zonefs: fix synchronous direct writes to sequential files

Commit 16d7fd3cfa72 ("zonefs: use iomap for synchronous direct writes")
changes zonefs code from a self-built zone append BIO to using iomap for
synchronous direct writes. This change relies on iomap submit BIO
callback to change the write BIO built by iomap to a zone append BIO.
However, this change overlooked the fact that a write BIO may be very
large as it is split when issued. The change from a regular write to a
zone append operation for the built BIO can result in a block layer
warning as zone append BIO are not allowed to be split.

WARNING: CPU: 18 PID: 202210 at block/bio.c:1644 bio_split+0x288/0x350
Call Trace:
? __warn+0xc9/0x2b0
? bio_split+0x288/0x350
? report_bug+0x2e6/0x390
? handle_bug+0x41/0x80
? exc_invalid_op+0x13/0x40
? asm_exc_invalid_op+0x16/0x20
? bio_split+0x288/0x350
bio_split_rw+0x4bc/0x810
? __pfx_bio_split_rw+0x10/0x10
? lockdep_unlock+0xf2/0x250
__bio_split_to_limits+0x1d8/0x900
blk_mq_submit_bio+0x1cf/0x18a0
? __pfx_iov_iter_extract_pages+0x10/0x10
? __pfx_blk_mq_submit_bio+0x10/0x10
? find_held_lock+0x2d/0x110
? lock_release+0x362/0x620
? mark_held_locks+0x9e/0xe0
__submit_bio+0x1ea/0x290
? __pfx___submit_bio+0x10/0x10
? seqcount_lockdep_reader_access.constprop.0+0x82/0x90
submit_bio_noacct_nocheck+0x675/0xa20
? __pfx_bio_iov_iter_get_pages+0x10/0x10
? __pfx_submit_bio_noacct_nocheck+0x10/0x10
iomap_dio_bio_iter+0x624/0x1280
__iomap_dio_rw+0xa22/0x18a0
? lock_is_held_type+0xe3/0x140
? __pfx___iomap_dio_rw+0x10/0x10
? lock_release+0x362/0x620
? zonefs_file_write_iter+0x74c/0xc80 [zonefs]
? down_write+0x13d/0x1e0
iomap_dio_rw+0xe/0x40
zonefs_file_write_iter+0x5ea/0xc80 [zonefs]
do_iter_readv_writev+0x18b/0x2c0
? __pfx_do_iter_readv_writev+0x10/0x10
? inode_security+0x54/0xf0
do_iter_write+0x13b/0x7c0
? lock_is_held_type+0xe3/0x140
vfs_writev+0x185/0x550
? __pfx_vfs_writev+0x10/0x10
? __handle_mm_fault+0x9bd/0x1c90
? find_held_lock+0x2d/0x110
? lock_release+0x362/0x620
? find_held_lock+0x2d/0x110
? lock_release+0x362/0x620
? __up_read+0x1ea/0x720
? do_pwritev+0x136/0x1f0
do_pwritev+0x136/0x1f0
? __pfx_do_pwritev+0x10/0x10
? syscall_enter_from_user_mode+0x22/0x90
? lockdep_hardirqs_on+0x7d/0x100
do_syscall_64+0x58/0x80

This error depends on the hardware used, specifically on the max zone
append bytes and max_[hw_]sectors limits. Tests using AMD Epyc machines
that have low limits did not reveal this issue while runs on Intel Xeon
machines with larger limits trigger it.

Manually splitting the zone append BIO using bio_split_rw() can solve
this issue but also requires issuing the fragment BIOs synchronously
with submit_bio_wait(), to avoid potential reordering of the zone append
BIO fragments, which would lead to data corruption. That is, this
solution is not better than using regular write BIOs which are subject
to serialization using zone write locking at the IO scheduler level.

Given this, fix the issue by removing zone append support and using
regular write BIOs for synchronous direct writes. This allows preseving
the use of iomap and having identical synchronous and asynchronous
sequential file write path. Zone append support will be reintroduced
later through io_uring commands to ensure that the needed special
handling is done correctly.

Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Fixes: 16d7fd3cfa72 ("zonefs: use iomap for synchronous direct writes")
Cc: stable@vger.kernel.org
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Tested-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>

zonefs: use iomap for synchronous direct writes

Remove the function zonefs_file_dio_append() that is used to manually
issue REQ_OP_ZONE_APPEND BIOs for processing synchronous direct writes
and use iomap instead.

To preserve the use of zone append operations for synchronous writes,
different struct iomap_dio_ops are defined. For synchronous direct
writes using zone append, zonefs_zone_append_dio_ops is introduced.
The submit_bio operation of this structure is defined as the function
zonefs_file_zone_append_dio_submit_io() which is used to change the BIO
opreation for synchronous direct IO writes to REQ_OP_ZONE_APPEND.

In order to preserve the write location check on completion of zone
append BIOs, the end_io operation is also defined using the function
zonefs_file_zone_append_dio_bio_end_io(). This check now relies on the
zonefs_zone_append_bio structure, allocated together with zone append
BIOs with a dedicated BIO set. This structure include the target inode
of a zone append BIO as well as the target append offset location for
the zone append operation. This is used to perform a check against
bio->bi_iter.bi_sector when the BIO completes, without needing to use
the zone information z_wpoffset field, thus removing the need for
taking the inode truncate mutex.

Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>

zonefs: Cache zone group directory inodes

Since looking up any zone file inode requires looking up first the inode
for the directory representing the zone group of the file, ensuring that
the zone group inodes are always cached is desired. To do so, take an
extra reference on the zone groups directory inodes on mount, thus
avoiding the eviction of these inodes from the inode cache until the
volume is unmounted.

Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>

zonefs: Dynamically create file inodes when needed

Allocating and initializing all inodes and dentries for all files
results in a very large memory usage with high capacity zoned block
devices. For instance, with a 26 TB SMR HDD with over 96000 zones,
mounting the disk with zonefs results in about 130 MB of memory used,
the vast majority of this space being used for vfs inodes and dentries.

However, since a user will rarely access all zones at the same time,
dynamically creating file inodes and dentries on demand, similarly to
regular file systems, can significantly reduce memory usage.

This patch modifies mount processing to not create the inodes and
dentries for zone files. Instead, the directory inode operation
zonefs_lookup() and directory file operation zonefs_readdir() are
introduced to allocate and initialize inodes on-demand using the helper
functions zonefs_get_dir_inode() and zonefs_get_zgroup_inode().

Implementation of these functions is simple, relying on the static
nature of zonefs directories and files. Directory inodes are linked to
the volume zone groups (struct zonefs_zone_group) they represent by
using the directory inode i_private field. This simplifies the
implementation of the lookup and readdir operations.

Unreferenced zone file inodes can be evicted from the inode cache at any
time. In such case, the only inode information that cannot be recreated
from the zone information that is saved in the zone group data
structures attached to the volume super block is the inode uid, gid and
access rights. These values may have been changed by the user. To keep
these attributes for the life time of the mount, as before, the inode
mode, uid and gid are saved in the inode zone information and the saved
values are used to initialize regular file inodes when an inode lookup
happens. The zone information mode, uid and gid are initialized in
zonefs_init_zgroup() using the default values.

With these changes, the static minimal memory usage of a zonefs volume
is mostly reduced to the array of zone information for each zone group.
For the 26 TB SMR hard-disk mentioned above, the memory usage after
mount becomes about 5.4 MB, a reduction by a factor of 24 from the
initial 130 MB memory use.

Co-developed-by: Jorgen Hansen <Jorgen.Hansen@wdc.com>
Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>

zonefs: Separate zone information from inode information

In preparation for adding dynamic inode allocation, separate an inode
zone information from the zonefs inode structure. The new data structure
zonefs_zone is introduced to store in memory information about a zone
that must be kept throughout the lifetime of the device mount.

Linking between a zone file inode and its zone information is done by
setting the inode i_private field to point to a struct zonefs_zone.
Using the i_private pointer avoids the need for adding a pointer in
struct zonefs_inode_info. Beside the vfs inode, this structure is
reduced to a mutex and a write open counter.

One struct zonefs_zone is created per file inode on mount. These
structures are organized in an array using the new struct
zonefs_zone_group data structure to represent zone groups. The
zonefs_zone arrays are indexed per file number (the index of a struct
zonefs_zone in its array directly gives the file number/name for that
zone file inode).

Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>

zonefs: Reduce struct zonefs_inode_info size

Instead of using the i_ztype field in struct zonefs_inode_info to
indicate the zone type of an inode, introduce the new inode flag
ZONEFS_ZONE_CNV to be set in the i_flags field of struct
zonefs_inode_info to identify conventional zones. If this flag is not
set, the zone of an inode is considered to be a sequential zone.

The helpers zonefs_zone_is_cnv(), zonefs_zone_is_seq(),
zonefs_inode_is_cnv() and zonefs_inode_is_seq() are introduced to
simplify testing the zone type of a struct zonefs_inode_info and of a
struct inode.

Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>

zonefs: Simplify IO error handling

Simplify zonefs_check_zone_condition() by moving the code that changes
an inode access rights to the new function zonefs_inode_update_mode().
Furthermore, since on mount an inode wpoffset is always zero when
zonefs_check_zone_condition() is called during an inode initialization,
the "mount" boolean argument is not necessary for the readonly zone
case. This argument is thus removed.

zonefs_io_error_cb() is also modified to use the inode offline and
zone state flags instead of checking the device zone condition. The
multiple calls to zonefs_check_zone_condition() are reduced to the first
call on entry, which allows removing the "warn" argument.
zonefs_inode_update_mode() is also used to update an inode access rights
as zonefs_io_error_cb() modifies the inode flags depending on the volume
error handling mode (defined with a mount option). Since an inode mode
change differs for read-only zones between mount time and IO error time,
the flag ZONEFS_ZONE_INIT_MODE is used to differentiate both cases.

Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>

zonefs: Reorganize code

Move all code related to zone file operations from super.c to the new
file.c file. Inode and zone management code remains in super.c.

Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>

zonefs: Fix active zone accounting

If a file zone transitions to the offline or readonly state from an
active state, we must clear the zone active flag and decrement the
active seq file counter. Do so in zonefs_account_active() using the new
zonefs inode flags ZONEFS_ZONE_OFFLINE and ZONEFS_ZONE_READONLY. These
flags are set if necessary in zonefs_check_zone_condition() based on the
result of report zones operation after an IO error.

Fixes: 87c9ce3ffec9 ("zonefs: Add active seq file accounting")
Cc: stable@vger.kernel.org
Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>

zonefs: Add active seq file accounting

Modify struct zonefs_sb_info to add the s_active_seq_files atomic to
count the number of seq files representing a zone that is partially
written or explicitly open, that is, to count sequential files with
a zone that is in an active state on the device.

The helper function zonefs_account_active() is introduced to update
this counter whenever a file is written or truncated. This helper is
also used in the zonefs_seq_file_write_open() and
zonefs_seq_file_write_close() functions when the explicit_open mount
option is used.

The s_active_seq_files counter is exported through sysfs using the
read-only attribute nr_active_seq_files. The device maximum number of
active zones is also exported through sysfs with the read-only attribute
max_active_seq_files.

Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Hans Holmberg <hans.holmberg@wdc.com>

zonefs: Export open zone resource information through sysfs

To allow applications to easily check the current usage status of the
open zone resources of the mounted device, export through sysfs the
counter of write open sequential files s_wro_seq_files field of
struct zonefs_sb_info. The attribute is named nr_wro_seq_files and is
read only.

The maximum number of write open sequential files (zones) indicated by
the s_max_wro_seq_files field of struct zonefs_sb_info is also exported
as the read only attribute max_wro_seq_files.

Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Hans Holmberg <hans.holmberg@wdc.com>

zonefs: Rename super block information fields

The s_open_zones field of struct zonefs_sb_info is used to count the
number of files that are open for writing and may not necessarilly
correspond to the number of open zones on the device. For instance, an
application may open for writing a sequential zone file, fully write it
and keep the file open. In such case, the zone of the file is not open
anymore (it is in the full state).

Avoid confusion about this counter meaning by renaming it to
s_wro_seq_files. To keep things consistent, the field s_max_open_zones
is renamed to s_max_wro_seq_files.

Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Hans Holmberg <hans.holmberg@wdc.com>

zonefs: Convert to using invalidate_lock

Use invalidate_lock instead of zonefs' private i_mmap_sem. The intended
purpose is exactly the same.

CC: Damien Le Moal <damien.lemoal@wdc.com>
CC: Johannes Thumshirn <jth@kernel.org>
CC: <linux-fsdevel@vger.kernel.org>
Acked-by: Damien Le Moal <damien.lemoal@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>

zonefs: open/close zone on file open/close

NVMe Zoned Namespace introduced the concept of active zones, which are
zones in the implicit open, explicit open or closed condition. Drives may
have a limit on the number of zones that can be simultaneously active.
This potential limitation translate into a risk for applications to see
write IO errors due to this limit if the zone of a file being written to is
not already active when a write request is issued.

To avoid these potential errors, the zone of a file can explicitly be made
active using an open zone command when the file is open for the first
time. If the zone open command succeeds, the application is then
guaranteed that write requests can be processed. This indirect management
of active zones relies on the maximum number of open zones of a drive,
which is always lower or equal to the maximum number of active zones.

On the first open of a sequential zone file, send a REQ_OP_ZONE_OPEN
command to the block device. Conversely, on the last release of a zone
file and send a REQ_OP_ZONE_CLOSE to the device if the zone is not full or
empty.

As truncating a zone file to 0 or max can deactivate a zone as well, we
need to serialize against truncates and also be careful not to close a
zone as the file may still be open for writing, e.g. the user called
ftruncate(). If the zone file is not open and a process does a truncate(),
then no close operation is needed.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>

zonefs: add zone-capacity support

In the zoned storage model, the sectors within a zone are typically all
writeable. With the introduction of the Zoned Namespace (ZNS) Command
Set in the NVM Express organization, the model was extended to have a
specific writeable capacity.

This zone capacity can be less than the overall zone size for a NVMe ZNS
device or null_blk in zoned-mode. For other ZBC/ZAC devices the zone
capacity is always equal to the zone size.

Use the zone capacity field instead from blk_zone for determining the
maximum inode size and inode blocks in zonefs.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>

fs: New zonefs file system

zonefs is a very simple file system exposing each zone of a zoned block
device as a file. Unlike a regular file system with zoned block device
support (e.g. f2fs), zonefs does not hide the sequential write
constraint of zoned block devices to the user. Files representing
sequential write zones of the device must be written sequentially
starting from the end of the file (append only writes).

As such, zonefs is in essence closer to a raw block device access
interface than to a full featured POSIX file system. The goal of zonefs
is to simplify the implementation of zoned block device support in
applications by replacing raw block device file accesses with a richer
file API, avoiding relying on direct block device file ioctls which may
be more obscure to developers. One example of this approach is the
implementation of LSM (log-structured merge) tree structures (such as
used in RocksDB and LevelDB) on zoned block devices by allowing SSTables
to be stored in a zone file similarly to a regular file system rather
than as a range of sectors of a zoned device. The introduction of the
higher level construct "one file is one zone" can help reducing the
amount of changes needed in the application as well as introducing
support for different application programming languages.

Zonefs on-disk metadata is reduced to an immutable super block to
persistently store a magic number and optional feature flags and
values. On mount, zonefs uses blkdev_report_zones() to obtain the device
zone configuration and populates the mount point with a static file tree
solely based on this information. E.g. file sizes come from the device
zone type and write pointer offset managed by the device itself.

The zone files created on mount have the following characteristics.
1) Files representing zones of the same type are grouped together
under a common sub-directory:
* For conventional zones, the sub-directory "cnv" is used.
* For sequential write zones, the sub-directory "seq" is used.
These two directories are the only directories that exist in zonefs.
Users cannot create other directories and cannot rename nor delete
the "cnv" and "seq" sub-directories.
2) The name of zone files is the number of the file within the zone
type sub-directory, in order of increasing zone start sector.
3) The size of conventional zone files is fixed to the device zone size.
Conventional zone files cannot be truncated.
4) The size of sequential zone files represent the file's zone write
pointer position relative to the zone start sector. Truncating these
files is allowed only down to 0, in which case, the zone is reset to
rewind the zone write pointer position to the start of the zone, or
up to the zone size, in which case the file's zone is transitioned
to the FULL state (finish zone operation).
5) All read and write operations to files are not allowed beyond the
file zone size. Any access exceeding the zone size is failed with
the -EFBIG error.
6) Creating, deleting, renaming or modifying any attribute of files and
sub-directories is not allowed.
7) There are no restrictions on the type of read and write operations
that can be issued to conventional zone files. Buffered, direct and
mmap read & write operations are accepted. For sequential zone files,
there are no restrictions on read operations, but all write
operations must be direct IO append writes. mmap write of sequential
files is not allowed.

Several optional features of zonefs can be enabled at format time.
* Conventional zone aggregation: ranges of contiguous conventional
zones can be aggregated into a single larger file instead of the
default one file per zone.
* File ownership: The owner UID and GID of zone files is by default 0
(root) but can be changed to any valid UID/GID.
* File access permissions: the default 640 access permissions can be
changed.

The mkzonefs tool is used to format zoned block devices for use with
zonefs. This tool is available on Github at:

git@github.com:damien-lemoal/zonefs-tools.git.

zonefs-tools also includes a test suite which can be run against any
zoned block device, including null_blk block device created with zoned
mode.

Example: the following formats a 15TB host-managed SMR HDD with 256 MB
zones with the conventional zones aggregation feature enabled.

$ sudo mkzonefs -o aggr_cnv /dev/sdX
$ sudo mount -t zonefs /dev/sdX /mnt
$ ls -l /mnt/
total 0
dr-xr-xr-x 2 root root 1 Nov 25 13:23 cnv
dr-xr-xr-x 2 root root 55356 Nov 25 13:23 seq

The size of the zone files sub-directories indicate the number of files
existing for each type of zones. In this example, there is only one
conventional zone file (all conventional zones are aggregated under a
single file).

$ ls -l /mnt/cnv
total 137101312
-rw-r----- 1 root root 140391743488 Nov 25 13:23 0

This aggregated conventional zone file can be used as a regular file.

$ sudo mkfs.ext4 /mnt/cnv/0
$ sudo mount -o loop /mnt/cnv/0 /data

The "seq" sub-directory grouping files for sequential write zones has
in this example 55356 zones.

$ ls -lv /mnt/seq
total 14511243264
-rw-r----- 1 root root 0 Nov 25 13:23 0
-rw-r----- 1 root root 0 Nov 25 13:23 1
-rw-r----- 1 root root 0 Nov 25 13:23 2
...
-rw-r----- 1 root root 0 Nov 25 13:23 55354
-rw-r----- 1 root root 0 Nov 25 13:23 55355

For sequential write zone files, the file size changes as data is
appended at the end of the file, similarly to any regular file system.

$ dd if=/dev/zero of=/mnt/seq/0 bs=4K count=1 conv=notrunc oflag=direct
1+0 records in
1+0 records out
4096 bytes (4.1 kB, 4.0 KiB) copied, 0.000452219 s, 9.1 MB/s

$ ls -l /mnt/seq/0
-rw-r----- 1 root root 4096 Nov 25 13:23 /mnt/seq/0

The written file can be truncated to the zone size, preventing any
further write operation.

$ truncate -s 268435456 /mnt/seq/0
$ ls -l /mnt/seq/0
-rw-r----- 1 root root 268435456 Nov 25 13:49 /mnt/seq/0

Truncation to 0 size allows freeing the file zone storage space and
restart append-writes to the file.

$ truncate -s 0 /mnt/seq/0
$ ls -l /mnt/seq/0
-rw-r----- 1 root root 0 Nov 25 13:49 /mnt/seq/0

Since files are statically mapped to zones on the disk, the number of
blocks of a file as reported by stat() and fstat() indicates the size
of the file zone.

$ stat /mnt/seq/0
File: /mnt/seq/0
Size: 0 Blocks: 524288 IO Block: 4096 regular empty file
Device: 870h/2160d Inode: 50431 Links: 1
Access: (0640/-rw-r-----) Uid: ( 0/ root) Gid: ( 0/ root)
Access: 2019-11-25 13:23:57.048971997 +0900
Modify: 2019-11-25 13:52:25.553805765 +0900
Change: 2019-11-25 13:52:25.553805765 +0900
Birth: -

The number of blocks of the file ("Blocks") in units of 512B blocks
gives the maximum file size of 524288 * 512 B = 256 MB, corresponding
to the device zone size in this example. Of note is that the "IO block"
field always indicates the minimum IO size for writes and corresponds
to the device physical sector size.

This code contains contributions from:
* Johannes Thumshirn <jthumshirn@suse.de>,
* Darrick J. Wong <darrick.wong@oracle.com>,
* Christoph Hellwig <hch@lst.de>,
* Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com> and
* Ting Yao <tingyao@hust.edu.cn>.

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>