Cancel and wait for all requests on exit
Subsystems (6)
Jens Axboe
maintainerv3 cover Cancel and wait for all requests on exit
Hi,
Currently, when a ring is being shut down, some cancelations may happen out-of-line. This means that an application cannot rely on the ring exit meaning that any IO has fully completed, or someone else waiting on an application (which has a ring with pending IO) being terminated will mean that all requests are done. This has also manifested itself as various testing sometimes finding a mount point busy after a test has exited, because it may take a brief period of time for things to quiesce and be fully done.
This patchset makes the task wait on the cancelations, if any, when the io_uring file fd is being put. That has the effect of ensuring that pending IO has fully completed, and files closed, before the ring exit returns.
io_uring runs cancelations off kthread based fallback work, and patch 1 enables these to run task_work so we'll know that we can put files inline and not need rely on deadlock prune flushing of the delayed fput work item. The rest of the patches are all io_uring based and pretty straight forward. This fundamentally doesn't change how cancelations work, it just waits on the out-of-line cancelations and request queiscing before exiting.
The switch away from percpu reference counts is done mostly because exiting those references will cost us an RCU grace period. That will noticeably slow down the ring tear down by anywhere from 10-100x.
The changes can also be found here:
https://git.kernel.dk/cgit/linux/log/?h=io_uring-exit-cancel.2
Since v2:
- Fix logic error io_queue_iowq()
fs/file_table.c | 2 +- include/linux/io_uring_types.h | 4 +- include/linux/sched.h | 2 +- io_uring/io_uring.c | 79 +++++++++++++++++++++++----------- io_uring/io_uring.h | 3 +- io_uring/msg_ring.c | 4 +- io_uring/refs.h | 43 ++++++++++++++++++ io_uring/register.c | 2 +- io_uring/rw.c | 2 +- io_uring/sqpoll.c | 2 +- io_uring/zcrx.c | 4 +- kernel/fork.c | 2 +- 12 files changed, 111 insertions(+), 38 deletions(-)
Signature · 1 line
Jens Axboe
Jens Axboe
maintainerPATCH v3 1/5 fs: gate final fput task_work on PF_NO_TASKWORK
fput currently gates whether or not a task can run task_work on the PF_KTHREAD flag, which excludes kernel threads as they don't usually run task_work as they never exit to userspace. This punts the final fput done from a kthread to a delayed work item instead of using task_work.
It's perfectly viable to have the final fput done by the kthread itself, as long as it will actually run the task_work. Add a PF_NO_TASKWORK flag which is set by default by a kernel thread, and gate the task_work fput on that instead. This enables a kernel thread to clear this flag temporarily while putting files, as long as it runs its task_work manually.
This enables users like io_uring to ensure that when the final fput of a file is done as part of ring teardown to run the local task_work and hence know that all files have been properly put, without needing to resort to workqueue flushing tricks which can deadlock.
No functional changes in this patch.
--- fs/file_table.c | 2 +- include/linux/sched.h | 2 +- kernel/fork.c | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-)
fs/file_table.c +1 -1
@@ -521,7 +521,7 @@ static void __fput_deferred(struct file *file)return ; } -if (likely (!in_interrupt () && !(task->flags &PF_KTHREAD ))) { +if (likely (!in_interrupt () && !(task->flags &PF_NO_TASKWORK ))) {init_task_work (&file->f_task_work, ____fput);if (!task_work_add (task, &file->f_task_work,TWA_RESUME ))return ;
include/linux/sched.h +1 -1
@@ -1736,7 +1736,7 @@ extern struct pid *cad_pid;* I am cleaning dirty pages from some other bdi. */ #define PF_KTHREAD 0x00200000/* I am a kernel thread */ #define PF_RANDOMIZE 0x00400000/* Randomize virtual address space */ -#define PF__HOLE__00800000 0x00800000 +#define PF_NO_TASKWORK 0x00800000/* task doesn't run task_work */ #define PF__HOLE__01000000 0x01000000#define PF__HOLE__02000000 0x02000000#define PF_NO_SETAFFINITY 0x04000000/* Userland is not allowed to meddle with cpus_mask */
kernel/fork.c +1 -1
@@ -2261,7 +2261,7 @@ __latent_entropy struct task_struct *copy_process( goto fork_out; p->flags &= ~PF_KTHREAD ;if (args->kthread) - p->flags |=PF_KTHREAD ; + p->flags |=PF_KTHREAD |PF_NO_TASKWORK ;if (args->user_worker) {/* * Mark us a user worker, and block any signal that isn't
Signature · 1 line
2.49.0
Re: [PATCH 1/5] fs: gate final fput task_work on PF_NO_TASKWORK
Quoted 17 lines · Jens Axboe
fput currently gates whether or not a task can run
task_workon thePF_KTHREADflag, which excludes kernel threads as they don't usually runtask_workas they never exit to userspace. This punts the final fput done from a kthread to a delayed work item instead of usingtask_work.It's perfectly viable to have the final fput done by the kthread itself, as long as it will actually run the
task_work. Add aPF_NO_TASKWORKflag which is set by default by a kernel thread, and gate thetask_workfput on that instead. This enables a kernel thread to clear this flag temporarily while putting files, as long as it runs itstask_workmanually.This enables users like
io_uringto ensure that when the final fput of a file is done as part of ring teardown to run the localtask_workand hence know that all files have been properly put, without needing to resort to workqueue flushing tricks which can deadlock.
Quoted 5 lines
No functional changes in this patch.
Cc Christian Brauner <[email protected]>Signed-off-by Jens Axboe <[email protected]>---
Seems fine. Although it has some potential for abuse. So maybe a VFS_WARN_ON_ONCE() that PF_NO_TASKWORK is only used with PF_KTHREAD would make sense.
Quoted 47 lines
fs/file_table.c | 2 +- include/linux/sched.h | 2 +- kernel/fork.c | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-)fs/file_table.c +1
-1@@ -521,7 +521,7 @@ static void __fput_deferred(struct file *file)return ; } -if (likely (!in_interrupt () && !(task->flags &PF_KTHREAD ))) { +if (likely (!in_interrupt () && !(task->flags &PF_NO_TASKWORK ))) {init_task_work (&file->f_task_work, ____fput);if (!task_work_add (task, &file->f_task_work,TWA_RESUME ))return ;include/linux/sched.h +1
-1@@ -1736,7 +1736,7 @@ extern struct pid *cad_pid;* I am cleaning dirty pages from some other bdi. */ #define PF_KTHREAD 0x00200000/* I am a kernel thread */ #define PF_RANDOMIZE 0x00400000/* Randomize virtual address space */ -#define PF__HOLE__00800000 0x00800000 +#define PF_NO_TASKWORK 0x00800000/* task doesn't run task_work */ #define PF__HOLE__01000000 0x01000000#define PF__HOLE__02000000 0x02000000#define PF_NO_SETAFFINITY 0x04000000/* Userland is not allowed to meddle with cpus_mask */ kernel/fork.c +1
-1@@ -2261,7 +2261,7 @@ __latent_entropy struct task_struct *copy_process( goto fork_out; p->flags &= ~PF_KTHREAD ;if (args->kthread) - p->flags |=PF_KTHREAD ; + p->flags |=PF_KTHREAD |PF_NO_TASKWORK ;if (args->user_worker) {/* * Mark us a user worker, and block any signal that isn't Signature · 1 line
2.49.0
Jens Axboe
maintainerRe: [PATCH 1/5] fs: gate final fput task_work on PF_NO_TASKWORK
Quoted 3 lines · Christian Brauner
Seems fine. Although it has some potential for abuse. So maybe a
VFS_WARN_ON_ONCE()thatPF_NO_TASKWORKis only used withPF_KTHREADwould make sense.
Can certainly add that. You'd want that before the check for in_interrupt and PF_NO_TASKWORK? Something ala
/* PF_NO_TASKWORK should only be used with PF_KTHREAD */ VFS_WARN_ON_ONCE((task->flags & PF_NO_TASKWORK) && !(task->flags & PF_KTHREAD));
?
Quoted 1 line
Thanks!
Signature · 1 line
Jens Axboe
Christian Brauner
maintainerRe: [PATCH 1/5] fs: gate final fput task_work on PF_NO_TASKWORK
Quoted 5 lines · Jens Axboe
Quoted 3 lines · Christian Brauner
Seems fine. Although it has some potential for abuse. So maybe a
VFS_WARN_ON_ONCE()thatPF_NO_TASKWORKis only used withPF_KTHREADwould make sense.
Quoted 7 lines
Can certainly add that. You'd want that before the check for
in_interruptandPF_NO_TASKWORK? Something ala/* PF_NO_TASKWORK should only be used with PF_KTHREAD */ VFS_WARN_ON_ONCE((task->flags & PF_NO_TASKWORK) && !(task->flags & PF_KTHREAD));?
Yeah, sounds good!
Jens Axboe
maintainerRe: [PATCH 1/5] fs: gate final fput task_work on PF_NO_TASKWORK
Quoted 9 lines · Christian Brauner
Quoted 8 lines · Jens Axboe
Quoted 3 lines · Christian Brauner
Seems fine. Although it has some potential for abuse. So maybe a
VFS_WARN_ON_ONCE()thatPF_NO_TASKWORKis only used withPF_KTHREADwould make sense.Can certainly add that. You'd want that before the check for
in_interruptandPF_NO_TASKWORK? Something ala
Quoted 6 lines
Quoted 4 lines
/* PF_NO_TASKWORK should only be used with PF_KTHREAD */ VFS_WARN_ON_ONCE((task->flags & PF_NO_TASKWORK) && !(task->flags & PF_KTHREAD));?
Yeah, sounds good!
I used the usual XOR trick for this kind of test, but placed in the same spot:
Signature · 1 line
Jens Axboe
Re: [PATCH 1/5] fs: gate final fput task_work on PF_NO_TASKWORK
In reply to Jens AxboeQuoted 58 lines · Jens Axboe
fput currently gates whether or not a task can run
task_workon thePF_KTHREADflag, which excludes kernel threads as they don't usually runtask_workas they never exit to userspace. This punts the final fput done from a kthread to a delayed work item instead of usingtask_work.It's perfectly viable to have the final fput done by the kthread itself, as long as it will actually run the
task_work. Add aPF_NO_TASKWORKflag which is set by default by a kernel thread, and gate thetask_workfput on that instead. This enables a kernel thread to clear this flag temporarily while putting files, as long as it runs itstask_workmanually.This enables users like
io_uringto ensure that when the final fput of a file is done as part of ring teardown to run the localtask_workand hence know that all files have been properly put, without needing to resort to workqueue flushing tricks which can deadlock.No functional changes in this patch.
Cc Christian Brauner <[email protected]>Signed-off-by Jens Axboe <[email protected]>--- fs/file_table.c | 2 +- include/linux/sched.h | 2 +- kernel/fork.c | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-)fs/file_table.c +1
-1@@ -521,7 +521,7 @@ static void __fput_deferred(struct file *file)return ; } -if (likely (!in_interrupt () && !(task->flags &PF_KTHREAD ))) { +if (likely (!in_interrupt () && !(task->flags &PF_NO_TASKWORK ))) {init_task_work (&file->f_task_work, ____fput);if (!task_work_add (task, &file->f_task_work,TWA_RESUME ))return ;include/linux/sched.h +1
-1@@ -1736,7 +1736,7 @@ extern struct pid *cad_pid;* I am cleaning dirty pages from some other bdi. */ #define PF_KTHREAD 0x00200000/* I am a kernel thread */ #define PF_RANDOMIZE 0x00400000/* Randomize virtual address space */ -#define PF__HOLE__00800000 0x00800000 +#define PF_NO_TASKWORK 0x00800000/* task doesn't run task_work */ #define PF__HOLE__01000000 0x01000000#define PF__HOLE__02000000 0x02000000#define PF_NO_SETAFFINITY 0x04000000/* Userland is not allowed to meddle with cpus_mask */ kernel/fork.c +0
-0
kernel/fork.c __latent_entropy struct task_struct *copy_process( · line 2267
Quoted 8 lines
kernel/fork.c +1
-1@@ -2261,7 +2261,7 @@ __latent_entropy struct task_struct *copy_process( goto fork_out; p->flags &= ~PF_KTHREAD ;if (args->kthread) - p->flags |=PF_KTHREAD ; + p->flags |=PF_KTHREAD |PF_NO_TASKWORK ;if (args->user_worker) {/* * Mark us a user worker, and block any signal that isn't
I don't have comments on the semantics here, I do have comments on some future-proofing.
To my reading kthreads on the stock kernel never execute task_work.
This suggests it would be nice for task_work_add() to at least WARN_ON when executing with a kthread. After all you don't want a task_work_add consumer adding work which will never execute.
But then for your patch to not produce any splats there would have to be a flag blessing select kthreads as legitimate task_work consumers.
So my suggestion would be to add the WARN_ON() in task_work_add() prior to anything in this patchset, then this patch would be extended with a flag (PF_KTHREAD_DOES_TASK_WORK?) and relevant io_uring threads would get the flag.
Then the machinery which sets/unsets PF_NO_TASKWORK can assert that:
1. it operates on a kthread...
2. ...with the PF_KTHREAD_DOES_TASK_WORK flag
This is just a suggestion though.
Jens Axboe
maintainerRe: [PATCH 1/5] fs: gate final fput task_work on PF_NO_TASKWORK
Quoted 71 lines · Mateusz Guzik
Quoted 58 lines · Jens Axboe
fput currently gates whether or not a task can run
task_workon thePF_KTHREADflag, which excludes kernel threads as they don't usually runtask_workas they never exit to userspace. This punts the final fput done from a kthread to a delayed work item instead of usingtask_work.It's perfectly viable to have the final fput done by the kthread itself, as long as it will actually run the
task_work. Add aPF_NO_TASKWORKflag which is set by default by a kernel thread, and gate thetask_workfput on that instead. This enables a kernel thread to clear this flag temporarily while putting files, as long as it runs itstask_workmanually.This enables users like
io_uringto ensure that when the final fput of a file is done as part of ring teardown to run the localtask_workand hence know that all files have been properly put, without needing to resort to workqueue flushing tricks which can deadlock.No functional changes in this patch.
Cc Christian Brauner <[email protected]>Signed-off-by Jens Axboe <[email protected]>--- fs/file_table.c | 2 +- include/linux/sched.h | 2 +- kernel/fork.c | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-)fs/file_table.c +1
-1@@ -521,7 +521,7 @@ static void __fput_deferred(struct file *file)return ; } -if (likely (!in_interrupt () && !(task->flags &PF_KTHREAD ))) { +if (likely (!in_interrupt () && !(task->flags &PF_NO_TASKWORK ))) {init_task_work (&file->f_task_work, ____fput);if (!task_work_add (task, &file->f_task_work,TWA_RESUME ))return ;include/linux/sched.h +1
-1@@ -1736,7 +1736,7 @@ extern struct pid *cad_pid;* I am cleaning dirty pages from some other bdi. */ #define PF_KTHREAD 0x00200000/* I am a kernel thread */ #define PF_RANDOMIZE 0x00400000/* Randomize virtual address space */ -#define PF__HOLE__00800000 0x00800000 +#define PF_NO_TASKWORK 0x00800000/* task doesn't run task_work */ #define PF__HOLE__01000000 0x01000000#define PF__HOLE__02000000 0x02000000#define PF_NO_SETAFFINITY 0x04000000/* Userland is not allowed to meddle with cpus_mask */ kernel/fork.c +0
-0kernel/fork.c __latent_entropy struct task_struct *copy_process( · line 2267
Quoted 8 lines
kernel/fork.c +1
-1@@ -2261,7 +2261,7 @@ __latent_entropy struct task_struct *copy_process( goto fork_out; p->flags &= ~PF_KTHREAD ;if (args->kthread) - p->flags |=PF_KTHREAD ; + p->flags |=PF_KTHREAD |PF_NO_TASKWORK ;if (args->user_worker) {/* * Mark us a user worker, and block any signal that isn't I don't have comments on the semantics here, I do have comments on some future-proofing.
Quoted 1 line
To my reading kthreads on the stock kernel never execute
task_work.
Correct
Quoted 3 lines
This suggests it would be nice for
task_work_add()to at leastWARN_ONwhen executing with a kthread. After all you don't want atask_work_addconsumer adding work which will never execute.
I don't think there's much need for that, as I'm not aware of any kernel usage that had a bug due to that. And if you did, you'd find it pretty quick during testing as that work would just never execute.
Quoted 2 lines
But then for your patch to not produce any splats there would have to be a flag blessing select kthreads as legitimate
task_workconsumers.
This patchset very much adds a specific flag for that, PF_NO_TASKWORK, and kernel threads have it set by default. It just separates the "do I run task_work" flag from PF_KTHREAD. So yes you could add:
WARN_ON_ONCE(task->flags & PF_NO_TASKWORK);
to task_work_add(), but I'm not really convinced it'd be super useful.
Signature · 1 line
Jens Axboe
Jens Axboe
maintainerPATCH v3 2/5 io_uring: mark exit side kworkers as task_work capable
In reply to Jens AxboeThere are two types of work here:
1) Fallback work, if the task is exiting
2) The exit side cancelations
and both of them may do the final fput() of a file. When this happens, fput() will schedule delayed work. This slows down exits when io_uring needs to wait for that work to finish. It is possible to flush this via flush_delayed_fput(), but that's a big hammer as other unrelated files could be involved, and from other tasks as well.
Add two io_uring helpers to temporarily clear PF_NO_TASKWORK for the worker threads, and run any queued task_work before setting the flag again. Then we can ensure we only flush related items that received their final fput as part of work cancelation and flushing.
For now these are io_uring private, but could obviously be made generically available, should there be a need to do so.
--- io_uring/io_uring.c | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+)
io_uring/io_uring.c +21 -0
@@ -238,6 +238,20 @@ static inline void io_req_add_to_cache(struct io_kiocb *req, struct io_ring_ctxwq_stack_add_head (&req->comp_list, &ctx->submit_state.free_list); } +static __cold void io_kworker_tw_start (void ) +{ +if (WARN_ON_ONCE (!(current->flags &PF_NO_TASKWORK ))) +return ; + current->flags &= ~PF_NO_TASKWORK ; +} + +static __cold void io_kworker_tw_end (void ) +{ +while (task_work_pending (current)) +task_work_run (); + current->flags |=PF_NO_TASKWORK ; +} +static __cold void io_ring_ctx_ref_free (struct percpu_ref *ref) { struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs);
@@ -253,6 +267,8 @@ static __cold void io_fallback_req_func(struct work_struct *work)struct io_kiocb *req, *tmp;struct io_tw_state ts = {}; +io_kworker_tw_start (); +percpu_ref_get (&ctx->refs);mutex_lock (&ctx->uring_lock);llist_for_each_entry_safe (req, tmp, node, io_task_work.node)
@@ -260,6 +276,7 @@ static __cold void io_fallback_req_func(struct work_struct *work)io_submit_flush_completions (ctx);mutex_unlock (&ctx->uring_lock);percpu_ref_put (&ctx->refs); +io_kworker_tw_end (); }static int io_alloc_hash_table (struct io_hash_table *table,unsigned bits)
@@ -2876,6 +2893,8 @@ static __cold void io_ring_exit_work(struct work_struct *work)struct io_tctx_node *node;int ret; +io_kworker_tw_start (); +/* * If we're doing polled IO and end up having requests being * submitted async (out-of-line), then completions can come in while
@@ -2932,6 +2951,8 @@ static __cold void io_ring_exit_work(struct work_struct *work)*/ }while (!wait_for_completion_interruptible_timeout (&ctx->ref_comp, interval)); +io_kworker_tw_end (); +init_completion (&exit.completion);init_task_work (&exit.task_work, io_tctx_exit_cb); exit.ctx = ctx;
Signature · 1 line
2.49.0
Christian Brauner
maintainerRe: [PATCH 2/5] io_uring: mark exit side kworkers as task_work capable
Quoted 7 lines · Jens Axboe
There are two types of work here:
1) Fallback work, if the task is exiting
2) The exit side cancelationsand both of them may do the final
fput()of a file. When this happens,fput()will schedule delayed work. This slows down exits whenio_uring
I was a bit surprised by this because it means that all those __fput()s are done with kthread credentials which is a bit surprising (but harmless afaict).
Quoted 8 lines
needs to wait for that work to finish. It is possible to flush this via
flush_delayed_fput(), but that's a big hammer as other unrelated files could be involved, and from other tasks as well.Add two
io_uringhelpers to temporarily clearPF_NO_TASKWORKfor the worker threads, and run any queuedtask_workbefore setting the flag again. Then we can ensure we only flush related items that received their final fput as part of work cancelation and flushing.
Ok, so the only change is that this isn't offloaded to the global delayed fput workqueue but to the task work that you're running off of your kthread helpers.
Quoted 72 lines
For now these are
io_uringprivate, but could obviously be made generically available, should there be a need to do so.Signed-off-by Jens Axboe <[email protected]>--- io_uring/io_uring.c | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+)io_uring/io_uring.c +21
-0@@ -238,6 +238,20 @@ static inline void io_req_add_to_cache(struct io_kiocb *req, struct io_ring_ctxwq_stack_add_head (&req->comp_list, &ctx->submit_state.free_list); } +static __cold void io_kworker_tw_start (void ) +{ +if (WARN_ON_ONCE (!(current->flags &PF_NO_TASKWORK ))) +return ; + current->flags &= ~PF_NO_TASKWORK ; +} + +static __cold void io_kworker_tw_end (void ) +{ +while (task_work_pending (current)) +task_work_run (); + current->flags |=PF_NO_TASKWORK ; +} +static __cold void io_ring_ctx_ref_free (struct percpu_ref *ref) { struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs);@@ -253,6 +267,8 @@ static __cold void io_fallback_req_func(struct work_struct *work)struct io_kiocb *req, *tmp;struct io_tw_state ts = {}; +io_kworker_tw_start (); +percpu_ref_get (&ctx->refs);mutex_lock (&ctx->uring_lock);llist_for_each_entry_safe (req, tmp, node, io_task_work.node)@@ -260,6 +276,7 @@ static __cold void io_fallback_req_func(struct work_struct *work)io_submit_flush_completions (ctx);mutex_unlock (&ctx->uring_lock);percpu_ref_put (&ctx->refs); +io_kworker_tw_end (); }static int io_alloc_hash_table (struct io_hash_table *table,unsigned bits)@@ -2876,6 +2893,8 @@ static __cold void io_ring_exit_work(struct work_struct *work)struct io_tctx_node *node;int ret; +io_kworker_tw_start (); +/* * If we're doing polled IO and end up having requests being * submitted async (out-of-line), then completions can come in while @@ -2932,6 +2951,8 @@ static __cold void io_ring_exit_work(struct work_struct *work)*/ }while (!wait_for_completion_interruptible_timeout (&ctx->ref_comp, interval)); +io_kworker_tw_end (); +init_completion (&exit.completion);init_task_work (&exit.task_work, io_tctx_exit_cb); exit.ctx = ctx;Signature · 1 line
2.49.0
Jens Axboe
maintainerRe: [PATCH 2/5] io_uring: mark exit side kworkers as task_work capable
Quoted 6 lines · Christian Brauner
Quoted 5 lines · Jens Axboe
There are two types of work here:
1) Fallback work, if the task is exiting
2) The exit side cancelations
Quoted 6 lines
Quoted 2 lines
and both of them may do the final
fput()of a file. When this happens,fput()will schedule delayed work. This slows down exits whenio_uringI was a bit surprised by this because it means that all those __fput()s are done with kthread credentials which is a bit surprising (but harmless afaict).
Sure hope it is, because that's already what happens off the delayed fput that it'd otherwise go through!
Quoted 4 lines
Quoted 4 lines
needs to wait for that work to finish. It is possible to flush this via
flush_delayed_fput(), but that's a big hammer as other unrelated files could be involved, and from other tasks as well.
Quoted 8 lines
Quoted 4 lines
Add two
io_uringhelpers to temporarily clearPF_NO_TASKWORKfor the worker threads, and run any queuedtask_workbefore setting the flag again. Then we can ensure we only flush related items that received their final fput as part of work cancelation and flushing.Ok, so the only change is that this isn't offloaded to the global delayed fput workqueue but to the task work that you're running off of your kthread helpers.
Exactly
Signature · 1 line
Jens Axboe
Jens Axboe
maintainerPATCH v3 3/5 io_uring: consider ring dead once the ref is marked dying
In reply to Jens AxboeFor queueing work to io-wq or adding normal task_work, io_uring will cancel the work items if the task is going away. If the ring is starting to go through teardown, the ref is marked as dying. Use that as well for the fallback/cancel mechanism.
For deferred task_work, this is done out-of-line as part of the exit work handling. Hence it doesn't need any extra checks in the hot path.
--- io_uring/io_uring.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
io_uring/io_uring.c +4 -2
@@ -555,7 +555,8 @@ static void io_queue_iowq(struct io_kiocb *req)* procedure rather than attempt to run this request (or create a new * worker for it). */ -if (WARN_ON_ONCE (!same_thread_group (tctx->task, current))) +if (WARN_ON_ONCE (!same_thread_group (tctx->task, current) || +percpu_ref_is_dying (&req->ctx->refs)))atomic_or (IO_WQ_WORK_CANCEL , &req->work.flags);trace_io_uring_queue_async_work (req,io_wq_is_hashed (&req->work));
@@ -1246,7 +1247,8 @@ static void io_req_normal_work_add(struct io_kiocb *req)return ; } -if (likely (!task_work_add (tctx->task, &tctx->task_work, ctx->notify_method))) +if (!percpu_ref_is_dying (&ctx->refs) && + !task_work_add (tctx->task, &tctx->task_work, ctx->notify_method))return ;io_fallback_tw (tctx, false);
Signature · 1 line
2.49.0
Jens Axboe
maintainerPATCH v3 4/5 io_uring: wait for cancelations on final ring put
In reply to Jens AxboeWe still offload the cancelation to a workqueue, as not to introduce dependencies between the exiting task waiting on cleanup, and that task needing to run task_work to complete the process.
This means that once the final ring put is done, any request that was inflight and needed cancelation will be done as well. Notably requests that hold references to files - once the ring fd close is done, we will have dropped any of those references too.
--- include/linux/io_uring_types.h | 2 ++ io_uring/io_uring.c | 17 +++++++++++++++++ 2 files changed, 19 insertions(+)
include/linux/io_uring_types.h +2 -0
@@ -450,6 +450,8 @@ struct io_ring_ctx {struct io_mapped_region param_region;/* just one zcrx per ring for now, will move to io_zcrx_ifq eventually */ struct io_mapped_region zcrx_region; + +struct completion *exit_comp; };/*
io_uring/io_uring.c +17 -0
@@ -2891,6 +2891,7 @@ static __cold void io_ring_exit_work(struct work_struct *work) struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work);unsigned long timeout = jiffies +HZ *60 *5 ;unsigned long interval =HZ /20 ; +struct completion *exit_comp;struct io_tctx_exit exit;struct io_tctx_node *node;int ret;
@@ -2955,6 +2956,10 @@ static __cold void io_ring_exit_work(struct work_struct *work)io_kworker_tw_end (); + exit_comp =READ_ONCE (ctx->exit_comp); +if (exit_comp) +complete (exit_comp); +init_completion (&exit.completion);init_task_work (&exit.task_work, io_tctx_exit_cb); exit.ctx = ctx;
@@ -3017,9 +3022,21 @@ static __cold void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)static int io_uring_release (struct inode *inode,struct file *file) {struct io_ring_ctx *ctx = file->private_data; +DECLARE_COMPLETION_ONSTACK (exit_comp); file->private_data =NULL ; +WRITE_ONCE (ctx->exit_comp, &exit_comp);io_ring_ctx_wait_and_kill (ctx); + +/* +* Wait for cancel to run before exiting task +*/ +do { +if (current->io_uring) +io_fallback_tw (current->io_uring, false); +cond_resched (); + }while (wait_for_completion_interruptible (&exit_comp)); +return 0 ; }
Signature · 1 line
2.49.0
Jens Axboe
maintainerPATCH v3 5/5 io_uring: switch away from percpu refcounts
In reply to Jens AxboeFor the common cases, the io_uring ref counts are all batched and hence need not be a percpu reference. This saves some memory on systems, but outside of that, it gets rid of needing a full RCU grace period on tearing down the reference. With io_uring now waiting on cancelations and IO during exit, this slows down the tear down a lot, up to 100x as slow.
--- include/linux/io_uring_types.h | 2 +- io_uring/io_uring.c | 47 ++++++++++++---------------------- io_uring/io_uring.h | 3 ++- io_uring/msg_ring.c | 4 +-- io_uring/refs.h | 43 +++++++++++++++++++++++++++++++ io_uring/register.c | 2 +- io_uring/rw.c | 2 +- io_uring/sqpoll.c | 2 +- io_uring/zcrx.c | 4 +-- 9 files changed, 70 insertions(+), 39 deletions(-)
include/linux/io_uring_types.h +1 -1
@@ -256,7 +256,7 @@ struct io_ring_ctx {struct task_struct *submitter_task;struct io_rings *rings; -struct percpu_ref refs; +atomic_long_t refs;clockid_t clockid;enum tk_offsets clock_offset;
io_uring/io_uring.c +17 -30
158 lines not on this page · raw message
io_uring/io_uring.h +2 -1
@@ -13,6 +13,7 @@#include "slist.h" #include "filetable.h" #include "opdef.h" +#include "refs.h" #ifndef CREATE_TRACE_POINTS #include <trace/events/io_uring.h>
@@ -142,7 +143,7 @@ static inline void io_lockdep_assert_cq_locked(struct io_ring_ctx *ctx)* Not from an SQE, as those cannot be submitted, but via * updating tagged resources. */ -if (!percpu_ref_is_dying (&ctx->refs)) +if (!io_ring_ref_is_dying (ctx)) lockdep_assert(current == ctx->submitter_task); }#endif
io_uring/msg_ring.c +2 -2
@@ -83,7 +83,7 @@ static void io_msg_tw_complete(struct io_kiocb *req, io_tw_token_t tw) }if (req)kmem_cache_free (req_cachep, req); -percpu_ref_put (&ctx->refs); +io_ring_ref_put (ctx); } static int io_msg_remote_post(struct io_ring_ctx *ctx, struct io_kiocb *req,
@@ -96,7 +96,7 @@ static int io_msg_remote_post(struct io_ring_ctx *ctx, struct io_kiocb *req, req->opcode =IORING_OP_NOP ; req->cqe.user_data = user_data;io_req_set_res (req, res, cflags); -percpu_ref_get (&ctx->refs); +io_ring_ref_get (ctx); req->ctx = ctx; req->tctx =NULL ; req->io_task_work.func = io_msg_tw_complete;
io_uring/refs.h +43 -0
47 lines not on this page · raw message
io_uring/register.c +1 -1
@@ -637,7 +637,7 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,* We don't quiesce the refs for register anymore and so it can't be * dying as we're holding a file ref here. */ -if (WARN_ON_ONCE (percpu_ref_is_dying (&ctx->refs))) +if (WARN_ON_ONCE (io_ring_ref_is_dying (ctx)))return -ENXIO ;if (ctx->submitter_task && ctx->submitter_task != current)
io_uring/rw.c +1 -1
@@ -496,7 +496,7 @@ static bool io_rw_should_reissue(struct io_kiocb *req)* Don't attempt to reissue from that path, just let it fail with * -EAGAIN. */ -if (percpu_ref_is_dying (&ctx->refs)) +if (io_ring_ref_is_dying (ctx))return false;io_meta_restore (io, &rw->kiocb);
io_uring/sqpoll.c +1 -1
@@ -184,7 +184,7 @@ static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)* Don't submit if refs are dying, good for io_uring_register(), * but also it is relied upon by io_ring_exit_work() */ -if (to_submit &&likely (!percpu_ref_is_dying (&ctx->refs)) && +if (to_submit &&likely (!io_ring_ref_is_dying (ctx)) && !(ctx->flags &IORING_SETUP_R_DISABLED )) ret =io_submit_sqes (ctx, to_submit);mutex_unlock (&ctx->uring_lock);
io_uring/zcrx.c +2 -2
@@ -630,7 +630,7 @@ static int io_pp_zc_init(struct page_pool *pp)if (pp->p.dma_dir !=DMA_FROM_DEVICE )return -EOPNOTSUPP ; -percpu_ref_get (&ifq->ctx->refs); +io_ring_ref_get (ifq->ctx);return 0 ; }
@@ -641,7 +641,7 @@ static void io_pp_zc_destroy(struct page_pool *pp)if (WARN_ON_ONCE (area->free_count != area->nia.num_niovs))return ; -percpu_ref_put (&ifq->ctx->refs); +io_ring_ref_put (ifq->ctx); } static int io_pp_nl_fill(void *mp_priv, struct sk_buff *rsp,
Signature · 1 line
2.49.0