Searched +hist:2 +hist:a9d6481 (Results 1 - 2 of 2) sorted by relevance
| /linux-master/ipc/ | ||
| H A D | util.h | diff 72d1e611 Tue Sep 13 13:25:38 MDT 2022 Jiebin Sun <jiebin.sun@intel.com> ipc/msg: mitigate the lock contention with percpu counter The msg_bytes and msg_hdrs atomic counters are frequently updated when IPC msg queue is in heavy use, causing heavy cache bounce and overhead. Change them to percpu_counter greatly improve the performance. Since there is one percpu struct per namespace, additional memory cost is minimal. Reading of the count done in msgctl call, which is infrequent. So the need to sum up the counts in each CPU is infrequent. Apply the patch and test the pts/stress-ng-1.4.0 -- system v message passing (160 threads). Score gain: 3.99x CPU: ICX 8380 x 2 sockets Core number: 40 x 2 physical cores Benchmark: pts/stress-ng-1.4.0 -- system v message passing (160 threads) [akpm@linux-foundation.org: coding-style cleanups] [jiebin.sun@intel.com: avoid negative value by overflow in msginfo] Link: https://lkml.kernel.org/r/20220920150809.4014944-1-jiebin.sun@intel.com [akpm@linux-foundation.org: fix min() warnings] Link: https://lkml.kernel.org/r/20220913192538.3023708-3-jiebin.sun@intel.com Signed-off-by: Jiebin Sun <jiebin.sun@intel.com> Reviewed-by: Tim Chen <tim.c.chen@linux.intel.com> Cc: Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com> Cc: Alexey Gladkov <legion@kernel.org> Cc: Christoph Lameter <cl@linux.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Dennis Zhou <dennis@kernel.org> Cc: "Eric W . Biederman" <ebiederm@xmission.com> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Tejun Heo <tj@kernel.org> Cc: Vasily Averin <vasily.averin@linux.dev> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> diff 72d1e611 Tue Sep 13 13:25:38 MDT 2022 Jiebin Sun <jiebin.sun@intel.com> ipc/msg: mitigate the lock contention with percpu counter The msg_bytes and msg_hdrs atomic counters are frequently updated when IPC msg queue is in heavy use, causing heavy cache bounce and overhead. Change them to percpu_counter greatly improve the performance. Since there is one percpu struct per namespace, additional memory cost is minimal. Reading of the count done in msgctl call, which is infrequent. So the need to sum up the counts in each CPU is infrequent. Apply the patch and test the pts/stress-ng-1.4.0 -- system v message passing (160 threads). Score gain: 3.99x CPU: ICX 8380 x 2 sockets Core number: 40 x 2 physical cores Benchmark: pts/stress-ng-1.4.0 -- system v message passing (160 threads) [akpm@linux-foundation.org: coding-style cleanups] [jiebin.sun@intel.com: avoid negative value by overflow in msginfo] Link: https://lkml.kernel.org/r/20220920150809.4014944-1-jiebin.sun@intel.com [akpm@linux-foundation.org: fix min() warnings] Link: https://lkml.kernel.org/r/20220913192538.3023708-3-jiebin.sun@intel.com Signed-off-by: Jiebin Sun <jiebin.sun@intel.com> Reviewed-by: Tim Chen <tim.c.chen@linux.intel.com> Cc: Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com> Cc: Alexey Gladkov <legion@kernel.org> Cc: Christoph Lameter <cl@linux.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Dennis Zhou <dennis@kernel.org> Cc: "Eric W . Biederman" <ebiederm@xmission.com> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Tejun Heo <tj@kernel.org> Cc: Vasily Averin <vasily.averin@linux.dev> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> diff b869d5be Wed Jun 30 19:57:18 MDT 2021 Manfred Spraul <manfred@colorfullife.com> ipc/util.c: use binary search for max_idx If semctl(), msgctl() and shmctl() are called with IPC_INFO, SEM_INFO, MSG_INFO or SHM_INFO, then the return value is the index of the highest used index in the kernel's internal array recording information about all SysV objects of the requested type for the current namespace. (This information can be used with repeated ..._STAT or ..._STAT_ANY operations to obtain information about all SysV objects on the system.) There is a cache for this value. But when the cache needs up be updated, then the highest used index is determined by looping over all possible values. With the introduction of IPCMNI_EXTEND_SHIFT, this could be a loop over 16 million entries. And due to /proc/sys/kernel/*next_id, the index values do not need to be consecutive. With <write 16000000 to msg_next_id>, msgget(), msgctl(,IPC_RMID) in a loop, I have observed a performance increase of around factor 13000. As there is no get_last() function for idr structures: Implement a "get_last()" using a binary search. As far as I see, ipc is the only user that needs get_last(), thus implement it in ipc/util.c and not in a central location. [akpm@linux-foundation.org: tweak comment, fix typo] Link: https://lkml.kernel.org/r/20210425075208.11777-2-manfred@colorfullife.com Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Acked-by: Davidlohr Bueso <dbueso@suse.de> Cc: <1vier1@web.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 99db46ea Tue May 14 16:46:36 MDT 2019 Manfred Spraul <manfred@colorfullife.com> ipc: do cyclic id allocation for the ipc object. For ipcmni_extend mode, the sequence number space is only 7 bits. So the chance of id reuse is relatively high compared with the non-extended mode. To alleviate this id reuse problem, this patch enables cyclic allocation for the index to the radix tree (idx). The disadvantage is that this can cause a slight slow-down of the fast path, as the radix tree could be higher than necessary. To limit the radix tree height, I have chosen the following limits: 1) The cycling is done over in_use*1.5. 2) At least, the cycling is done over "normal" ipcnmi mode: RADIX_TREE_MAP_SIZE elements "ipcmni_extended": 4096 elements Result: - for normal mode: No change for <= 42 active ipc elements. With more than 42 active ipc elements, a 2nd level would be added to the radix tree. Without cyclic allocation, a 2nd level would be added only with more than 63 active elements. - for extended mode: Cycling creates always at least a 2-level radix tree. With more than 2730 active objects, a 3rd level would be added, instead of > 4095 active objects until the 3rd level is added without cyclic allocation. For a 2-level radix tree compared to a 1-level radix tree, I have observed < 1% performance impact. Notes: 1) Normal "x=semget();y=semget();" is unaffected: Then the idx is e.g. a and a+1, regardless if idr_alloc() or idr_alloc_cyclic() is used. 2) The -1% happens in a microbenchmark after this situation: x=semget(); for(i=0;i<4000;i++) {t=semget();semctl(t,0,IPC_RMID);} y=semget(); Now perform semget calls on x and y that do not sleep. 3) The worst-case reuse cycle time is unfortunately unaffected: If you have 2^24-1 ipc objects allocated, and get/remove the last possible element in a loop, then the id is reused after 128 get/remove pairs. Performance check: A microbenchmark that performes no-op semop() randomly on two IDs, with only these two IDs allocated. The IDs were set using /proc/sys/kernel/sem_next_id. The test was run 5 times, averages are shown. 1 & 2: Base (6.22 seconds for 10.000.000 semops) 1 & 40: -0.2% 1 & 3348: - 0.8% 1 & 27348: - 1.6% 1 & 15777204: - 3.2% Or: ~12.6 cpu cycles per additional radix tree level. The cpu is an Intel I3-5010U. ~1300 cpu cycles/syscall is slower than what I remember (spectre impact?). V2 of the patch: - use "min" and "max" - use RADIX_TREE_MAP_SIZE * RADIX_TREE_MAP_SIZE instead of (2<<12). [akpm@linux-foundation.org: fix max() warning] Link: http://lkml.kernel.org/r/20190329204930.21620-3-longman@redhat.com Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Acked-by: Waiman Long <longman@redhat.com> Cc: "Luis R. Rodriguez" <mcgrof@kernel.org> Cc: Kees Cook <keescook@chromium.org> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Matthew Wilcox <willy@infradead.org> Cc: "Eric W . Biederman" <ebiederm@xmission.com> Cc: Takashi Iwai <tiwai@suse.de> Cc: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 99db46ea Tue May 14 16:46:36 MDT 2019 Manfred Spraul <manfred@colorfullife.com> ipc: do cyclic id allocation for the ipc object. For ipcmni_extend mode, the sequence number space is only 7 bits. So the chance of id reuse is relatively high compared with the non-extended mode. To alleviate this id reuse problem, this patch enables cyclic allocation for the index to the radix tree (idx). The disadvantage is that this can cause a slight slow-down of the fast path, as the radix tree could be higher than necessary. To limit the radix tree height, I have chosen the following limits: 1) The cycling is done over in_use*1.5. 2) At least, the cycling is done over "normal" ipcnmi mode: RADIX_TREE_MAP_SIZE elements "ipcmni_extended": 4096 elements Result: - for normal mode: No change for <= 42 active ipc elements. With more than 42 active ipc elements, a 2nd level would be added to the radix tree. Without cyclic allocation, a 2nd level would be added only with more than 63 active elements. - for extended mode: Cycling creates always at least a 2-level radix tree. With more than 2730 active objects, a 3rd level would be added, instead of > 4095 active objects until the 3rd level is added without cyclic allocation. For a 2-level radix tree compared to a 1-level radix tree, I have observed < 1% performance impact. Notes: 1) Normal "x=semget();y=semget();" is unaffected: Then the idx is e.g. a and a+1, regardless if idr_alloc() or idr_alloc_cyclic() is used. 2) The -1% happens in a microbenchmark after this situation: x=semget(); for(i=0;i<4000;i++) {t=semget();semctl(t,0,IPC_RMID);} y=semget(); Now perform semget calls on x and y that do not sleep. 3) The worst-case reuse cycle time is unfortunately unaffected: If you have 2^24-1 ipc objects allocated, and get/remove the last possible element in a loop, then the id is reused after 128 get/remove pairs. Performance check: A microbenchmark that performes no-op semop() randomly on two IDs, with only these two IDs allocated. The IDs were set using /proc/sys/kernel/sem_next_id. The test was run 5 times, averages are shown. 1 & 2: Base (6.22 seconds for 10.000.000 semops) 1 & 40: -0.2% 1 & 3348: - 0.8% 1 & 27348: - 1.6% 1 & 15777204: - 3.2% Or: ~12.6 cpu cycles per additional radix tree level. The cpu is an Intel I3-5010U. ~1300 cpu cycles/syscall is slower than what I remember (spectre impact?). V2 of the patch: - use "min" and "max" - use RADIX_TREE_MAP_SIZE * RADIX_TREE_MAP_SIZE instead of (2<<12). [akpm@linux-foundation.org: fix max() warning] Link: http://lkml.kernel.org/r/20190329204930.21620-3-longman@redhat.com Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Acked-by: Waiman Long <longman@redhat.com> Cc: "Luis R. Rodriguez" <mcgrof@kernel.org> Cc: Kees Cook <keescook@chromium.org> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Matthew Wilcox <willy@infradead.org> Cc: "Eric W . Biederman" <ebiederm@xmission.com> Cc: Takashi Iwai <tiwai@suse.de> Cc: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 99db46ea Tue May 14 16:46:36 MDT 2019 Manfred Spraul <manfred@colorfullife.com> ipc: do cyclic id allocation for the ipc object. For ipcmni_extend mode, the sequence number space is only 7 bits. So the chance of id reuse is relatively high compared with the non-extended mode. To alleviate this id reuse problem, this patch enables cyclic allocation for the index to the radix tree (idx). The disadvantage is that this can cause a slight slow-down of the fast path, as the radix tree could be higher than necessary. To limit the radix tree height, I have chosen the following limits: 1) The cycling is done over in_use*1.5. 2) At least, the cycling is done over "normal" ipcnmi mode: RADIX_TREE_MAP_SIZE elements "ipcmni_extended": 4096 elements Result: - for normal mode: No change for <= 42 active ipc elements. With more than 42 active ipc elements, a 2nd level would be added to the radix tree. Without cyclic allocation, a 2nd level would be added only with more than 63 active elements. - for extended mode: Cycling creates always at least a 2-level radix tree. With more than 2730 active objects, a 3rd level would be added, instead of > 4095 active objects until the 3rd level is added without cyclic allocation. For a 2-level radix tree compared to a 1-level radix tree, I have observed < 1% performance impact. Notes: 1) Normal "x=semget();y=semget();" is unaffected: Then the idx is e.g. a and a+1, regardless if idr_alloc() or idr_alloc_cyclic() is used. 2) The -1% happens in a microbenchmark after this situation: x=semget(); for(i=0;i<4000;i++) {t=semget();semctl(t,0,IPC_RMID);} y=semget(); Now perform semget calls on x and y that do not sleep. 3) The worst-case reuse cycle time is unfortunately unaffected: If you have 2^24-1 ipc objects allocated, and get/remove the last possible element in a loop, then the id is reused after 128 get/remove pairs. Performance check: A microbenchmark that performes no-op semop() randomly on two IDs, with only these two IDs allocated. The IDs were set using /proc/sys/kernel/sem_next_id. The test was run 5 times, averages are shown. 1 & 2: Base (6.22 seconds for 10.000.000 semops) 1 & 40: -0.2% 1 & 3348: - 0.8% 1 & 27348: - 1.6% 1 & 15777204: - 3.2% Or: ~12.6 cpu cycles per additional radix tree level. The cpu is an Intel I3-5010U. ~1300 cpu cycles/syscall is slower than what I remember (spectre impact?). V2 of the patch: - use "min" and "max" - use RADIX_TREE_MAP_SIZE * RADIX_TREE_MAP_SIZE instead of (2<<12). [akpm@linux-foundation.org: fix max() warning] Link: http://lkml.kernel.org/r/20190329204930.21620-3-longman@redhat.com Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Acked-by: Waiman Long <longman@redhat.com> Cc: "Luis R. Rodriguez" <mcgrof@kernel.org> Cc: Kees Cook <keescook@chromium.org> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Matthew Wilcox <willy@infradead.org> Cc: "Eric W . Biederman" <ebiederm@xmission.com> Cc: Takashi Iwai <tiwai@suse.de> Cc: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 99db46ea Tue May 14 16:46:36 MDT 2019 Manfred Spraul <manfred@colorfullife.com> ipc: do cyclic id allocation for the ipc object. For ipcmni_extend mode, the sequence number space is only 7 bits. So the chance of id reuse is relatively high compared with the non-extended mode. To alleviate this id reuse problem, this patch enables cyclic allocation for the index to the radix tree (idx). The disadvantage is that this can cause a slight slow-down of the fast path, as the radix tree could be higher than necessary. To limit the radix tree height, I have chosen the following limits: 1) The cycling is done over in_use*1.5. 2) At least, the cycling is done over "normal" ipcnmi mode: RADIX_TREE_MAP_SIZE elements "ipcmni_extended": 4096 elements Result: - for normal mode: No change for <= 42 active ipc elements. With more than 42 active ipc elements, a 2nd level would be added to the radix tree. Without cyclic allocation, a 2nd level would be added only with more than 63 active elements. - for extended mode: Cycling creates always at least a 2-level radix tree. With more than 2730 active objects, a 3rd level would be added, instead of > 4095 active objects until the 3rd level is added without cyclic allocation. For a 2-level radix tree compared to a 1-level radix tree, I have observed < 1% performance impact. Notes: 1) Normal "x=semget();y=semget();" is unaffected: Then the idx is e.g. a and a+1, regardless if idr_alloc() or idr_alloc_cyclic() is used. 2) The -1% happens in a microbenchmark after this situation: x=semget(); for(i=0;i<4000;i++) {t=semget();semctl(t,0,IPC_RMID);} y=semget(); Now perform semget calls on x and y that do not sleep. 3) The worst-case reuse cycle time is unfortunately unaffected: If you have 2^24-1 ipc objects allocated, and get/remove the last possible element in a loop, then the id is reused after 128 get/remove pairs. Performance check: A microbenchmark that performes no-op semop() randomly on two IDs, with only these two IDs allocated. The IDs were set using /proc/sys/kernel/sem_next_id. The test was run 5 times, averages are shown. 1 & 2: Base (6.22 seconds for 10.000.000 semops) 1 & 40: -0.2% 1 & 3348: - 0.8% 1 & 27348: - 1.6% 1 & 15777204: - 3.2% Or: ~12.6 cpu cycles per additional radix tree level. The cpu is an Intel I3-5010U. ~1300 cpu cycles/syscall is slower than what I remember (spectre impact?). V2 of the patch: - use "min" and "max" - use RADIX_TREE_MAP_SIZE * RADIX_TREE_MAP_SIZE instead of (2<<12). [akpm@linux-foundation.org: fix max() warning] Link: http://lkml.kernel.org/r/20190329204930.21620-3-longman@redhat.com Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Acked-by: Waiman Long <longman@redhat.com> Cc: "Luis R. Rodriguez" <mcgrof@kernel.org> Cc: Kees Cook <keescook@chromium.org> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Matthew Wilcox <willy@infradead.org> Cc: "Eric W . Biederman" <ebiederm@xmission.com> Cc: Takashi Iwai <tiwai@suse.de> Cc: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 99db46ea Tue May 14 16:46:36 MDT 2019 Manfred Spraul <manfred@colorfullife.com> ipc: do cyclic id allocation for the ipc object. For ipcmni_extend mode, the sequence number space is only 7 bits. So the chance of id reuse is relatively high compared with the non-extended mode. To alleviate this id reuse problem, this patch enables cyclic allocation for the index to the radix tree (idx). The disadvantage is that this can cause a slight slow-down of the fast path, as the radix tree could be higher than necessary. To limit the radix tree height, I have chosen the following limits: 1) The cycling is done over in_use*1.5. 2) At least, the cycling is done over "normal" ipcnmi mode: RADIX_TREE_MAP_SIZE elements "ipcmni_extended": 4096 elements Result: - for normal mode: No change for <= 42 active ipc elements. With more than 42 active ipc elements, a 2nd level would be added to the radix tree. Without cyclic allocation, a 2nd level would be added only with more than 63 active elements. - for extended mode: Cycling creates always at least a 2-level radix tree. With more than 2730 active objects, a 3rd level would be added, instead of > 4095 active objects until the 3rd level is added without cyclic allocation. For a 2-level radix tree compared to a 1-level radix tree, I have observed < 1% performance impact. Notes: 1) Normal "x=semget();y=semget();" is unaffected: Then the idx is e.g. a and a+1, regardless if idr_alloc() or idr_alloc_cyclic() is used. 2) The -1% happens in a microbenchmark after this situation: x=semget(); for(i=0;i<4000;i++) {t=semget();semctl(t,0,IPC_RMID);} y=semget(); Now perform semget calls on x and y that do not sleep. 3) The worst-case reuse cycle time is unfortunately unaffected: If you have 2^24-1 ipc objects allocated, and get/remove the last possible element in a loop, then the id is reused after 128 get/remove pairs. Performance check: A microbenchmark that performes no-op semop() randomly on two IDs, with only these two IDs allocated. The IDs were set using /proc/sys/kernel/sem_next_id. The test was run 5 times, averages are shown. 1 & 2: Base (6.22 seconds for 10.000.000 semops) 1 & 40: -0.2% 1 & 3348: - 0.8% 1 & 27348: - 1.6% 1 & 15777204: - 3.2% Or: ~12.6 cpu cycles per additional radix tree level. The cpu is an Intel I3-5010U. ~1300 cpu cycles/syscall is slower than what I remember (spectre impact?). V2 of the patch: - use "min" and "max" - use RADIX_TREE_MAP_SIZE * RADIX_TREE_MAP_SIZE instead of (2<<12). [akpm@linux-foundation.org: fix max() warning] Link: http://lkml.kernel.org/r/20190329204930.21620-3-longman@redhat.com Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Acked-by: Waiman Long <longman@redhat.com> Cc: "Luis R. Rodriguez" <mcgrof@kernel.org> Cc: Kees Cook <keescook@chromium.org> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Matthew Wilcox <willy@infradead.org> Cc: "Eric W . Biederman" <ebiederm@xmission.com> Cc: Takashi Iwai <tiwai@suse.de> Cc: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 99db46ea Tue May 14 16:46:36 MDT 2019 Manfred Spraul <manfred@colorfullife.com> ipc: do cyclic id allocation for the ipc object. For ipcmni_extend mode, the sequence number space is only 7 bits. So the chance of id reuse is relatively high compared with the non-extended mode. To alleviate this id reuse problem, this patch enables cyclic allocation for the index to the radix tree (idx). The disadvantage is that this can cause a slight slow-down of the fast path, as the radix tree could be higher than necessary. To limit the radix tree height, I have chosen the following limits: 1) The cycling is done over in_use*1.5. 2) At least, the cycling is done over "normal" ipcnmi mode: RADIX_TREE_MAP_SIZE elements "ipcmni_extended": 4096 elements Result: - for normal mode: No change for <= 42 active ipc elements. With more than 42 active ipc elements, a 2nd level would be added to the radix tree. Without cyclic allocation, a 2nd level would be added only with more than 63 active elements. - for extended mode: Cycling creates always at least a 2-level radix tree. With more than 2730 active objects, a 3rd level would be added, instead of > 4095 active objects until the 3rd level is added without cyclic allocation. For a 2-level radix tree compared to a 1-level radix tree, I have observed < 1% performance impact. Notes: 1) Normal "x=semget();y=semget();" is unaffected: Then the idx is e.g. a and a+1, regardless if idr_alloc() or idr_alloc_cyclic() is used. 2) The -1% happens in a microbenchmark after this situation: x=semget(); for(i=0;i<4000;i++) {t=semget();semctl(t,0,IPC_RMID);} y=semget(); Now perform semget calls on x and y that do not sleep. 3) The worst-case reuse cycle time is unfortunately unaffected: If you have 2^24-1 ipc objects allocated, and get/remove the last possible element in a loop, then the id is reused after 128 get/remove pairs. Performance check: A microbenchmark that performes no-op semop() randomly on two IDs, with only these two IDs allocated. The IDs were set using /proc/sys/kernel/sem_next_id. The test was run 5 times, averages are shown. 1 & 2: Base (6.22 seconds for 10.000.000 semops) 1 & 40: -0.2% 1 & 3348: - 0.8% 1 & 27348: - 1.6% 1 & 15777204: - 3.2% Or: ~12.6 cpu cycles per additional radix tree level. The cpu is an Intel I3-5010U. ~1300 cpu cycles/syscall is slower than what I remember (spectre impact?). V2 of the patch: - use "min" and "max" - use RADIX_TREE_MAP_SIZE * RADIX_TREE_MAP_SIZE instead of (2<<12). [akpm@linux-foundation.org: fix max() warning] Link: http://lkml.kernel.org/r/20190329204930.21620-3-longman@redhat.com Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Acked-by: Waiman Long <longman@redhat.com> Cc: "Luis R. Rodriguez" <mcgrof@kernel.org> Cc: Kees Cook <keescook@chromium.org> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Matthew Wilcox <willy@infradead.org> Cc: "Eric W . Biederman" <ebiederm@xmission.com> Cc: Takashi Iwai <tiwai@suse.de> Cc: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 99db46ea Tue May 14 16:46:36 MDT 2019 Manfred Spraul <manfred@colorfullife.com> ipc: do cyclic id allocation for the ipc object. For ipcmni_extend mode, the sequence number space is only 7 bits. So the chance of id reuse is relatively high compared with the non-extended mode. To alleviate this id reuse problem, this patch enables cyclic allocation for the index to the radix tree (idx). The disadvantage is that this can cause a slight slow-down of the fast path, as the radix tree could be higher than necessary. To limit the radix tree height, I have chosen the following limits: 1) The cycling is done over in_use*1.5. 2) At least, the cycling is done over "normal" ipcnmi mode: RADIX_TREE_MAP_SIZE elements "ipcmni_extended": 4096 elements Result: - for normal mode: No change for <= 42 active ipc elements. With more than 42 active ipc elements, a 2nd level would be added to the radix tree. Without cyclic allocation, a 2nd level would be added only with more than 63 active elements. - for extended mode: Cycling creates always at least a 2-level radix tree. With more than 2730 active objects, a 3rd level would be added, instead of > 4095 active objects until the 3rd level is added without cyclic allocation. For a 2-level radix tree compared to a 1-level radix tree, I have observed < 1% performance impact. Notes: 1) Normal "x=semget();y=semget();" is unaffected: Then the idx is e.g. a and a+1, regardless if idr_alloc() or idr_alloc_cyclic() is used. 2) The -1% happens in a microbenchmark after this situation: x=semget(); for(i=0;i<4000;i++) {t=semget();semctl(t,0,IPC_RMID);} y=semget(); Now perform semget calls on x and y that do not sleep. 3) The worst-case reuse cycle time is unfortunately unaffected: If you have 2^24-1 ipc objects allocated, and get/remove the last possible element in a loop, then the id is reused after 128 get/remove pairs. Performance check: A microbenchmark that performes no-op semop() randomly on two IDs, with only these two IDs allocated. The IDs were set using /proc/sys/kernel/sem_next_id. The test was run 5 times, averages are shown. 1 & 2: Base (6.22 seconds for 10.000.000 semops) 1 & 40: -0.2% 1 & 3348: - 0.8% 1 & 27348: - 1.6% 1 & 15777204: - 3.2% Or: ~12.6 cpu cycles per additional radix tree level. The cpu is an Intel I3-5010U. ~1300 cpu cycles/syscall is slower than what I remember (spectre impact?). V2 of the patch: - use "min" and "max" - use RADIX_TREE_MAP_SIZE * RADIX_TREE_MAP_SIZE instead of (2<<12). [akpm@linux-foundation.org: fix max() warning] Link: http://lkml.kernel.org/r/20190329204930.21620-3-longman@redhat.com Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Acked-by: Waiman Long <longman@redhat.com> Cc: "Luis R. Rodriguez" <mcgrof@kernel.org> Cc: Kees Cook <keescook@chromium.org> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Matthew Wilcox <willy@infradead.org> Cc: "Eric W . Biederman" <ebiederm@xmission.com> Cc: Takashi Iwai <tiwai@suse.de> Cc: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 99db46ea Tue May 14 16:46:36 MDT 2019 Manfred Spraul <manfred@colorfullife.com> ipc: do cyclic id allocation for the ipc object. For ipcmni_extend mode, the sequence number space is only 7 bits. So the chance of id reuse is relatively high compared with the non-extended mode. To alleviate this id reuse problem, this patch enables cyclic allocation for the index to the radix tree (idx). The disadvantage is that this can cause a slight slow-down of the fast path, as the radix tree could be higher than necessary. To limit the radix tree height, I have chosen the following limits: 1) The cycling is done over in_use*1.5. 2) At least, the cycling is done over "normal" ipcnmi mode: RADIX_TREE_MAP_SIZE elements "ipcmni_extended": 4096 elements Result: - for normal mode: No change for <= 42 active ipc elements. With more than 42 active ipc elements, a 2nd level would be added to the radix tree. Without cyclic allocation, a 2nd level would be added only with more than 63 active elements. - for extended mode: Cycling creates always at least a 2-level radix tree. With more than 2730 active objects, a 3rd level would be added, instead of > 4095 active objects until the 3rd level is added without cyclic allocation. For a 2-level radix tree compared to a 1-level radix tree, I have observed < 1% performance impact. Notes: 1) Normal "x=semget();y=semget();" is unaffected: Then the idx is e.g. a and a+1, regardless if idr_alloc() or idr_alloc_cyclic() is used. 2) The -1% happens in a microbenchmark after this situation: x=semget(); for(i=0;i<4000;i++) {t=semget();semctl(t,0,IPC_RMID);} y=semget(); Now perform semget calls on x and y that do not sleep. 3) The worst-case reuse cycle time is unfortunately unaffected: If you have 2^24-1 ipc objects allocated, and get/remove the last possible element in a loop, then the id is reused after 128 get/remove pairs. Performance check: A microbenchmark that performes no-op semop() randomly on two IDs, with only these two IDs allocated. The IDs were set using /proc/sys/kernel/sem_next_id. The test was run 5 times, averages are shown. 1 & 2: Base (6.22 seconds for 10.000.000 semops) 1 & 40: -0.2% 1 & 3348: - 0.8% 1 & 27348: - 1.6% 1 & 15777204: - 3.2% Or: ~12.6 cpu cycles per additional radix tree level. The cpu is an Intel I3-5010U. ~1300 cpu cycles/syscall is slower than what I remember (spectre impact?). V2 of the patch: - use "min" and "max" - use RADIX_TREE_MAP_SIZE * RADIX_TREE_MAP_SIZE instead of (2<<12). [akpm@linux-foundation.org: fix max() warning] Link: http://lkml.kernel.org/r/20190329204930.21620-3-longman@redhat.com Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Acked-by: Waiman Long <longman@redhat.com> Cc: "Luis R. Rodriguez" <mcgrof@kernel.org> Cc: Kees Cook <keescook@chromium.org> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Matthew Wilcox <willy@infradead.org> Cc: "Eric W . Biederman" <ebiederm@xmission.com> Cc: Takashi Iwai <tiwai@suse.de> Cc: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 99db46ea Tue May 14 16:46:36 MDT 2019 Manfred Spraul <manfred@colorfullife.com> ipc: do cyclic id allocation for the ipc object. For ipcmni_extend mode, the sequence number space is only 7 bits. So the chance of id reuse is relatively high compared with the non-extended mode. To alleviate this id reuse problem, this patch enables cyclic allocation for the index to the radix tree (idx). The disadvantage is that this can cause a slight slow-down of the fast path, as the radix tree could be higher than necessary. To limit the radix tree height, I have chosen the following limits: 1) The cycling is done over in_use*1.5. 2) At least, the cycling is done over "normal" ipcnmi mode: RADIX_TREE_MAP_SIZE elements "ipcmni_extended": 4096 elements Result: - for normal mode: No change for <= 42 active ipc elements. With more than 42 active ipc elements, a 2nd level would be added to the radix tree. Without cyclic allocation, a 2nd level would be added only with more than 63 active elements. - for extended mode: Cycling creates always at least a 2-level radix tree. With more than 2730 active objects, a 3rd level would be added, instead of > 4095 active objects until the 3rd level is added without cyclic allocation. For a 2-level radix tree compared to a 1-level radix tree, I have observed < 1% performance impact. Notes: 1) Normal "x=semget();y=semget();" is unaffected: Then the idx is e.g. a and a+1, regardless if idr_alloc() or idr_alloc_cyclic() is used. 2) The -1% happens in a microbenchmark after this situation: x=semget(); for(i=0;i<4000;i++) {t=semget();semctl(t,0,IPC_RMID);} y=semget(); Now perform semget calls on x and y that do not sleep. 3) The worst-case reuse cycle time is unfortunately unaffected: If you have 2^24-1 ipc objects allocated, and get/remove the last possible element in a loop, then the id is reused after 128 get/remove pairs. Performance check: A microbenchmark that performes no-op semop() randomly on two IDs, with only these two IDs allocated. The IDs were set using /proc/sys/kernel/sem_next_id. The test was run 5 times, averages are shown. 1 & 2: Base (6.22 seconds for 10.000.000 semops) 1 & 40: -0.2% 1 & 3348: - 0.8% 1 & 27348: - 1.6% 1 & 15777204: - 3.2% Or: ~12.6 cpu cycles per additional radix tree level. The cpu is an Intel I3-5010U. ~1300 cpu cycles/syscall is slower than what I remember (spectre impact?). V2 of the patch: - use "min" and "max" - use RADIX_TREE_MAP_SIZE * RADIX_TREE_MAP_SIZE instead of (2<<12). [akpm@linux-foundation.org: fix max() warning] Link: http://lkml.kernel.org/r/20190329204930.21620-3-longman@redhat.com Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Acked-by: Waiman Long <longman@redhat.com> Cc: "Luis R. Rodriguez" <mcgrof@kernel.org> Cc: Kees Cook <keescook@chromium.org> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Matthew Wilcox <willy@infradead.org> Cc: "Eric W . Biederman" <ebiederm@xmission.com> Cc: Takashi Iwai <tiwai@suse.de> Cc: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> |
| H A D | util.c | diff 359745d7 Fri Jan 21 23:14:23 MST 2022 Muchun Song <songmuchun@bytedance.com> proc: remove PDE_DATA() completely Remove PDE_DATA() completely and replace it with pde_data(). [akpm@linux-foundation.org: fix naming clash in drivers/nubus/proc.c] [akpm@linux-foundation.org: now fix it properly] Link: https://lkml.kernel.org/r/20211124081956.87711-2-songmuchun@bytedance.com Signed-off-by: Muchun Song <songmuchun@bytedance.com> Acked-by: Christian Brauner <christian.brauner@ubuntu.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: Alexey Gladkov <gladkov.alexey@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 126e8bee Fri Nov 19 17:43:18 MST 2021 Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com> ipc: WARN if trying to remove ipc object which is absent Patch series "shm: shm_rmid_forced feature fixes". Some time ago I met kernel crash after CRIU restore procedure, fortunately, it was CRIU restore, so, I had dump files and could do restore many times and crash reproduced easily. After some investigation I've constructed the minimal reproducer. It was found that it's use-after-free and it happens only if sysctl kernel.shm_rmid_forced = 1. The key of the problem is that the exit_shm() function not handles shp's object destroy when task->sysvshm.shm_clist contains items from different IPC namespaces. In most cases this list will contain only items from one IPC namespace. How can this list contain object from different namespaces? The exit_shm() function is designed to clean up this list always when process leaves IPC namespace. But we made a mistake a long time ago and did not add a exit_shm() call into the setns() syscall procedures. The first idea was just to add this call to setns() syscall but it obviously changes semantics of setns() syscall and that's userspace-visible change. So, I gave up on this idea. The first real attempt to address the issue was just to omit forced destroy if we meet shp object not from current task IPC namespace [1]. But that was not the best idea because task->sysvshm.shm_clist was protected by rwsem which belongs to current task IPC namespace. It means that list corruption may occur. Second approach is just extend exit_shm() to properly handle shp's from different IPC namespaces [2]. This is really non-trivial thing, I've put a lot of effort into that but not believed that it's possible to make it fully safe, clean and clear. Thanks to the efforts of Manfred Spraul working an elegant solution was designed. Thanks a lot, Manfred! Eric also suggested the way to address the issue in ("[RFC][PATCH] shm: In shm_exit destroy all created and never attached segments") Eric's idea was to maintain a list of shm_clists one per IPC namespace, use lock-less lists. But there is some extra memory consumption-related concerns. An alternative solution which was suggested by me was implemented in ("shm: reset shm_clist on setns but omit forced shm destroy"). The idea is pretty simple, we add exit_shm() syscall to setns() but DO NOT destroy shm segments even if sysctl kernel.shm_rmid_forced = 1, we just clean up the task->sysvshm.shm_clist list. This chages semantics of setns() syscall a little bit but in comparision to the "naive" solution when we just add exit_shm() without any special exclusions this looks like a safer option. [1] https://lkml.org/lkml/2021/7/6/1108 [2] https://lkml.org/lkml/2021/7/14/736 This patch (of 2): Let's produce a warning if we trying to remove non-existing IPC object from IPC namespace kht/idr structures. This allows us to catch possible bugs when the ipc_rmid() function was called with inconsistent struct ipc_ids*, struct kern_ipc_perm* arguments. Link: https://lkml.kernel.org/r/20211027224348.611025-1-alexander.mikhalitsyn@virtuozzo.com Link: https://lkml.kernel.org/r/20211027224348.611025-2-alexander.mikhalitsyn@virtuozzo.com Co-developed-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Andrei Vagin <avagin@gmail.com> Cc: Pavel Tikhomirov <ptikhomirov@virtuozzo.com> Cc: Vasily Averin <vvs@virtuozzo.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 126e8bee Fri Nov 19 17:43:18 MST 2021 Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com> ipc: WARN if trying to remove ipc object which is absent Patch series "shm: shm_rmid_forced feature fixes". Some time ago I met kernel crash after CRIU restore procedure, fortunately, it was CRIU restore, so, I had dump files and could do restore many times and crash reproduced easily. After some investigation I've constructed the minimal reproducer. It was found that it's use-after-free and it happens only if sysctl kernel.shm_rmid_forced = 1. The key of the problem is that the exit_shm() function not handles shp's object destroy when task->sysvshm.shm_clist contains items from different IPC namespaces. In most cases this list will contain only items from one IPC namespace. How can this list contain object from different namespaces? The exit_shm() function is designed to clean up this list always when process leaves IPC namespace. But we made a mistake a long time ago and did not add a exit_shm() call into the setns() syscall procedures. The first idea was just to add this call to setns() syscall but it obviously changes semantics of setns() syscall and that's userspace-visible change. So, I gave up on this idea. The first real attempt to address the issue was just to omit forced destroy if we meet shp object not from current task IPC namespace [1]. But that was not the best idea because task->sysvshm.shm_clist was protected by rwsem which belongs to current task IPC namespace. It means that list corruption may occur. Second approach is just extend exit_shm() to properly handle shp's from different IPC namespaces [2]. This is really non-trivial thing, I've put a lot of effort into that but not believed that it's possible to make it fully safe, clean and clear. Thanks to the efforts of Manfred Spraul working an elegant solution was designed. Thanks a lot, Manfred! Eric also suggested the way to address the issue in ("[RFC][PATCH] shm: In shm_exit destroy all created and never attached segments") Eric's idea was to maintain a list of shm_clists one per IPC namespace, use lock-less lists. But there is some extra memory consumption-related concerns. An alternative solution which was suggested by me was implemented in ("shm: reset shm_clist on setns but omit forced shm destroy"). The idea is pretty simple, we add exit_shm() syscall to setns() but DO NOT destroy shm segments even if sysctl kernel.shm_rmid_forced = 1, we just clean up the task->sysvshm.shm_clist list. This chages semantics of setns() syscall a little bit but in comparision to the "naive" solution when we just add exit_shm() without any special exclusions this looks like a safer option. [1] https://lkml.org/lkml/2021/7/6/1108 [2] https://lkml.org/lkml/2021/7/14/736 This patch (of 2): Let's produce a warning if we trying to remove non-existing IPC object from IPC namespace kht/idr structures. This allows us to catch possible bugs when the ipc_rmid() function was called with inconsistent struct ipc_ids*, struct kern_ipc_perm* arguments. Link: https://lkml.kernel.org/r/20211027224348.611025-1-alexander.mikhalitsyn@virtuozzo.com Link: https://lkml.kernel.org/r/20211027224348.611025-2-alexander.mikhalitsyn@virtuozzo.com Co-developed-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Andrei Vagin <avagin@gmail.com> Cc: Pavel Tikhomirov <ptikhomirov@virtuozzo.com> Cc: Vasily Averin <vvs@virtuozzo.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 126e8bee Fri Nov 19 17:43:18 MST 2021 Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com> ipc: WARN if trying to remove ipc object which is absent Patch series "shm: shm_rmid_forced feature fixes". Some time ago I met kernel crash after CRIU restore procedure, fortunately, it was CRIU restore, so, I had dump files and could do restore many times and crash reproduced easily. After some investigation I've constructed the minimal reproducer. It was found that it's use-after-free and it happens only if sysctl kernel.shm_rmid_forced = 1. The key of the problem is that the exit_shm() function not handles shp's object destroy when task->sysvshm.shm_clist contains items from different IPC namespaces. In most cases this list will contain only items from one IPC namespace. How can this list contain object from different namespaces? The exit_shm() function is designed to clean up this list always when process leaves IPC namespace. But we made a mistake a long time ago and did not add a exit_shm() call into the setns() syscall procedures. The first idea was just to add this call to setns() syscall but it obviously changes semantics of setns() syscall and that's userspace-visible change. So, I gave up on this idea. The first real attempt to address the issue was just to omit forced destroy if we meet shp object not from current task IPC namespace [1]. But that was not the best idea because task->sysvshm.shm_clist was protected by rwsem which belongs to current task IPC namespace. It means that list corruption may occur. Second approach is just extend exit_shm() to properly handle shp's from different IPC namespaces [2]. This is really non-trivial thing, I've put a lot of effort into that but not believed that it's possible to make it fully safe, clean and clear. Thanks to the efforts of Manfred Spraul working an elegant solution was designed. Thanks a lot, Manfred! Eric also suggested the way to address the issue in ("[RFC][PATCH] shm: In shm_exit destroy all created and never attached segments") Eric's idea was to maintain a list of shm_clists one per IPC namespace, use lock-less lists. But there is some extra memory consumption-related concerns. An alternative solution which was suggested by me was implemented in ("shm: reset shm_clist on setns but omit forced shm destroy"). The idea is pretty simple, we add exit_shm() syscall to setns() but DO NOT destroy shm segments even if sysctl kernel.shm_rmid_forced = 1, we just clean up the task->sysvshm.shm_clist list. This chages semantics of setns() syscall a little bit but in comparision to the "naive" solution when we just add exit_shm() without any special exclusions this looks like a safer option. [1] https://lkml.org/lkml/2021/7/6/1108 [2] https://lkml.org/lkml/2021/7/14/736 This patch (of 2): Let's produce a warning if we trying to remove non-existing IPC object from IPC namespace kht/idr structures. This allows us to catch possible bugs when the ipc_rmid() function was called with inconsistent struct ipc_ids*, struct kern_ipc_perm* arguments. Link: https://lkml.kernel.org/r/20211027224348.611025-1-alexander.mikhalitsyn@virtuozzo.com Link: https://lkml.kernel.org/r/20211027224348.611025-2-alexander.mikhalitsyn@virtuozzo.com Co-developed-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Andrei Vagin <avagin@gmail.com> Cc: Pavel Tikhomirov <ptikhomirov@virtuozzo.com> Cc: Vasily Averin <vvs@virtuozzo.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 126e8bee Fri Nov 19 17:43:18 MST 2021 Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com> ipc: WARN if trying to remove ipc object which is absent Patch series "shm: shm_rmid_forced feature fixes". Some time ago I met kernel crash after CRIU restore procedure, fortunately, it was CRIU restore, so, I had dump files and could do restore many times and crash reproduced easily. After some investigation I've constructed the minimal reproducer. It was found that it's use-after-free and it happens only if sysctl kernel.shm_rmid_forced = 1. The key of the problem is that the exit_shm() function not handles shp's object destroy when task->sysvshm.shm_clist contains items from different IPC namespaces. In most cases this list will contain only items from one IPC namespace. How can this list contain object from different namespaces? The exit_shm() function is designed to clean up this list always when process leaves IPC namespace. But we made a mistake a long time ago and did not add a exit_shm() call into the setns() syscall procedures. The first idea was just to add this call to setns() syscall but it obviously changes semantics of setns() syscall and that's userspace-visible change. So, I gave up on this idea. The first real attempt to address the issue was just to omit forced destroy if we meet shp object not from current task IPC namespace [1]. But that was not the best idea because task->sysvshm.shm_clist was protected by rwsem which belongs to current task IPC namespace. It means that list corruption may occur. Second approach is just extend exit_shm() to properly handle shp's from different IPC namespaces [2]. This is really non-trivial thing, I've put a lot of effort into that but not believed that it's possible to make it fully safe, clean and clear. Thanks to the efforts of Manfred Spraul working an elegant solution was designed. Thanks a lot, Manfred! Eric also suggested the way to address the issue in ("[RFC][PATCH] shm: In shm_exit destroy all created and never attached segments") Eric's idea was to maintain a list of shm_clists one per IPC namespace, use lock-less lists. But there is some extra memory consumption-related concerns. An alternative solution which was suggested by me was implemented in ("shm: reset shm_clist on setns but omit forced shm destroy"). The idea is pretty simple, we add exit_shm() syscall to setns() but DO NOT destroy shm segments even if sysctl kernel.shm_rmid_forced = 1, we just clean up the task->sysvshm.shm_clist list. This chages semantics of setns() syscall a little bit but in comparision to the "naive" solution when we just add exit_shm() without any special exclusions this looks like a safer option. [1] https://lkml.org/lkml/2021/7/6/1108 [2] https://lkml.org/lkml/2021/7/14/736 This patch (of 2): Let's produce a warning if we trying to remove non-existing IPC object from IPC namespace kht/idr structures. This allows us to catch possible bugs when the ipc_rmid() function was called with inconsistent struct ipc_ids*, struct kern_ipc_perm* arguments. Link: https://lkml.kernel.org/r/20211027224348.611025-1-alexander.mikhalitsyn@virtuozzo.com Link: https://lkml.kernel.org/r/20211027224348.611025-2-alexander.mikhalitsyn@virtuozzo.com Co-developed-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Andrei Vagin <avagin@gmail.com> Cc: Pavel Tikhomirov <ptikhomirov@virtuozzo.com> Cc: Vasily Averin <vvs@virtuozzo.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 20401d10 Tue Sep 07 21:00:53 MDT 2021 Rafael Aquini <aquini@redhat.com> ipc: replace costly bailout check in sysvipc_find_ipc() sysvipc_find_ipc() was left with a costly way to check if the offset position fed to it is bigger than the total number of IPC IDs in use. So much so that the time it takes to iterate over /proc/sysvipc/* files grows exponentially for a custom benchmark that creates "N" SYSV shm segments and then times the read of /proc/sysvipc/shm (milliseconds): 12 msecs to read 1024 segs from /proc/sysvipc/shm 18 msecs to read 2048 segs from /proc/sysvipc/shm 65 msecs to read 4096 segs from /proc/sysvipc/shm 325 msecs to read 8192 segs from /proc/sysvipc/shm 1303 msecs to read 16384 segs from /proc/sysvipc/shm 5182 msecs to read 32768 segs from /proc/sysvipc/shm The root problem lies with the loop that computes the total amount of ids in use to check if the "pos" feeded to sysvipc_find_ipc() grew bigger than "ids->in_use". That is a quite inneficient way to get to the maximum index in the id lookup table, specially when that value is already provided by struct ipc_ids.max_idx. This patch follows up on the optimization introduced via commit 15df03c879836 ("sysvipc: make get_maxid O(1) again") and gets rid of the aforementioned costly loop replacing it by a simpler checkpoint based on ipc_get_maxidx() returned value, which allows for a smooth linear increase in time complexity for the same custom benchmark: 2 msecs to read 1024 segs from /proc/sysvipc/shm 2 msecs to read 2048 segs from /proc/sysvipc/shm 4 msecs to read 4096 segs from /proc/sysvipc/shm 9 msecs to read 8192 segs from /proc/sysvipc/shm 19 msecs to read 16384 segs from /proc/sysvipc/shm 39 msecs to read 32768 segs from /proc/sysvipc/shm Link: https://lkml.kernel.org/r/20210809203554.1562989-1-aquini@redhat.com Signed-off-by: Rafael Aquini <aquini@redhat.com> Acked-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Waiman Long <llong@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 20401d10 Tue Sep 07 21:00:53 MDT 2021 Rafael Aquini <aquini@redhat.com> ipc: replace costly bailout check in sysvipc_find_ipc() sysvipc_find_ipc() was left with a costly way to check if the offset position fed to it is bigger than the total number of IPC IDs in use. So much so that the time it takes to iterate over /proc/sysvipc/* files grows exponentially for a custom benchmark that creates "N" SYSV shm segments and then times the read of /proc/sysvipc/shm (milliseconds): 12 msecs to read 1024 segs from /proc/sysvipc/shm 18 msecs to read 2048 segs from /proc/sysvipc/shm 65 msecs to read 4096 segs from /proc/sysvipc/shm 325 msecs to read 8192 segs from /proc/sysvipc/shm 1303 msecs to read 16384 segs from /proc/sysvipc/shm 5182 msecs to read 32768 segs from /proc/sysvipc/shm The root problem lies with the loop that computes the total amount of ids in use to check if the "pos" feeded to sysvipc_find_ipc() grew bigger than "ids->in_use". That is a quite inneficient way to get to the maximum index in the id lookup table, specially when that value is already provided by struct ipc_ids.max_idx. This patch follows up on the optimization introduced via commit 15df03c879836 ("sysvipc: make get_maxid O(1) again") and gets rid of the aforementioned costly loop replacing it by a simpler checkpoint based on ipc_get_maxidx() returned value, which allows for a smooth linear increase in time complexity for the same custom benchmark: 2 msecs to read 1024 segs from /proc/sysvipc/shm 2 msecs to read 2048 segs from /proc/sysvipc/shm 4 msecs to read 4096 segs from /proc/sysvipc/shm 9 msecs to read 8192 segs from /proc/sysvipc/shm 19 msecs to read 16384 segs from /proc/sysvipc/shm 39 msecs to read 32768 segs from /proc/sysvipc/shm Link: https://lkml.kernel.org/r/20210809203554.1562989-1-aquini@redhat.com Signed-off-by: Rafael Aquini <aquini@redhat.com> Acked-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Waiman Long <llong@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff b869d5be Wed Jun 30 19:57:18 MDT 2021 Manfred Spraul <manfred@colorfullife.com> ipc/util.c: use binary search for max_idx If semctl(), msgctl() and shmctl() are called with IPC_INFO, SEM_INFO, MSG_INFO or SHM_INFO, then the return value is the index of the highest used index in the kernel's internal array recording information about all SysV objects of the requested type for the current namespace. (This information can be used with repeated ..._STAT or ..._STAT_ANY operations to obtain information about all SysV objects on the system.) There is a cache for this value. But when the cache needs up be updated, then the highest used index is determined by looping over all possible values. With the introduction of IPCMNI_EXTEND_SHIFT, this could be a loop over 16 million entries. And due to /proc/sys/kernel/*next_id, the index values do not need to be consecutive. With <write 16000000 to msg_next_id>, msgget(), msgctl(,IPC_RMID) in a loop, I have observed a performance increase of around factor 13000. As there is no get_last() function for idr structures: Implement a "get_last()" using a binary search. As far as I see, ipc is the only user that needs get_last(), thus implement it in ipc/util.c and not in a central location. [akpm@linux-foundation.org: tweak comment, fix typo] Link: https://lkml.kernel.org/r/20210425075208.11777-2-manfred@colorfullife.com Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Acked-by: Davidlohr Bueso <dbueso@suse.de> Cc: <1vier1@web.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 5e698222 Wed May 13 18:50:48 MDT 2020 Vasily Averin <vvs@virtuozzo.com> ipc/util.c: sysvipc_find_ipc() incorrectly updates position index Commit 89163f93c6f9 ("ipc/util.c: sysvipc_find_ipc() should increase position index") is causing this bug (seen on 5.6.8): # ipcs -q ------ Message Queues -------- key msqid owner perms used-bytes messages # ipcmk -Q Message queue id: 0 # ipcs -q ------ Message Queues -------- key msqid owner perms used-bytes messages 0x82db8127 0 root 644 0 0 # ipcmk -Q Message queue id: 1 # ipcs -q ------ Message Queues -------- key msqid owner perms used-bytes messages 0x82db8127 0 root 644 0 0 0x76d1fb2a 1 root 644 0 0 # ipcrm -q 0 # ipcs -q ------ Message Queues -------- key msqid owner perms used-bytes messages 0x76d1fb2a 1 root 644 0 0 0x76d1fb2a 1 root 644 0 0 # ipcmk -Q Message queue id: 2 # ipcrm -q 2 # ipcs -q ------ Message Queues -------- key msqid owner perms used-bytes messages 0x76d1fb2a 1 root 644 0 0 0x76d1fb2a 1 root 644 0 0 # ipcmk -Q Message queue id: 3 # ipcrm -q 1 # ipcs -q ------ Message Queues -------- key msqid owner perms used-bytes messages 0x7c982867 3 root 644 0 0 0x7c982867 3 root 644 0 0 0x7c982867 3 root 644 0 0 0x7c982867 3 root 644 0 0 Whenever an IPC item with a low id is deleted, the items with higher ids are duplicated, as if filling a hole. new_pos should jump through hole of unused ids, pos can be updated inside "for" cycle. Fixes: 89163f93c6f9 ("ipc/util.c: sysvipc_find_ipc() should increase position index") Reported-by: Andreas Schwab <schwab@suse.de> Reported-by: Randy Dunlap <rdunlap@infradead.org> Signed-off-by: Vasily Averin <vvs@virtuozzo.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Acked-by: Waiman Long <longman@redhat.com> Cc: NeilBrown <neilb@suse.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Peter Oberparleiter <oberpar@linux.ibm.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: <stable@vger.kernel.org> Link: http://lkml.kernel.org/r/4921fe9b-9385-a2b4-1dc4-1099be6d2e39@virtuozzo.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> diff 5e698222 Wed May 13 18:50:48 MDT 2020 Vasily Averin <vvs@virtuozzo.com> ipc/util.c: sysvipc_find_ipc() incorrectly updates position index Commit 89163f93c6f9 ("ipc/util.c: sysvipc_find_ipc() should increase position index") is causing this bug (seen on 5.6.8): # ipcs -q ------ Message Queues -------- key msqid owner perms used-bytes messages # ipcmk -Q Message queue id: 0 # ipcs -q ------ Message Queues -------- key msqid owner perms used-bytes messages 0x82db8127 0 root 644 0 0 # ipcmk -Q Message queue id: 1 # ipcs -q ------ Message Queues -------- key msqid owner perms used-bytes messages 0x82db8127 0 root 644 0 0 0x76d1fb2a 1 root 644 0 0 # ipcrm -q 0 # ipcs -q ------ Message Queues -------- key msqid owner perms used-bytes messages 0x76d1fb2a 1 root 644 0 0 0x76d1fb2a 1 root 644 0 0 # ipcmk -Q Message queue id: 2 # ipcrm -q 2 # ipcs -q ------ Message Queues -------- key msqid owner perms used-bytes messages 0x76d1fb2a 1 root 644 0 0 0x76d1fb2a 1 root 644 0 0 # ipcmk -Q Message queue id: 3 # ipcrm -q 1 # ipcs -q ------ Message Queues -------- key msqid owner perms used-bytes messages 0x7c982867 3 root 644 0 0 0x7c982867 3 root 644 0 0 0x7c982867 3 root 644 0 0 0x7c982867 3 root 644 0 0 Whenever an IPC item with a low id is deleted, the items with higher ids are duplicated, as if filling a hole. new_pos should jump through hole of unused ids, pos can be updated inside "for" cycle. Fixes: 89163f93c6f9 ("ipc/util.c: sysvipc_find_ipc() should increase position index") Reported-by: Andreas Schwab <schwab@suse.de> Reported-by: Randy Dunlap <rdunlap@infradead.org> Signed-off-by: Vasily Averin <vvs@virtuozzo.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Acked-by: Waiman Long <longman@redhat.com> Cc: NeilBrown <neilb@suse.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Peter Oberparleiter <oberpar@linux.ibm.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: <stable@vger.kernel.org> Link: http://lkml.kernel.org/r/4921fe9b-9385-a2b4-1dc4-1099be6d2e39@virtuozzo.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> |
Completed in 339 milliseconds