeing 5 hours. So in that 5 hours, 180208 (NOT 360416) log bytes are reserved. "With my test case, other transactions (update timestamps) then happen. As my hacking patch pins the journal tail, those timestamp-updating transactions finally use up (almost) all the left available log space (in memory in on disk). And finally the on disk (and in memory) available log space goes down near to 180208 bytes. Those 180208 bytes are reserved by [the] second (extent-free) transaction [in the chain]." Wengang and I noticed that EFI recovery starts a transaction, completes one step of the chain, and commits the transaction without completing any other steps of the chain. Those subsequent steps are completed by xlog_finish_defer_ops, which allocates yet another transaction to finish the rest of the chain. That transaction gets the same tr_logres as the head transaction, but with tr_logcount = 1 to force regranting with every roll to avoid livelocks. In other words, we already figured this out in commit 929b92f64048d ("xfs: xfs_defer_capture should absorb remaining transaction reservation"), but should have applied that logic to each intent item's recovery function. For Wengang's case, the xfs_trans_alloc call in the EFI recovery function should only be asking for a single transaction's worth of log reservation -- 180K, not 360K. Quoting Wengang again: "With log recovery, during EFI recovery, we use tr_itruncate again to reserve two transactions that needs 360416 log bytes. Reserving 360416 bytes fails [stalls] because we now only have about 180208 available. "Actually during the EFI recover, we only need one transaction to free the extents just like the 2nd transaction at RUNTIME. So it only needs to reserve 180208 rather than 360416 bytes. We have (a bit) more than 180208 available log bytes on disk, so [if we decrease the reservation to 180K] the reservation goes and the recovery [finishes]. That is to say: we can fix the log recover part to fix the issue. We can introduce a new xfs_trans_res xfs_mount->tr_ext_free { tr_logres = 175488, tr_logcount = 0, tr_logflags = 0, } "and use tr_ext_free instead of tr_itruncate in EFI recover." However, I don't think it quite makes sense to create an entirely new transaction reservation type to handle single-stepping during log recovery. Instead, we should copy the transaction reservation information in the xfs_mount, change tr_logcount to 1, and pass that into xfs_trans_alloc. We know this won't risk changing the min log size computation since we always ask for a fraction of the reservation for all known transaction types. This looks like it's been lurking in the codebase since commit 3d3c8b5222b92, which changed the xfs_trans_reserve call in xlog_recover_process_efi to use the tr_logcount in tr_itruncate. That changed the EFI recovery transaction from making a non-XFS_TRANS_PERM_LOG_RES request for one transaction's worth of log space to a XFS_TRANS_PERM_LOG_RES request for two transactions worth. Fixes: 3d3c8b5222b92 ("xfs: refactor xfs_trans_reserve() interface") Complements: 929b92f64048d ("xfs: xfs_defer_capture should absorb remaining transaction reservation") Suggested-by: Wengang Wang <wen.gang.wang@oracle.com> Cc: Srikanth C S <srikanth.c.s@oracle.com> [djwong: apply the same transformation to all log intent recovery] Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> diff 3c919b09 Mon Sep 11 09:39:05 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: reserve less log space when recovering log intent items Wengang Wang reports that a customer's system was running a number of truncate operations on a filesystem with a very small log. Contention on the reserve heads lead to other threads stalling on smaller updates (e.g. mtime updates) long enough to result in the node being rebooted on account of the lack of responsivenes. The node failed to recover because log recovery of an EFI became stuck waiting for a grant of reserve space. From Wengang's report: "For the file deletion, log bytes are reserved basing on xfs_mount->tr_itruncate which is: tr_logres = 175488, tr_logcount = 2, tr_logflags = XFS_TRANS_PERM_LOG_RES, "You see it's a permanent log reservation with two log operations (two transactions in rolling mode). After calculation (xlog_calc_unit_res() adds space for various log headers), the final log space needed per transaction changes from 175488 to 180208 bytes. So the total log space needed is 360416 bytes (180208 * 2). [That quantity] of log space (360416 bytes) needs to be reserved for both run time inode removing (xfs_inactive_truncate()) and EFI recover (xfs_efi_item_recover())." In other words, runtime pre-reserves 360K of space in anticipation of running a chain of two transactions in which each transaction gets a 180K reservation. Now that we've allocated the transaction, we delete the bmap mapping, log an EFI to free the space, and roll the transaction as part of finishing the deferops chain. Rolling creates a new xfs_trans which shares its ticket with the old transaction. Next, xfs_trans_roll calls __xfs_trans_commit with regrant == true, which calls xlog_cil_commit with the same regrant parameter. xlog_cil_commit calls xfs_log_ticket_regrant, which decrements t_cnt and subtracts t_curr_res from the reservation and write heads. If the filesystem is fresh and the first transaction only used (say) 20K, then t_curr_res will be 160K, and we give that much reservation back to the reservation head. Or if the file is really fragmented and the first transaction actually uses 170K, then t_curr_res will be 10K, and that's what we give back to the reservation. Having done that, we're now headed into the second transaction with an EFI and 180K of reservation. Other threads apparently consumed all the reservation for smaller transactions, such as timestamp updates. Now let's say the first transaction gets written to disk and we crash without ever completing the second transaction. Now we remount the fs, log recovery finds the unfinished EFI, and calls xfs_efi_recover to finish the EFI. However, xfs_efi_recover starts a new tr_itruncate tranasction, which asks for 360K log reservation. This is a lot more than the 180K that we had reserved at the time of the crash. If the first EFI to be recovered is also pinning the tail of the log, we will be unable to free any space in the log, and recovery livelocks. Wengang confirmed this: "Now we have the second transaction which has 180208 log bytes reserved too. The second transaction is supposed to process intents including extent freeing. With my hacking patch, I blocked the extent freeing 5 hours. So in that 5 hours, 180208 (NOT 360416) log bytes are reserved. "With my test case, other transactions (update timestamps) then happen. As my hacking patch pins the journal tail, those timestamp-updating transactions finally use up (almost) all the left available log space (in memory in on disk). And finally the on disk (and in memory) available log space goes down near to 180208 bytes. Those 180208 bytes are reserved by [the] second (extent-free) transaction [in the chain]." Wengang and I noticed that EFI recovery starts a transaction, completes one step of the chain, and commits the transaction without completing any other steps of the chain. Those subsequent steps are completed by xlog_finish_defer_ops, which allocates yet another transaction to finish the rest of the chain. That transaction gets the same tr_logres as the head transaction, but with tr_logcount = 1 to force regranting with every roll to avoid livelocks. In other words, we already figured this out in commit 929b92f64048d ("xfs: xfs_defer_capture should absorb remaining transaction reservation"), but should have applied that logic to each intent item's recovery function. For Wengang's case, the xfs_trans_alloc call in the EFI recovery function should only be asking for a single transaction's worth of log reservation -- 180K, not 360K. Quoting Wengang again: "With log recovery, during EFI recovery, we use tr_itruncate again to reserve two transactions that needs 360416 log bytes. Reserving 360416 bytes fails [stalls] because we now only have about 180208 available. "Actually during the EFI recover, we only need one transaction to free the extents just like the 2nd transaction at RUNTIME. So it only needs to reserve 180208 rather than 360416 bytes. We have (a bit) more than 180208 available log bytes on disk, so [if we decrease the reservation to 180K] the reservation goes and the recovery [finishes]. That is to say: we can fix the log recover part to fix the issue. We can introduce a new xfs_trans_res xfs_mount->tr_ext_free { tr_logres = 175488, tr_logcount = 0, tr_logflags = 0, } "and use tr_ext_free instead of tr_itruncate in EFI recover." However, I don't think it quite makes sense to create an entirely new transaction reservation type to handle single-stepping during log recovery. Instead, we should copy the transaction reservation information in the xfs_mount, change tr_logcount to 1, and pass that into xfs_trans_alloc. We know this won't risk changing the min log size computation since we always ask for a fraction of the reservation for all known transaction types. This looks like it's been lurking in the codebase since commit 3d3c8b5222b92, which changed the xfs_trans_reserve call in xlog_recover_process_efi to use the tr_logcount in tr_itruncate. That changed the EFI recovery transaction from making a non-XFS_TRANS_PERM_LOG_RES request for one transaction's worth of log space to a XFS_TRANS_PERM_LOG_RES request for two transactions worth. Fixes: 3d3c8b5222b92 ("xfs: refactor xfs_trans_reserve() interface") Complements: 929b92f64048d ("xfs: xfs_defer_capture should absorb remaining transaction reservation") Suggested-by: Wengang Wang <wen.gang.wang@oracle.com> Cc: Srikanth C S <srikanth.c.s@oracle.com> [djwong: apply the same transformation to all log intent recovery] Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> diff 707e0dda Mon Aug 26 01:06:22 MDT 2019 Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> fs: xfs: Remove KM_NOSLEEP and KM_SLEEP. Since no caller is using KM_NOSLEEP and no callee branches on KM_SLEEP, we can remove KM_NOSLEEP and replace KM_SLEEP with 0. Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> |
H A D | xfs_reflink.c | diff d6211330 Thu Aug 04 09:59:27 MDT 2022 Chandan Babu R <chandan.babu@oracle.com> xfs: Fix false ENOSPC when performing direct write on a delalloc extent in cow fork On a higly fragmented filesystem a Direct IO write can fail with -ENOSPC error even though the filesystem has sufficient number of free blocks. This occurs if the file offset range on which the write operation is being performed has a delalloc extent in the cow fork and this delalloc extent begins much before the Direct IO range. In such a scenario, xfs_reflink_allocate_cow() invokes xfs_bmapi_write() to allocate the blocks mapped by the delalloc extent. The extent thus allocated may not cover the beginning of file offset range on which the Direct IO write was issued. Hence xfs_reflink_allocate_cow() ends up returning -ENOSPC. The following script reliably recreates the bug described above. #!/usr/bin/bash device=/dev/loop0 shortdev=$(basename $device) mntpnt=/mnt/ file1=${mntpnt}/file1 file2=${mntpnt}/file2 fragmentedfile=${mntpnt}/fragmentedfile punchprog=/root/repos/xfstests-dev/src/punch-alternating errortag=/sys/fs/xfs/${shortdev}/errortag/bmap_alloc_minlen_extent umount $device > /dev/null 2>&1 echo "Create FS" mkfs.xfs -f -m reflink=1 $device > /dev/null 2>&1 if [[ $? != 0 ]]; then echo "mkfs failed." exit 1 fi echo "Mount FS" mount $device $mntpnt > /dev/null 2>&1 if [[ $? != 0 ]]; then echo "mount failed." exit 1 fi echo "Create source file" xfs_io -f -c "pwrite 0 32M" $file1 > /dev/null 2>&1 sync echo "Create Reflinked file" xfs_io -f -c "reflink $file1" $file2 &>/dev/null echo "Set cowextsize" xfs_io -c "cowextsize 16M" $file1 > /dev/null 2>&1 echo "Fragment FS" xfs_io -f -c "pwrite 0 64M" $fragmentedfile > /dev/null 2>&1 sync $punchprog $fragmentedfile echo "Allocate block sized extent from now onwards" echo -n 1 > $errortag echo "Create 16MiB delalloc extent in CoW fork" xfs_io -c "pwrite 0 4k" $file1 > /dev/null 2>&1 sync echo "Direct I/O write at offset 12k" xfs_io -d -c "pwrite 12k 8k" $file1 This commit fixes the bug by invoking xfs_bmapi_write() in a loop until disk blocks are allocated for atleast the starting file offset of the Direct IO write range. Fixes: 3c68d44a2b49 ("xfs: allocate direct I/O COW blocks in iomap_begin") Reported-and-Root-caused-by: Wengang Wang <wen.gang.wang@oracle.com> Signed-off-by: Chandan Babu R <chandan.babu@oracle.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> [djwong: slight editing to make the locking less grody, and fix some style things] Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff d6211330 Thu Aug 04 09:59:27 MDT 2022 Chandan Babu R <chandan.babu@oracle.com> xfs: Fix false ENOSPC when performing direct write on a delalloc extent in cow fork On a higly fragmented filesystem a Direct IO write can fail with -ENOSPC error even though the filesystem has sufficient number of free blocks. This occurs if the file offset range on which the write operation is being performed has a delalloc extent in the cow fork and this delalloc extent begins much before the Direct IO range. In such a scenario, xfs_reflink_allocate_cow() invokes xfs_bmapi_write() to allocate the blocks mapped by the delalloc extent. The extent thus allocated may not cover the beginning of file offset range on which the Direct IO write was issued. Hence xfs_reflink_allocate_cow() ends up returning -ENOSPC. The following script reliably recreates the bug described above. #!/usr/bin/bash device=/dev/loop0 shortdev=$(basename $device) mntpnt=/mnt/ file1=${mntpnt}/file1 file2=${mntpnt}/file2 fragmentedfile=${mntpnt}/fragmentedfile punchprog=/root/repos/xfstests-dev/src/punch-alternating errortag=/sys/fs/xfs/${shortdev}/errortag/bmap_alloc_minlen_extent umount $device > /dev/null 2>&1 echo "Create FS" mkfs.xfs -f -m reflink=1 $device > /dev/null 2>&1 if [[ $? != 0 ]]; then echo "mkfs failed." exit 1 fi echo "Mount FS" mount $device $mntpnt > /dev/null 2>&1 if [[ $? != 0 ]]; then echo "mount failed." exit 1 fi echo "Create source file" xfs_io -f -c "pwrite 0 32M" $file1 > /dev/null 2>&1 sync echo "Create Reflinked file" xfs_io -f -c "reflink $file1" $file2 &>/dev/null echo "Set cowextsize" xfs_io -c "cowextsize 16M" $file1 > /dev/null 2>&1 echo "Fragment FS" xfs_io -f -c "pwrite 0 64M" $fragmentedfile > /dev/null 2>&1 sync $punchprog $fragmentedfile echo "Allocate block sized extent from now onwards" echo -n 1 > $errortag echo "Create 16MiB delalloc extent in CoW fork" xfs_io -c "pwrite 0 4k" $file1 > /dev/null 2>&1 sync echo "Direct I/O write at offset 12k" xfs_io -d -c "pwrite 12k 8k" $file1 This commit fixes the bug by invoking xfs_bmapi_write() in a loop until disk blocks are allocated for atleast the starting file offset of the Direct IO write range. Fixes: 3c68d44a2b49 ("xfs: allocate direct I/O COW blocks in iomap_begin") Reported-and-Root-caused-by: Wengang Wang <wen.gang.wang@oracle.com> Signed-off-by: Chandan Babu R <chandan.babu@oracle.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> [djwong: slight editing to make the locking less grody, and fix some style things] Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff d6211330 Thu Aug 04 09:59:27 MDT 2022 Chandan Babu R <chandan.babu@oracle.com> xfs: Fix false ENOSPC when performing direct write on a delalloc extent in cow fork On a higly fragmented filesystem a Direct IO write can fail with -ENOSPC error even though the filesystem has sufficient number of free blocks. This occurs if the file offset range on which the write operation is being performed has a delalloc extent in the cow fork and this delalloc extent begins much before the Direct IO range. In such a scenario, xfs_reflink_allocate_cow() invokes xfs_bmapi_write() to allocate the blocks mapped by the delalloc extent. The extent thus allocated may not cover the beginning of file offset range on which the Direct IO write was issued. Hence xfs_reflink_allocate_cow() ends up returning -ENOSPC. The following script reliably recreates the bug described above. #!/usr/bin/bash device=/dev/loop0 shortdev=$(basename $device) mntpnt=/mnt/ file1=${mntpnt}/file1 file2=${mntpnt}/file2 fragmentedfile=${mntpnt}/fragmentedfile punchprog=/root/repos/xfstests-dev/src/punch-alternating errortag=/sys/fs/xfs/${shortdev}/errortag/bmap_alloc_minlen_extent umount $device > /dev/null 2>&1 echo "Create FS" mkfs.xfs -f -m reflink=1 $device > /dev/null 2>&1 if [[ $? != 0 ]]; then echo "mkfs failed." exit 1 fi echo "Mount FS" mount $device $mntpnt > /dev/null 2>&1 if [[ $? != 0 ]]; then echo "mount failed." exit 1 fi echo "Create source file" xfs_io -f -c "pwrite 0 32M" $file1 > /dev/null 2>&1 sync echo "Create Reflinked file" xfs_io -f -c "reflink $file1" $file2 &>/dev/null echo "Set cowextsize" xfs_io -c "cowextsize 16M" $file1 > /dev/null 2>&1 echo "Fragment FS" xfs_io -f -c "pwrite 0 64M" $fragmentedfile > /dev/null 2>&1 sync $punchprog $fragmentedfile echo "Allocate block sized extent from now onwards" echo -n 1 > $errortag echo "Create 16MiB delalloc extent in CoW fork" xfs_io -c "pwrite 0 4k" $file1 > /dev/null 2>&1 sync echo "Direct I/O write at offset 12k" xfs_io -d -c "pwrite 12k 8k" $file1 This commit fixes the bug by invoking xfs_bmapi_write() in a loop until disk blocks are allocated for atleast the starting file offset of the Direct IO write range. Fixes: 3c68d44a2b49 ("xfs: allocate direct I/O COW blocks in iomap_begin") Reported-and-Root-caused-by: Wengang Wang <wen.gang.wang@oracle.com> Signed-off-by: Chandan Babu R <chandan.babu@oracle.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> [djwong: slight editing to make the locking less grody, and fix some style things] Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff d6211330 Thu Aug 04 09:59:27 MDT 2022 Chandan Babu R <chandan.babu@oracle.com> xfs: Fix false ENOSPC when performing direct write on a delalloc extent in cow fork On a higly fragmented filesystem a Direct IO write can fail with -ENOSPC error even though the filesystem has sufficient number of free blocks. This occurs if the file offset range on which the write operation is being performed has a delalloc extent in the cow fork and this delalloc extent begins much before the Direct IO range. In such a scenario, xfs_reflink_allocate_cow() invokes xfs_bmapi_write() to allocate the blocks mapped by the delalloc extent. The extent thus allocated may not cover the beginning of file offset range on which the Direct IO write was issued. Hence xfs_reflink_allocate_cow() ends up returning -ENOSPC. The following script reliably recreates the bug described above. #!/usr/bin/bash device=/dev/loop0 shortdev=$(basename $device) mntpnt=/mnt/ file1=${mntpnt}/file1 file2=${mntpnt}/file2 fragmentedfile=${mntpnt}/fragmentedfile punchprog=/root/repos/xfstests-dev/src/punch-alternating errortag=/sys/fs/xfs/${shortdev}/errortag/bmap_alloc_minlen_extent umount $device > /dev/null 2>&1 echo "Create FS" mkfs.xfs -f -m reflink=1 $device > /dev/null 2>&1 if [[ $? != 0 ]]; then echo "mkfs failed." exit 1 fi echo "Mount FS" mount $device $mntpnt > /dev/null 2>&1 if [[ $? != 0 ]]; then echo "mount failed." exit 1 fi echo "Create source file" xfs_io -f -c "pwrite 0 32M" $file1 > /dev/null 2>&1 sync echo "Create Reflinked file" xfs_io -f -c "reflink $file1" $file2 &>/dev/null echo "Set cowextsize" xfs_io -c "cowextsize 16M" $file1 > /dev/null 2>&1 echo "Fragment FS" xfs_io -f -c "pwrite 0 64M" $fragmentedfile > /dev/null 2>&1 sync $punchprog $fragmentedfile echo "Allocate block sized extent from now onwards" echo -n 1 > $errortag echo "Create 16MiB delalloc extent in CoW fork" xfs_io -c "pwrite 0 4k" $file1 > /dev/null 2>&1 sync echo "Direct I/O write at offset 12k" xfs_io -d -c "pwrite 12k 8k" $file1 This commit fixes the bug by invoking xfs_bmapi_write() in a loop until disk blocks are allocated for atleast the starting file offset of the Direct IO write range. Fixes: 3c68d44a2b49 ("xfs: allocate direct I/O COW blocks in iomap_begin") Reported-and-Root-caused-by: Wengang Wang <wen.gang.wang@oracle.com> Signed-off-by: Chandan Babu R <chandan.babu@oracle.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> [djwong: slight editing to make the locking less grody, and fix some style things] Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff d6211330 Thu Aug 04 09:59:27 MDT 2022 Chandan Babu R <chandan.babu@oracle.com> xfs: Fix false ENOSPC when performing direct write on a delalloc extent in cow fork On a higly fragmented filesystem a Direct IO write can fail with -ENOSPC error even though the filesystem has sufficient number of free blocks. This occurs if the file offset range on which the write operation is being performed has a delalloc extent in the cow fork and this delalloc extent begins much before the Direct IO range. In such a scenario, xfs_reflink_allocate_cow() invokes xfs_bmapi_write() to allocate the blocks mapped by the delalloc extent. The extent thus allocated may not cover the beginning of file offset range on which the Direct IO write was issued. Hence xfs_reflink_allocate_cow() ends up returning -ENOSPC. The following script reliably recreates the bug described above. #!/usr/bin/bash device=/dev/loop0 shortdev=$(basename $device) mntpnt=/mnt/ file1=${mntpnt}/file1 file2=${mntpnt}/file2 fragmentedfile=${mntpnt}/fragmentedfile punchprog=/root/repos/xfstests-dev/src/punch-alternating errortag=/sys/fs/xfs/${shortdev}/errortag/bmap_alloc_minlen_extent umount $device > /dev/null 2>&1 echo "Create FS" mkfs.xfs -f -m reflink=1 $device > /dev/null 2>&1 if [[ $? != 0 ]]; then echo "mkfs failed." exit 1 fi echo "Mount FS" mount $device $mntpnt > /dev/null 2>&1 if [[ $? != 0 ]]; then echo "mount failed." exit 1 fi echo "Create source file" xfs_io -f -c "pwrite 0 32M" $file1 > /dev/null 2>&1 sync echo "Create Reflinked file" xfs_io -f -c "reflink $file1" $file2 &>/dev/null echo "Set cowextsize" xfs_io -c "cowextsize 16M" $file1 > /dev/null 2>&1 echo "Fragment FS" xfs_io -f -c "pwrite 0 64M" $fragmentedfile > /dev/null 2>&1 sync $punchprog $fragmentedfile echo "Allocate block sized extent from now onwards" echo -n 1 > $errortag echo "Create 16MiB delalloc extent in CoW fork" xfs_io -c "pwrite 0 4k" $file1 > /dev/null 2>&1 sync echo "Direct I/O write at offset 12k" xfs_io -d -c "pwrite 12k 8k" $file1 This commit fixes the bug by invoking xfs_bmapi_write() in a loop until disk blocks are allocated for atleast the starting file offset of the Direct IO write range. Fixes: 3c68d44a2b49 ("xfs: allocate direct I/O COW blocks in iomap_begin") Reported-and-Root-caused-by: Wengang Wang <wen.gang.wang@oracle.com> Signed-off-by: Chandan Babu R <chandan.babu@oracle.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> [djwong: slight editing to make the locking less grody, and fix some style things] Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 7993f1a4 Wed Dec 15 12:52:23 MST 2021 Darrick J. Wong <djwong@kernel.org> xfs: only run COW extent recovery when there are no live extents As part of multiple customer escalations due to file data corruption after copy on write operations, I wrote some fstests that use fsstress to hammer on COW to shake things loose. Regrettably, I caught some filesystem shutdowns due to incorrect rmap operations with the following loop: mount <filesystem> # (0) fsstress <run only readonly ops> & # (1) while true; do fsstress <run all ops> mount -o remount,ro # (2) fsstress <run only readonly ops> mount -o remount,rw # (3) done When (2) happens, notice that (1) is still running. xfs_remount_ro will call xfs_blockgc_stop to walk the inode cache to free all the COW extents, but the blockgc mechanism races with (1)'s reader threads to take IOLOCKs and loses, which means that it doesn't clean them all out. Call such a file (A). When (3) happens, xfs_remount_rw calls xfs_reflink_recover_cow, which walks the ondisk refcount btree and frees any COW extent that it finds. This function does not check the inode cache, which means that incore COW forks of inode (A) is now inconsistent with the ondisk metadata. If one of those former COW extents are allocated and mapped into another file (B) and someone triggers a COW to the stale reservation in (A), A's dirty data will be written into (B) and once that's done, those blocks will be transferred to (A)'s data fork without bumping the refcount. The results are catastrophic -- file (B) and the refcount btree are now corrupt. In the first patch, we fixed the race condition in (2) so that (A) will always flush the COW fork. In this second patch, we move the _recover_cow call to the initial mount call in (0) for safety. As mentioned previously, xfs_reflink_recover_cow walks the refcount btree looking for COW staging extents, and frees them. This was intended to be run at mount time (when we know there are no live inodes) to clean up any leftover staging events that may have been left behind during an unclean shutdown. As a time "optimization" for readonly mounts, we deferred this to the ro->rw transition, not realizing that any failure to clean all COW forks during a rw->ro transition would result in catastrophic corruption. Therefore, remove this optimization and only run the recovery routine when we're guaranteed not to have any COW staging extents anywhere, which means we always run this at mount time. While we're at it, move the callsite to xfs_log_mount_finish because any refcount btree expansion (however unlikely given that we're removing records from the right side of the index) must be fed by a per-AG reservation, which doesn't exist in its current location. Fixes: 174edb0e46e5 ("xfs: store in-progress CoW allocations in the refcount btree") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Chandan Babu R <chandan.babu@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> diff 7993f1a4 Wed Dec 15 12:52:23 MST 2021 Darrick J. Wong <djwong@kernel.org> xfs: only run COW extent recovery when there are no live extents As part of multiple customer escalations due to file data corruption after copy on write operations, I wrote some fstests that use fsstress to hammer on COW to shake things loose. Regrettably, I caught some filesystem shutdowns due to incorrect rmap operations with the following loop: mount <filesystem> # (0) fsstress <run only readonly ops> & # (1) while true; do fsstress <run all ops> mount -o remount,ro # (2) fsstress <run only readonly ops> mount -o remount,rw # (3) done When (2) happens, notice that (1) is still running. xfs_remount_ro will call xfs_blockgc_stop to walk the inode cache to free all the COW extents, but the blockgc mechanism races with (1)'s reader threads to take IOLOCKs and loses, which means that it doesn't clean them all out. Call such a file (A). When (3) happens, xfs_remount_rw calls xfs_reflink_recover_cow, which walks the ondisk refcount btree and frees any COW extent that it finds. This function does not check the inode cache, which means that incore COW forks of inode (A) is now inconsistent with the ondisk metadata. If one of those former COW extents are allocated and mapped into another file (B) and someone triggers a COW to the stale reservation in (A), A's dirty data will be written into (B) and once that's done, those blocks will be transferred to (A)'s data fork without bumping the refcount. The results are catastrophic -- file (B) and the refcount btree are now corrupt. In the first patch, we fixed the race condition in (2) so that (A) will always flush the COW fork. In this second patch, we move the _recover_cow call to the initial mount call in (0) for safety. As mentioned previously, xfs_reflink_recover_cow walks the refcount btree looking for COW staging extents, and frees them. This was intended to be run at mount time (when we know there are no live inodes) to clean up any leftover staging events that may have been left behind during an unclean shutdown. As a time "optimization" for readonly mounts, we deferred this to the ro->rw transition, not realizing that any failure to clean all COW forks during a rw->ro transition would result in catastrophic corruption. Therefore, remove this optimization and only run the recovery routine when we're guaranteed not to have any COW staging extents anywhere, which means we always run this at mount time. While we're at it, move the callsite to xfs_log_mount_finish because any refcount btree expansion (however unlikely given that we're removing records from the right side of the index) must be fed by a per-AG reservation, which doesn't exist in its current location. Fixes: 174edb0e46e5 ("xfs: store in-progress CoW allocations in the refcount btree") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Chandan Babu R <chandan.babu@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> diff aa124436 Mon Jan 20 15:34:47 MST 2020 zhengbin <zhengbin13@huawei.com> xfs: change return value of xfs_inode_need_cow to int Fixes coccicheck warning: fs/xfs/xfs_reflink.c:236:9-10: WARNING: return of 0/1 in function 'xfs_inode_need_cow' with return type bool Reported-by: Hulk Robot <hulkci@huawei.com> Signed-off-by: zhengbin <zhengbin13@huawei.com> [darrick: rename the function so it doesn't sound like a predicate] Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 66ae56a5 Mon Feb 18 10:38:49 MST 2019 Christoph Hellwig <hch@lst.de> xfs: introduce an always_cow mode Add a mode where XFS never overwrites existing blocks in place. This is to aid debugging our COW code, and also put infatructure in place for things like possible future support for zoned block devices, which can't support overwrites. This mode is enabled globally by doing a: echo 1 > /sys/fs/xfs/debug/always_cow Note that the parameter is global to allow running all tests in xfstests easily in this mode, which would not easily be possible with a per-fs sysfs file. In always_cow mode persistent preallocations are disabled, and fallocate will fail when called with a 0 mode (with our without FALLOC_FL_KEEP_SIZE), and not create unwritten extent for zeroed space when called with FALLOC_FL_ZERO_RANGE or FALLOC_FL_UNSHARE_RANGE. There are a few interesting xfstests failures when run in always_cow mode: - generic/392 fails because the bytes used in the file used to test hole punch recovery are less after the log replay. This is because the blocks written and then punched out are only freed with a delay due to the logging mechanism. - xfs/170 will fail as the already fragile file streams mechanism doesn't seem to interact well with the COW allocator - xfs/180 xfs/182 xfs/192 xfs/198 xfs/204 and xfs/208 will claim the file system is badly fragmented, but there is not much we can do to avoid that when always writing out of place - xfs/205 fails because overwriting a file in always_cow mode will require new space allocation and the assumption in the test thus don't work anymore. - xfs/326 fails to modify the file at all in always_cow mode after injecting the refcount error, leading to an unexpected md5sum after the remount, but that again is expected Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> diff d6f215f3 Wed Dec 12 09:46:19 MST 2018 Darrick J. Wong <darrick.wong@oracle.com> xfs: split up the xfs_reflink_end_cow work into smaller transactions In xfs_reflink_end_cow, we allocate a single transaction for the entire end_cow operation and then loop the CoW fork mappings to move them to the data fork. This design fails on a heavily fragmented filesystem where an inode's data fork has exactly one more extent than would fit in an extents-format fork, because the unmap can collapse the data fork into extents format (freeing the bmbt block) but the remap can expand the data fork back into a (newly allocated) bmbt block. If the number of extents we end up remapping is large, we can overflow the block reservation because we reserved blocks assuming that we were adding mappings into an already-cleared area of the data fork. Let's say we have 8 extents in the data fork, 8 extents in the CoW fork, and the data fork can hold at most 7 extents before needing to convert to btree format; and that blocks A-P are discontiguous single-block extents: 0......7 D: ABCDEFGH C: IJKLMNOP When a write to file blocks 0-7 completes, we must remap I-P into the data fork. We start by removing H from the btree-format data fork. Now we have 7 extents, so we convert the fork to extents format, freeing the bmbt block. We then move P into the data fork and it now has 8 extents again. We must convert the data fork back to btree format, requiring a block allocation. If we repeat this sequence for blocks 6-5-4-3-2-1-0, we'll need a total of 8 block allocations to remap all 8 blocks. We reserved only enough blocks to handle one btree split (5 blocks on a 4k block filesystem), which means we overflow the block reservation. To fix this issue, create a separate helper function to remap a single extent, and change _reflink_end_cow to call it in a tight loop over the entire range we're completing. As a side effect this also removes the size restrictions on how many extents we can end_cow at a time, though nobody ever hit that. It is not reasonable to reserve N blocks to remap N blocks. Note that this can be reproduced after ~320 million fsx ops while running generic/938 (long soak directio fsx exerciser): XFS: Assertion failed: tp->t_blk_res >= tp->t_blk_res_used, file: fs/xfs/xfs_trans.c, line: 116 <machine registers snipped> Call Trace: xfs_trans_dup+0x211/0x250 [xfs] xfs_trans_roll+0x6d/0x180 [xfs] xfs_defer_trans_roll+0x10c/0x3b0 [xfs] xfs_defer_finish_noroll+0xdf/0x740 [xfs] xfs_defer_finish+0x13/0x70 [xfs] xfs_reflink_end_cow+0x2c6/0x680 [xfs] xfs_dio_write_end_io+0x115/0x220 [xfs] iomap_dio_complete+0x3f/0x130 iomap_dio_rw+0x3c3/0x420 xfs_file_dio_aio_write+0x132/0x3c0 [xfs] xfs_file_write_iter+0x8b/0xc0 [xfs] __vfs_write+0x193/0x1f0 vfs_write+0xba/0x1c0 ksys_write+0x52/0xc0 do_syscall_64+0x50/0x160 entry_SYSCALL_64_after_hwframe+0x49/0xbe Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> |
H A D | xfs_refcount_item.c | diff 3c919b09 Mon Sep 11 09:39:05 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: reserve less log space when recovering log intent items Wengang Wang reports that a customer's system was running a number of truncate operations on a filesystem with a very small log. Contention on the reserve heads lead to other threads stalling on smaller updates (e.g. mtime updates) long enough to result in the node being rebooted on account of the lack of responsivenes. The node failed to recover because log recovery of an EFI became stuck waiting for a grant of reserve space. From Wengang's report: "For the file deletion, log bytes are reserved basing on xfs_mount->tr_itruncate which is: tr_logres = 175488, tr_logcount = 2, tr_logflags = XFS_TRANS_PERM_LOG_RES, "You see it's a permanent log reservation with two log operations (two transactions in rolling mode). After calculation (xlog_calc_unit_res() adds space for various log headers), the final log space needed per transaction changes from 175488 to 180208 bytes. So the total log space needed is 360416 bytes (180208 * 2). [That quantity] of log space (360416 bytes) needs to be reserved for both run time inode removing (xfs_inactive_truncate()) and EFI recover (xfs_efi_item_recover())." In other words, runtime pre-reserves 360K of space in anticipation of running a chain of two transactions in which each transaction gets a 180K reservation. Now that we've allocated the transaction, we delete the bmap mapping, log an EFI to free the space, and roll the transaction as part of finishing the deferops chain. Rolling creates a new xfs_trans which shares its ticket with the old transaction. Next, xfs_trans_roll calls __xfs_trans_commit with regrant == true, which calls xlog_cil_commit with the same regrant parameter. xlog_cil_commit calls xfs_log_ticket_regrant, which decrements t_cnt and subtracts t_curr_res from the reservation and write heads. If the filesystem is fresh and the first transaction only used (say) 20K, then t_curr_res will be 160K, and we give that much reservation back to the reservation head. Or if the file is really fragmented and the first transaction actually uses 170K, then t_curr_res will be 10K, and that's what we give back to the reservation. Having done that, we're now headed into the second transaction with an EFI and 180K of reservation. Other threads apparently consumed all the reservation for smaller transactions, such as timestamp updates. Now let's say the first transaction gets written to disk and we crash without ever completing the second transaction. Now we remount the fs, log recovery finds the unfinished EFI, and calls xfs_efi_recover to finish the EFI. However, xfs_efi_recover starts a new tr_itruncate tranasction, which asks for 360K log reservation. This is a lot more than the 180K that we had reserved at the time of the crash. If the first EFI to be recovered is also pinning the tail of the log, we will be unable to free any space in the log, and recovery livelocks. Wengang confirmed this: "Now we have the second transaction which has 180208 log bytes reserved too. The second transaction is supposed to process intents including extent freeing. With my hacking patch, I blocked the extent freeing 5 hours. So in that 5 hours, 180208 (NOT 360416) log bytes are reserved. "With my test case, other transactions (update timestamps) then happen. As my hacking patch pins the journal tail, those timestamp-updating transactions finally use up (almost) all the left available log space (in memory in on disk). And finally the on disk (and in memory) available log space goes down near to 180208 bytes. Those 180208 bytes are reserved by [the] second (extent-free) transaction [in the chain]." Wengang and I noticed that EFI recovery starts a transaction, completes one step of the chain, and commits the transaction without completing any other steps of the chain. Those subsequent steps are completed by xlog_finish_defer_ops, which allocates yet another transaction to finish the rest of the chain. That transaction gets the same tr_logres as the head transaction, but with tr_logcount = 1 to force regranting with every roll to avoid livelocks. In other words, we already figured this out in commit 929b92f64048d ("xfs: xfs_defer_capture should absorb remaining transaction reservation"), but should have applied that logic to each intent item's recovery function. For Wengang's case, the xfs_trans_alloc call in the EFI recovery function should only be asking for a single transaction's worth of log reservation -- 180K, not 360K. Quoting Wengang again: "With log recovery, during EFI recovery, we use tr_itruncate again to reserve two transactions that needs 360416 log bytes. Reserving 360416 bytes fails [stalls] because we now only have about 180208 available. "Actually during the EFI recover, we only need one transaction to free the extents just like the 2nd transaction at RUNTIME. So it only needs to reserve 180208 rather than 360416 bytes. We have (a bit) more than 180208 available log bytes on disk, so [if we decrease the reservation to 180K] the reservation goes and the recovery [finishes]. That is to say: we can fix the log recover part to fix the issue. We can introduce a new xfs_trans_res xfs_mount->tr_ext_free { tr_logres = 175488, tr_logcount = 0, tr_logflags = 0, } "and use tr_ext_free instead of tr_itruncate in EFI recover." However, I don't think it quite makes sense to create an entirely new transaction reservation type to handle single-stepping during log recovery. Instead, we should copy the transaction reservation information in the xfs_mount, change tr_logcount to 1, and pass that into xfs_trans_alloc. We know this won't risk changing the min log size computation since we always ask for a fraction of the reservation for all known transaction types. This looks like it's been lurking in the codebase since commit 3d3c8b5222b92, which changed the xfs_trans_reserve call in xlog_recover_process_efi to use the tr_logcount in tr_itruncate. That changed the EFI recovery transaction from making a non-XFS_TRANS_PERM_LOG_RES request for one transaction's worth of log space to a XFS_TRANS_PERM_LOG_RES request for two transactions worth. Fixes: 3d3c8b5222b92 ("xfs: refactor xfs_trans_reserve() interface") Complements: 929b92f64048d ("xfs: xfs_defer_capture should absorb remaining transaction reservation") Suggested-by: Wengang Wang <wen.gang.wang@oracle.com> Cc: Srikanth C S <srikanth.c.s@oracle.com> [djwong: apply the same transformation to all log intent recovery] Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> diff 3c919b09 Mon Sep 11 09:39:05 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: reserve less log space when recovering log intent items Wengang Wang reports that a customer's system was running a number of truncate operations on a filesystem with a very small log. Contention on the reserve heads lead to other threads stalling on smaller updates (e.g. mtime updates) long enough to result in the node being rebooted on account of the lack of responsivenes. The node failed to recover because log recovery of an EFI became stuck waiting for a grant of reserve space. From Wengang's report: "For the file deletion, log bytes are reserved basing on xfs_mount->tr_itruncate which is: tr_logres = 175488, tr_logcount = 2, tr_logflags = XFS_TRANS_PERM_LOG_RES, "You see it's a permanent log reservation with two log operations (two transactions in rolling mode). After calculation (xlog_calc_unit_res() adds space for various log headers), the final log space needed per transaction changes from 175488 to 180208 bytes. So the total log space needed is 360416 bytes (180208 * 2). [That quantity] of log space (360416 bytes) needs to be reserved for both run time inode removing (xfs_inactive_truncate()) and EFI recover (xfs_efi_item_recover())." In other words, runtime pre-reserves 360K of space in anticipation of running a chain of two transactions in which each transaction gets a 180K reservation. Now that we've allocated the transaction, we delete the bmap mapping, log an EFI to free the space, and roll the transaction as part of finishing the deferops chain. Rolling creates a new xfs_trans which shares its ticket with the old transaction. Next, xfs_trans_roll calls __xfs_trans_commit with regrant == true, which calls xlog_cil_commit with the same regrant parameter. xlog_cil_commit calls xfs_log_ticket_regrant, which decrements t_cnt and subtracts t_curr_res from the reservation and write heads. If the filesystem is fresh and the first transaction only used (say) 20K, then t_curr_res will be 160K, and we give that much reservation back to the reservation head. Or if the file is really fragmented and the first transaction actually uses 170K, then t_curr_res will be 10K, and that's what we give back to the reservation. Having done that, we're now headed into the second transaction with an EFI and 180K of reservation. Other threads apparently consumed all the reservation for smaller transactions, such as timestamp updates. Now let's say the first transaction gets written to disk and we crash without ever completing the second transaction. Now we remount the fs, log recovery finds the unfinished EFI, and calls xfs_efi_recover to finish the EFI. However, xfs_efi_recover starts a new tr_itruncate tranasction, which asks for 360K log reservation. This is a lot more than the 180K that we had reserved at the time of the crash. If the first EFI to be recovered is also pinning the tail of the log, we will be unable to free any space in the log, and recovery livelocks. Wengang confirmed this: "Now we have the second transaction which has 180208 log bytes reserved too. The second transaction is supposed to process intents including extent freeing. With my hacking patch, I blocked the extent freeing 5 hours. So in that 5 hours, 180208 (NOT 360416) log bytes are reserved. "With my test case, other transactions (update timestamps) then happen. As my hacking patch pins the journal tail, those timestamp-updating transactions finally use up (almost) all the left available log space (in memory in on disk). And finally the on disk (and in memory) available log space goes down near to 180208 bytes. Those 180208 bytes are reserved by [the] second (extent-free) transaction [in the chain]." Wengang and I noticed that EFI recovery starts a transaction, completes one step of the chain, and commits the transaction without completing any other steps of the chain. Those subsequent steps are completed by xlog_finish_defer_ops, which allocates yet another transaction to finish the rest of the chain. That transaction gets the same tr_logres as the head transaction, but with tr_logcount = 1 to force regranting with every roll to avoid livelocks. In other words, we already figured this out in commit 929b92f64048d ("xfs: xfs_defer_capture should absorb remaining transaction reservation"), but should have applied that logic to each intent item's recovery function. For Wengang's case, the xfs_trans_alloc call in the EFI recovery function should only be asking for a single transaction's worth of log reservation -- 180K, not 360K. Quoting Wengang again: "With log recovery, during EFI recovery, we use tr_itruncate again to reserve two transactions that needs 360416 log bytes. Reserving 360416 bytes fails [stalls] because we now only have about 180208 available. "Actually during the EFI recover, we only need one transaction to free the extents just like the 2nd transaction at RUNTIME. So it only needs to reserve 180208 rather than 360416 bytes. We have (a bit) more than 180208 available log bytes on disk, so [if we decrease the reservation to 180K] the reservation goes and the recovery [finishes]. That is to say: we can fix the log recover part to fix the issue. We can introduce a new xfs_trans_res xfs_mount->tr_ext_free { tr_logres = 175488, tr_logcount = 0, tr_logflags = 0, } "and use tr_ext_free instead of tr_itruncate in EFI recover." However, I don't think it quite makes sense to create an entirely new transaction reservation type to handle single-stepping during log recovery. Instead, we should copy the transaction reservation information in the xfs_mount, change tr_logcount to 1, and pass that into xfs_trans_alloc. We know this won't risk changing the min log size computation since we always ask for a fraction of the reservation for all known transaction types. This looks like it's been lurking in the codebase since commit 3d3c8b5222b92, which changed the xfs_trans_reserve call in xlog_recover_process_efi to use the tr_logcount in tr_itruncate. That changed the EFI recovery transaction from making a non-XFS_TRANS_PERM_LOG_RES request for one transaction's worth of log space to a XFS_TRANS_PERM_LOG_RES request for two transactions worth. Fixes: 3d3c8b5222b92 ("xfs: refactor xfs_trans_reserve() interface") Complements: 929b92f64048d ("xfs: xfs_defer_capture should absorb remaining transaction reservation") Suggested-by: Wengang Wang <wen.gang.wang@oracle.com> Cc: Srikanth C S <srikanth.c.s@oracle.com> [djwong: apply the same transformation to all log intent recovery] Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> diff 0b11553e Wed Feb 01 11:16:04 MST 2023 Darrick J. Wong <djwong@kernel.org> xfs: pass refcount intent directly through the log intent code Pass the incore refcount intent through the CUI logging code instead of repeatedly boxing and unboxing parameters. Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0d79781a Sun Nov 29 17:33:37 MST 2020 Darrick J. Wong <darrick.wong@oracle.com> xfs: improve the code that checks recovered refcount intent items The code that validates recovered refcount intent items is kind of a mess -- it doesn't use the standard xfs type validators, and it doesn't check for things that it should. Fix the validator function to use the standard validation helpers and look for more types of obvious errors. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Brian Foster <bfoster@redhat.com> diff 707e0dda Mon Aug 26 01:06:22 MDT 2019 Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> fs: xfs: Remove KM_NOSLEEP and KM_SLEEP. Since no caller is using KM_NOSLEEP and no callee branches on KM_SLEEP, we can remove KM_NOSLEEP and replace KM_SLEEP with 0. Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0f37d178 Wed Aug 01 08:20:34 MDT 2018 Brian Foster <bfoster@redhat.com> xfs: pass transaction to xfs_defer_add() The majority of remaining references to struct xfs_defer_ops in XFS are associated with xfs_defer_add(). At this point, there are no more external xfs_defer_ops users left. All instances of xfs_defer_ops are embedded in the transaction, which means we can safely pass the transaction down to the dfops add interface. Update xfs_defer_add() to receive the transaction as a parameter. Various subsystems implement wrappers to allocate and construct the context specific data structures for the associated deferred operation type. Update these to also carry the transaction down as needed and clean up unused dfops parameters along the way. This removes most of the remaining references to struct xfs_defer_ops throughout the code and facilitates removal of the structure. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Christoph Hellwig <hch@lst.de> [darrick: fix unused variable warnings with ftrace disabled] Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> |
/linux-master/fs/xfs/scrub/ | ||
H A D | common.c | diff 6c728952 Fri Dec 15 11:03:37 MST 2023 Darrick J. Wong <djwong@kernel.org> xfs: abort directory parent scrub scans if we encounter a zapped directory In a previous patch, we added some code to perform sufficient repairs to an ondisk inode record such that the inode cache would be willing to load the inode. If the broken inode was a shortform directory, it will reset the directory to something plausible, which is to say an empty subdirectory of the root. The telltale signs that something is seriously wrong is the broken link count. Such directories look clean, but they shouldn't participate in a filesystem scan to find or confirm a directory parent pointer. Create a predicate that identifies such directories and abort the scrub. Found by fuzzing xfs/1554 with multithreaded xfs_scrub enabled and u3.bmx[0].startblock = zeroes. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 0d296634 Thu Aug 10 08:48:12 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: hide xfs_inode_is_allocated in scrub common code This function is only used by online fsck, so let's move it there. In the next patch, we'll fix it to work properly and to require that the caller hold the AGI buffer locked. No major changes aside from adjusting the signature a bit. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> diff 38bb1310 Tue Apr 11 20:00:22 MDT 2023 Darrick J. Wong <djwong@kernel.org> xfs: retain the AGI when we can't iget an inode to scrub the core xchk_get_inode is not quite the right function to be calling from the inode scrubber setup function. The common get_inode function either gets an inode and installs it in the scrub context, or it returns an error code explaining what happened. This is acceptable for most file scrubbers because it is not in their scope to fix corruptions in the inode core and fork areas that cause iget to fail. Dealing with these problems is within the scope of the inode scrubber, however. If iget fails with EFSCORRUPTED, we need to xchk_inode to flag that as corruption. Since we can't get our hands on an incore inode, we need to hold the AGI to prevent inode allocation activity so that nothing changes in the inode metadata. Looking ahead to the inode core repair patches, we will also need to hold the AGI buffer into xrep_inode so that we can make modifications to the xfs_dinode structure without any other thread swooping in to allocate or free the inode. Adapt the xchk_get_inode into xchk_setup_inode since this is a one-off use case where the error codes we check for are a little different, and the return state is much different from the common function. xchk_setup_inode prepares to check or repair an inode record, so it must continue the scrub operation even if the inode/inobt verifiers cause xfs_iget to return EFSCORRUPTED. This is done by attaching the locked AGI buffer to the scrub transaction and returning 0 to move on to the actual scrub. (Later, the online inode repair code will also want the xfs_imap structure so that it can reset the ondisk xfs_dinode structure.) xchk_get_inode retrieves an inode on behalf of a scrubber that operates on an incore inode -- data/attr/cow forks, directories, xattrs, symlinks, parent pointers, etc. If the inode/inobt verifiers fail and xfs_iget returns EFSCORRUPTED, we want to exit to userspace (because the caller should be fix the inode first) and drop everything we acquired along the way. A behavior common to both functions is that it's possible that xfs_scrub asked for a scrub-by-handle concurrent with the inode being freed or the passed-in inumber is invalid. In this case, we call xfs_imap to see if the inobt index thinks the inode is allocated, and return ENOENT ("nothing to check here") to userspace if this is not the case. The imap lookup is why both functions call xchk_iget_agi. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> diff ebd9027d Wed Aug 18 19:46:55 MDT 2021 Dave Chinner <dchinner@redhat.com> xfs: convert xfs_sb_version_has checks to use mount features This is a conversion of the remaining xfs_sb_version_has..(sbp) checks to use xfs_has_..(mp) feature checks. This was largely done with a vim replacement macro that did: :0,$s/xfs_sb_version_has\(.*\)&\(.*\)->m_sb/xfs_has_\1\2/g<CR> A couple of other variants were also used, and the rest touched up by hand. $ size -t fs/xfs/built-in.a text data bss dec hex filename before 1127533 311352 484 1439369 15f689 (TOTALS) after 1125360 311352 484 1437196 15ee0c (TOTALS) Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> |
/linux-master/fs/xfs/libxfs/ | ||
H A D | xfs_sb.c | diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff de6077ec Thu Feb 22 01:31:03 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: report ag header corruption errors to the health tracking system Whenever we encounter a corrupt AG header, we should report that to the health monitoring system for later reporting. Buffer readers that don't respond to corruption events with a _mark_sick call can be detected with the following script: #!/bin/bash # Detect missing calls to xfs_*_mark_sick filter=cat tty -s && filter=less git grep -A10 -E '( = xfs_trans_read_buf| = xfs_buf_read\()' fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] | awk ' BEGIN { ignore = 0; lineno = 0; delete lines; } { if ($0 == "--") { if (!ignore) { for (i = 0; i < lineno; i++) { print(lines[i]); } printf("--\n"); } delete lines; lineno = 0; ignore = 0; } else if ($0 ~ /mark_sick/) { ignore = 1; } else { lines[lineno++] = $0; } } ' | $filter Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> diff 881f78f4 Mon Jan 29 21:27:23 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: remove conditional building of rt geometry validator functions I mistakenly turned off CONFIG_XFS_RT in the Kconfig file for arm64 variant of the djwong-wtf git branch. Unfortunately, it took me a good hour to figure out that RT wasn't built because this is what got printed to dmesg: XFS (sda2): realtime geometry sanity check failed XFS (sda2): Metadata corruption detected at xfs_sb_read_verify+0x170/0x190 [xfs], xfs_sb block 0x0 Whereas I would have expected: XFS (sda2): Not built with CONFIG_XFS_RT XFS (sda2): RT mount failed The root cause of these problems is the conditional compilation of the new functions xfs_validate_rtextents and xfs_compute_rextslog that I introduced in the two commits listed below. The !RT versions of these functions return false and 0, respectively, which causes primary superblock validation to fail, which explains the first message. Move the two functions to other parts of libxfs that are not conditionally defined by CONFIG_XFS_RT and remove the broken stubs so that validation works again. Fixes: e14293803f4e ("xfs: don't allow overly small or large realtime volumes") Fixes: a6a38f309afc ("xfs: make rextslog computation consistent with mkfs") Signed-off-by: "Darrick J. Wong" <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff 881f78f4 Mon Jan 29 21:27:23 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: remove conditional building of rt geometry validator functions I mistakenly turned off CONFIG_XFS_RT in the Kconfig file for arm64 variant of the djwong-wtf git branch. Unfortunately, it took me a good hour to figure out that RT wasn't built because this is what got printed to dmesg: XFS (sda2): realtime geometry sanity check failed XFS (sda2): Metadata corruption detected at xfs_sb_read_verify+0x170/0x190 [xfs], xfs_sb block 0x0 Whereas I would have expected: XFS (sda2): Not built with CONFIG_XFS_RT XFS (sda2): RT mount failed The root cause of these problems is the conditional compilation of the new functions xfs_validate_rtextents and xfs_compute_rextslog that I introduced in the two commits listed below. The !RT versions of these functions return false and 0, respectively, which causes primary superblock validation to fail, which explains the first message. Move the two functions to other parts of libxfs that are not conditionally defined by CONFIG_XFS_RT and remove the broken stubs so that validation works again. Fixes: e14293803f4e ("xfs: don't allow overly small or large realtime volumes") Fixes: a6a38f309afc ("xfs: make rextslog computation consistent with mkfs") Signed-off-by: "Darrick J. Wong" <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff 881f78f4 Mon Jan 29 21:27:23 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: remove conditional building of rt geometry validator functions I mistakenly turned off CONFIG_XFS_RT in the Kconfig file for arm64 variant of the djwong-wtf git branch. Unfortunately, it took me a good hour to figure out that RT wasn't built because this is what got printed to dmesg: XFS (sda2): realtime geometry sanity check failed XFS (sda2): Metadata corruption detected at xfs_sb_read_verify+0x170/0x190 [xfs], xfs_sb block 0x0 Whereas I would have expected: XFS (sda2): Not built with CONFIG_XFS_RT XFS (sda2): RT mount failed The root cause of these problems is the conditional compilation of the new functions xfs_validate_rtextents and xfs_compute_rextslog that I introduced in the two commits listed below. The !RT versions of these functions return false and 0, respectively, which causes primary superblock validation to fail, which explains the first message. Move the two functions to other parts of libxfs that are not conditionally defined by CONFIG_XFS_RT and remove the broken stubs so that validation works again. Fixes: e14293803f4e ("xfs: don't allow overly small or large realtime volumes") Fixes: a6a38f309afc ("xfs: make rextslog computation consistent with mkfs") Signed-off-by: "Darrick J. Wong" <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff 881f78f4 Mon Jan 29 21:27:23 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: remove conditional building of rt geometry validator functions I mistakenly turned off CONFIG_XFS_RT in the Kconfig file for arm64 variant of the djwong-wtf git branch. Unfortunately, it took me a good hour to figure out that RT wasn't built because this is what got printed to dmesg: XFS (sda2): realtime geometry sanity check failed XFS (sda2): Metadata corruption detected at xfs_sb_read_verify+0x170/0x190 [xfs], xfs_sb block 0x0 Whereas I would have expected: XFS (sda2): Not built with CONFIG_XFS_RT XFS (sda2): RT mount failed The root cause of these problems is the conditional compilation of the new functions xfs_validate_rtextents and xfs_compute_rextslog that I introduced in the two commits listed below. The !RT versions of these functions return false and 0, respectively, which causes primary superblock validation to fail, which explains the first message. Move the two functions to other parts of libxfs that are not conditionally defined by CONFIG_XFS_RT and remove the broken stubs so that validation works again. Fixes: e14293803f4e ("xfs: don't allow overly small or large realtime volumes") Fixes: a6a38f309afc ("xfs: make rextslog computation consistent with mkfs") Signed-off-by: "Darrick J. Wong" <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> |
H A D | xfs_sb.h | diff 881f78f4 Mon Jan 29 21:27:23 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: remove conditional building of rt geometry validator functions I mistakenly turned off CONFIG_XFS_RT in the Kconfig file for arm64 variant of the djwong-wtf git branch. Unfortunately, it took me a good hour to figure out that RT wasn't built because this is what got printed to dmesg: XFS (sda2): realtime geometry sanity check failed XFS (sda2): Metadata corruption detected at xfs_sb_read_verify+0x170/0x190 [xfs], xfs_sb block 0x0 Whereas I would have expected: XFS (sda2): Not built with CONFIG_XFS_RT XFS (sda2): RT mount failed The root cause of these problems is the conditional compilation of the new functions xfs_validate_rtextents and xfs_compute_rextslog that I introduced in the two commits listed below. The !RT versions of these functions return false and 0, respectively, which causes primary superblock validation to fail, which explains the first message. Move the two functions to other parts of libxfs that are not conditionally defined by CONFIG_XFS_RT and remove the broken stubs so that validation works again. Fixes: e14293803f4e ("xfs: don't allow overly small or large realtime volumes") Fixes: a6a38f309afc ("xfs: make rextslog computation consistent with mkfs") Signed-off-by: "Darrick J. Wong" <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff 881f78f4 Mon Jan 29 21:27:23 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: remove conditional building of rt geometry validator functions I mistakenly turned off CONFIG_XFS_RT in the Kconfig file for arm64 variant of the djwong-wtf git branch. Unfortunately, it took me a good hour to figure out that RT wasn't built because this is what got printed to dmesg: XFS (sda2): realtime geometry sanity check failed XFS (sda2): Metadata corruption detected at xfs_sb_read_verify+0x170/0x190 [xfs], xfs_sb block 0x0 Whereas I would have expected: XFS (sda2): Not built with CONFIG_XFS_RT XFS (sda2): RT mount failed The root cause of these problems is the conditional compilation of the new functions xfs_validate_rtextents and xfs_compute_rextslog that I introduced in the two commits listed below. The !RT versions of these functions return false and 0, respectively, which causes primary superblock validation to fail, which explains the first message. Move the two functions to other parts of libxfs that are not conditionally defined by CONFIG_XFS_RT and remove the broken stubs so that validation works again. Fixes: e14293803f4e ("xfs: don't allow overly small or large realtime volumes") Fixes: a6a38f309afc ("xfs: make rextslog computation consistent with mkfs") Signed-off-by: "Darrick J. Wong" <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff 881f78f4 Mon Jan 29 21:27:23 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: remove conditional building of rt geometry validator functions I mistakenly turned off CONFIG_XFS_RT in the Kconfig file for arm64 variant of the djwong-wtf git branch. Unfortunately, it took me a good hour to figure out that RT wasn't built because this is what got printed to dmesg: XFS (sda2): realtime geometry sanity check failed XFS (sda2): Metadata corruption detected at xfs_sb_read_verify+0x170/0x190 [xfs], xfs_sb block 0x0 Whereas I would have expected: XFS (sda2): Not built with CONFIG_XFS_RT XFS (sda2): RT mount failed The root cause of these problems is the conditional compilation of the new functions xfs_validate_rtextents and xfs_compute_rextslog that I introduced in the two commits listed below. The !RT versions of these functions return false and 0, respectively, which causes primary superblock validation to fail, which explains the first message. Move the two functions to other parts of libxfs that are not conditionally defined by CONFIG_XFS_RT and remove the broken stubs so that validation works again. Fixes: e14293803f4e ("xfs: don't allow overly small or large realtime volumes") Fixes: a6a38f309afc ("xfs: make rextslog computation consistent with mkfs") Signed-off-by: "Darrick J. Wong" <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff 881f78f4 Mon Jan 29 21:27:23 MST 2024 Darrick J. Wong <djwong@kernel.org> xfs: remove conditional building of rt geometry validator functions I mistakenly turned off CONFIG_XFS_RT in the Kconfig file for arm64 variant of the djwong-wtf git branch. Unfortunately, it took me a good hour to figure out that RT wasn't built because this is what got printed to dmesg: XFS (sda2): realtime geometry sanity check failed XFS (sda2): Metadata corruption detected at xfs_sb_read_verify+0x170/0x190 [xfs], xfs_sb block 0x0 Whereas I would have expected: XFS (sda2): Not built with CONFIG_XFS_RT XFS (sda2): RT mount failed The root cause of these problems is the conditional compilation of the new functions xfs_validate_rtextents and xfs_compute_rextslog that I introduced in the two commits listed below. The !RT versions of these functions return false and 0, respectively, which causes primary superblock validation to fail, which explains the first message. Move the two functions to other parts of libxfs that are not conditionally defined by CONFIG_XFS_RT and remove the broken stubs so that validation works again. Fixes: e14293803f4e ("xfs: don't allow overly small or large realtime volumes") Fixes: a6a38f309afc ("xfs: make rextslog computation consistent with mkfs") Signed-off-by: "Darrick J. Wong" <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanbabu@kernel.org> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> diff 0b61f8a4 Tue Jun 05 20:42:14 MDT 2018 Dave Chinner <dchinner@redhat.com> xfs: convert to SPDX license tags Remove the verbose license text from XFS files and replace them with SPDX tags. This does not change the license of any of the code, merely refers to the common, up-to-date license files in LICENSES/ This change was mostly scripted. fs/xfs/Makefile and fs/xfs/libxfs/xfs_fs.h were modified by hand, the rest were detected and modified by the following command: for f in `git grep -l "GNU General" fs/xfs/` ; do echo $f cat $f | awk -f hdr.awk > $f.new mv -f $f.new $f done And the hdr.awk script that did the modification (including detecting the difference between GPL-2.0 and GPL-2.0+ licenses) is as follows: $ cat hdr.awk BEGIN { hdr = 1.0 tag = "GPL-2.0" str = "" } /^ \* This program is free software/ { hdr = 2.0; next } /any later version./ { tag = "GPL-2.0+" next } /^ \*\// { if (hdr > 0.0) { print "// SPDX-License-Identifier: " tag print str print $0 str="" hdr = 0.0 next } print $0 next } /^ \* / { if (hdr > 1.0) next if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 next } /^ \*/ { if (hdr > 0.0) next print $0 next } // { if (hdr > 0.0) { if (str != "") str = str "\n" str = str $0 next } print $0 } END { } $ Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> |
Completed in 1388 milliseconds