fix deadlock in __io_req_task_submit()
cover fix deadlock in __io_req_task_submit()
This patchset is to fix a deadlock issue in __io_req_task_submit(), the first patch is a prep one and the second patch does the work.
Hao Xu (2):
io_uring: add uring_lock as an argument to io_sqe_files_unregister()
io_uring: don't hold uring_lock when calling io_run_task_work*
fs/io_uring.c | 29 ++++++++++++++++++----------- 1 file changed, 18 insertions(+), 11 deletions(-)
Signature · 1 line
1.8.3.1
PATCH 1/2 io_uring: add uring_lock as an argument to io_sqe_files_unregister()
io_sqe_files_unregister is currently called from several places:
- syscall io_uring_register (with uring_lock)
- io_ring_ctx_wait_and_kill() (without uring_lock)
There is a AA type deadlock in io_sqe_files_unregister(), thus we need to know if we hold uring_lock in io_sqe_files_unregister() to fix the issue.
--- fs/io_uring.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
fs/io_uring.c +5 -5
@@ -7339,7 +7339,7 @@ static void io_sqe_files_set_node(struct fixed_file_data *file_data,percpu_ref_get (&file_data->refs); } -static int io_sqe_files_unregister (struct io_ring_ctx *ctx) +static int io_sqe_files_unregister (struct io_ring_ctx *ctx,bool locked) {struct fixed_file_data *data = ctx->file_data;struct fixed_file_ref_node *backup_node, *ref_node =NULL ;
@@ -7872,13 +7872,13 @@ static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, ret =io_sqe_files_scm (ctx);if (ret) { -io_sqe_files_unregister (ctx); +io_sqe_files_unregister (ctx, true);return ret; } ref_node =alloc_fixed_file_ref_node (ctx);if (!ref_node) { -io_sqe_files_unregister (ctx); +io_sqe_files_unregister (ctx, true);return -ENOMEM ; }
@@ -8682,7 +8682,7 @@ static void io_ring_ctx_free(struct io_ring_ctx *ctx)css_put (ctx->sqo_blkcg_css);#endif -io_sqe_files_unregister (ctx); +io_sqe_files_unregister (ctx, false);io_eventfd_unregister (ctx);io_destroy_buffers (ctx);idr_destroy (&ctx->personality_idr);
@@ -10065,7 +10065,7 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, ret = -EINVAL ;if (arg || nr_args)break ; - ret =io_sqe_files_unregister (ctx); + ret =io_sqe_files_unregister (ctx, true);break ;case IORING_REGISTER_FILES_UPDATE : ret =io_sqe_files_update (ctx, arg, nr_args);
Signature · 1 line
1.8.3.1
Re: [PATCH 1/2] io_uring: add uring_lock as an argument to io_sqe_files_unregister()
Quoted 7 lines · Hao Xu
io_sqe_files_unregisteris currently called from several places:
- syscallio_uring_register(withuring_lock)
-io_ring_ctx_wait_and_kill()(withouturing_lock)There is a AA type deadlock in
io_sqe_files_unregister(), thus we need to know if we holduring_lockinio_sqe_files_unregister()to fix the issue.
It's ugly, just take the lock and kill the patch. There can't be any contention during io_ring_ctx_free anyway.
Quoted 54 lines
Signed-off-by Hao Xu <[email protected]>--- fs/io_uring.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)fs/io_uring.c +5
-5@@ -7339,7 +7339,7 @@ static void io_sqe_files_set_node(struct fixed_file_data *file_data,percpu_ref_get (&file_data->refs); } -static int io_sqe_files_unregister (struct io_ring_ctx *ctx) +static int io_sqe_files_unregister (struct io_ring_ctx *ctx,bool locked) {struct fixed_file_data *data = ctx->file_data;struct fixed_file_ref_node *backup_node, *ref_node =NULL ;@@ -7872,13 +7872,13 @@ static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, ret =io_sqe_files_scm (ctx);if (ret) { -io_sqe_files_unregister (ctx); +io_sqe_files_unregister (ctx, true);return ret; } ref_node =alloc_fixed_file_ref_node (ctx);if (!ref_node) { -io_sqe_files_unregister (ctx); +io_sqe_files_unregister (ctx, true);return -ENOMEM ; }@@ -8682,7 +8682,7 @@ static void io_ring_ctx_free(struct io_ring_ctx *ctx)css_put (ctx->sqo_blkcg_css);#endif -io_sqe_files_unregister (ctx); +io_sqe_files_unregister (ctx, false);io_eventfd_unregister (ctx);io_destroy_buffers (ctx);idr_destroy (&ctx->personality_idr);@@ -10065,7 +10065,7 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, ret = -EINVAL ;if (arg || nr_args)break ; - ret =io_sqe_files_unregister (ctx); + ret =io_sqe_files_unregister (ctx, true);break ;case IORING_REGISTER_FILES_UPDATE : ret =io_sqe_files_update (ctx, arg, nr_args);
Signature · 1 line
Pavel Begunkov
Re: [PATCH 1/2] io_uring: add uring_lock as an argument to io_sqe_files_unregister()
Quoted 11 lines · Pavel Begunkov
Quoted 7 lines · Hao Xu
io_sqe_files_unregisteris currently called from several places:
- syscallio_uring_register(withuring_lock)
-io_ring_ctx_wait_and_kill()(withouturing_lock)There is a AA type deadlock in
io_sqe_files_unregister(), thus we need to know if we holduring_lockinio_sqe_files_unregister()to fix the issue.It's ugly, just take the lock and kill the patch. There can't be any contention during
io_ring_ctx_freeanyway.
Hi Pavel, I don't get it, do you mean this patch isn't needed, and we can just unlock(&uring_lock) before io_run_task_work_sig() and lock(&uring_lock) after it? I knew there won't be contention during io_ring_ctx_free that's why there is no uring_lock in it. I tried to just do unlock(&uring_lock) before io_run_task_sig() without if(locked) check, it reports something like "there are unpaired mutex lock/unlock" since we cannot just unlock if it's from io_ring_ctx_free.
Quoted 56 lines
Quoted 55 lines
Signed-off-by Hao Xu <[email protected]>--- fs/io_uring.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)fs/io_uring.c +5
-5@@ -7339,7 +7339,7 @@ static void io_sqe_files_set_node(struct fixed_file_data *file_data,percpu_ref_get (&file_data->refs); } -static int io_sqe_files_unregister (struct io_ring_ctx *ctx) +static int io_sqe_files_unregister (struct io_ring_ctx *ctx,bool locked) {struct fixed_file_data *data = ctx->file_data;struct fixed_file_ref_node *backup_node, *ref_node =NULL ;@@ -7872,13 +7872,13 @@ static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, ret =io_sqe_files_scm (ctx);if (ret) { -io_sqe_files_unregister (ctx); +io_sqe_files_unregister (ctx, true);return ret; } ref_node =alloc_fixed_file_ref_node (ctx);if (!ref_node) { -io_sqe_files_unregister (ctx); +io_sqe_files_unregister (ctx, true);return -ENOMEM ; }@@ -8682,7 +8682,7 @@ static void io_ring_ctx_free(struct io_ring_ctx *ctx)css_put (ctx->sqo_blkcg_css);#endif -io_sqe_files_unregister (ctx); +io_sqe_files_unregister (ctx, false);io_eventfd_unregister (ctx);io_destroy_buffers (ctx);idr_destroy (&ctx->personality_idr);@@ -10065,7 +10065,7 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, ret = -EINVAL ;if (arg || nr_args)break ; - ret =io_sqe_files_unregister (ctx); + ret =io_sqe_files_unregister (ctx, true);break ;case IORING_REGISTER_FILES_UPDATE : ret =io_sqe_files_update (ctx, arg, nr_args);
Re: [PATCH 1/2] io_uring: add uring_lock as an argument to io_sqe_files_unregister()
Quoted 6 lines · Hao Xu
Quoted 5 lines · Pavel Begunkov
Quoted 4 lines · Hao Xu
io_sqe_files_unregisteris currently called from several places:
- syscallio_uring_register(withuring_lock)
-io_ring_ctx_wait_and_kill()(withouturing_lock)
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
There is a AA type deadlock in
io_sqe_files_unregister(), thus we need to know if we holduring_lockinio_sqe_files_unregister()to fix the issue.It's ugly, just take the lock and kill the patch. There can't be any contention during
io_ring_ctx_freeanyway.Hi Pavel, I don't get it, do you mean this patch isn't needed, and we can just
unlock(&uring_lock)beforeio_run_task_work_sig()andlock(&uring_lock)after it? I knew there won't be contention duringio_ring_ctx_freethat's why there is nouring_lockin it. I tried to just dounlock(&uring_lock)beforeio_run_task_sig()without if(locked) check, it reports something like "there are unpaired mutex lock/unlock" since we cannot just unlock if it's fromio_ring_ctx_free.
The ugly part is @locked. I know that there is already similar stuff around, but I may go long why and how much I don't like it.
io_ring_ctx_free()
{
...
lock(uring_lock);
files_unregister();
unlock(uring_lock);
...
}With this you'll always have the mutex locked in unregister, so can drop it unconditionally (if that will ever be needed). It's also cleaner from the synchronisation perspective.
Quoted 52 lines
Quoted 51 lines
Quoted 50 lines
Signed-off-by Hao Xu <[email protected]>---
fs/io_uring.c| 10 +++++-----1 file changed, 5 insertions(+), 5 deletions(-)fs/io_uring.c +3
-2@@ -7339,7 +7339,7 @@ static void io_sqe_files_set_node(struct fixed_file_data *file_data, percpu_ref_get(&file_data->refs); } -static int io_sqe_files_unregister(struct io_ring_ctx *ctx) +static int io_sqe_files_unregister(struct io_ring_ctx *ctx, bool locked) { struct fixed_file_data *data = ctx->file_data; struct fixed_file_ref_node *backup_node, *ref_node = NULL;@@ -7872,13 +7872,13 @@ static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, ret = io_sqe_files_scm(ctx); if (ret) { - io_sqe_files_unregister(ctx); + io_sqe_files_unregister(ctx, true); return ret; } ref_node = alloc_fixed_file_ref_node(ctx); if (!ref_node) { - io_sqe_files_unregister(ctx); + io_sqe_files_unregister(ctx, true); return -ENOMEM; } @@ -8682,7 +8682,7 @@ static void io_ring_ctx_free(struct io_ring_ctx *ctx) css_put(ctx->sqo_blkcg_css); #endif-
io_sqe_files_unregister(ctx);
+io_sqe_files_unregister(ctx, false);
io_eventfd_unregister(ctx);
io_destroy_buffers(ctx);
idr_destroy(&ctx->personality_idr);fs/io_uring.c +1
-1@@ -10065,7 +10065,7 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, ret = -EINVAL; if (arg || nr_args) break; - ret = io_sqe_files_unregister(ctx); + ret = io_sqe_files_unregister(ctx, true); break;case
IORING_REGISTER_FILES_UPDATE:
ret =io_sqe_files_update(ctx, arg, nr_args);
Signature · 1 line
Pavel Begunkov
Jens Axboe
maintainerRe: [PATCH 1/2] io_uring: add uring_lock as an argument to io_sqe_files_unregister()
Quoted 29 lines · Pavel Begunkov
Quoted 6 lines · Hao Xu
Quoted 5 lines · Pavel Begunkov
Quoted 4 lines · Hao Xu
io_sqe_files_unregisteris currently called from several places:
- syscallio_uring_register(withuring_lock)
-io_ring_ctx_wait_and_kill()(withouturing_lock)Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
There is a AA type deadlock in
io_sqe_files_unregister(), thus we need to know if we holduring_lockinio_sqe_files_unregister()to fix the issue.It's ugly, just take the lock and kill the patch. There can't be any contention during
io_ring_ctx_freeanyway.Hi Pavel, I don't get it, do you mean this patch isn't needed, and we can just
unlock(&uring_lock)beforeio_run_task_work_sig()andlock(&uring_lock)after it? I knew there won't be contention duringio_ring_ctx_freethat's why there is nouring_lockin it. I tried to just dounlock(&uring_lock)beforeio_run_task_sig()without if(locked) check, it reports something like "there are unpaired mutex lock/unlock" since we cannot just unlock if it's fromio_ring_ctx_free.The ugly part is @locked. I know that there is already similar stuff around, but I may go long why and how much I don't like it.
io_ring_ctx_free(){ ... lock(uring_lock); files_unregister(); unlock(uring_lock); ... }
Quoted 3 lines
With this you'll always have the mutex locked in unregister, so can drop it unconditionally (if that will ever be needed). It's also cleaner from the synchronisation perspective.
Yes that's a much better approach - passing around a 'locked' flag is not very pretty, and the fewer we have of those the better.
Signature · 1 line
Jens Axboe
PATCH 2/2 io_uring: don't hold uring_lock when calling io_run_task_work*
In reply to Hao XuAbaci reported the below issue:
[ 141.400455] hrtimer: interrupt took 205853 ns
[ 189.869316] process 'usr/local/ilogtail/ilogtail_0.16.26' started with executable stack
[ 250.188042]
[ 250.188327] ============================================
[ 250.189015] WARNING: possible recursive locking detected
[ 250.189732] 5.11.0-rc4 #1 Not tainted
[ 250.190267] --------------------------------------------
[ 250.190917] a.out/7363 is trying to acquire lock:
[ 250.191506] ffff888114dbcbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __io_req_task_submit+0x29/0xa0
[ 250.192599]
[ 250.192599] but task is already holding lock:
[ 250.193309] ffff888114dbfbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_register+0xad/0x210
[ 250.194426]
[ 250.194426] other info that might help us debug this:
[ 250.195238] Possible unsafe locking scenario:
[ 250.195238]
[ 250.196019] CPU0
[ 250.196411] ----
[ 250.196803] lock(&ctx->uring_lock);
[ 250.197420] lock(&ctx->uring_lock);
[ 250.197966]
[ 250.197966] *** DEADLOCK ***
[ 250.197966]
[ 250.198837] May be due to missing lock nesting notation
[ 250.198837]
[ 250.199780] 1 lock held by a.out/7363:
[ 250.200373] #0: ffff888114dbfbe8 (&ctx->uring_lock){+.+.}-{3:3}, at: __x64_sys_io_uring_register+0xad/0x210
[ 250.201645]
[ 250.201645] stack backtrace:
[ 250.202298] CPU: 0 PID: 7363 Comm: a.out Not tainted 5.11.0-rc4 #1
[ 250.203144] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011
[ 250.203887] Call Trace:
[ 250.204302] dump_stack+0xac/0xe3
[ 250.204804] __lock_acquire+0xab6/0x13a0
[ 250.205392] lock_acquire+0x2c3/0x390
[ 250.205928] ? __io_req_task_submit+0x29/0xa0
[ 250.206541] __mutex_lock+0xae/0x9f0
[ 250.207071] ? __io_req_task_submit+0x29/0xa0
[ 250.207745] ? 0xffffffffa0006083
[ 250.208248] ? __io_req_task_submit+0x29/0xa0
[ 250.208845] ? __io_req_task_submit+0x29/0xa0
[ 250.209452] ? __io_req_task_submit+0x5/0xa0
[ 250.210083] __io_req_task_submit+0x29/0xa0
[ 250.210687] io_async_task_func+0x23d/0x4c0
[ 250.211278] task_work_run+0x89/0xd0
[ 250.211884] io_run_task_work_sig+0x50/0xc0
[ 250.212464] io_sqe_files_unregister+0xb2/0x1f0
[ 250.213109] __io_uring_register+0x115a/0x1750
[ 250.213718] ? __x64_sys_io_uring_register+0xad/0x210
[ 250.214395] ? __fget_files+0x15a/0x260
[ 250.214956] __x64_sys_io_uring_register+0xbe/0x210
[ 250.215620] ? trace_hardirqs_on+0x46/0x110
[ 250.216205] do_syscall_64+0x2d/0x40
[ 250.216731] entry_SYSCALL_64_after_hwframe+0x44/0xa9
[ 250.217455] RIP: 0033:0x7f0fa17e5239
[ 250.218034] Code: 01 00 48 81 c4 80 00 00 00 e9 f1 fe ff ff 0f 1f 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 3d 01 f0 ff ff 73 01 c3 48 8b 0d 27 ec 2c 00 f7 d8 64 89 01 48
[ 250.220343] RSP: 002b:00007f0fa1eeac48 EFLAGS: 00000246 ORIG_RAX: 00000000000001ab
[ 250.221360] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f0fa17e5239
[ 250.222272] RDX: 0000000000000000 RSI: 0000000000000003 RDI: 0000000000000008
[ 250.223185] RBP: 00007f0fa1eeae20 R08: 0000000000000000 R09: 0000000000000000
[ 250.224091] R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000
[ 250.224999] R13: 0000000000021000 R14: 0000000000000000 R15: 00007f0fa1eeb700This is caused by calling io_run_task_work_sig() to do work under uring_lock while the caller io_sqe_files_unregister() already held uring_lock.
we need to check if uring_lock is held by us when doing unlock around io_run_task_work_sig() since there are code paths down to that place without uring_lock held.
--- fs/io_uring.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)
fs/io_uring.c +13 -6
@@ -7362,18 +7362,25 @@ static int io_sqe_files_unregister(struct io_ring_ctx *ctx, bool locked)/* wait for all refs nodes to complete */ flush_delayed_work (&ctx->file_put_work); +if (locked) +mutex_unlock (&ctx->uring_lock);do { ret =wait_for_completion_interruptible (&data->done);if (!ret)break ; ret =io_run_task_work_sig (); -if (ret <0 ) { -percpu_ref_resurrect (&data->refs); -reinit_completion (&data->done); -io_sqe_files_set_node (data, backup_node); -return ret; - } +if (ret <0 ) +break ; }while (1 ); +if (locked) +mutex_lock (&ctx->uring_lock); + +if (ret <0 ) { +percpu_ref_resurrect (&data->refs); +reinit_completion (&data->done); +io_sqe_files_set_node (data, backup_node); +return ret; + }__io_sqe_files_unregister (ctx); nr_tables =DIV_ROUND_UP (ctx->nr_user_files,IORING_MAX_FILES_TABLE );
Signature · 1 line
1.8.3.1
Re: [PATCH 2/2] io_uring: don't hold uring_lock when calling io_run_task_work*
Quoted 6 lines · Hao Xu
This is caused by calling
io_run_task_work_sig()to do work underuring_lockwhile the callerio_sqe_files_unregister()already helduring_lock.
we need to check ifuring_lockis held by us when doing unlock aroundio_run_task_work_sig()since there are code paths down to that place withouturing_lockheld.
1. we don't want to allow parallel io_sqe_files_unregister()s happening, it's synchronised by uring_lock atm. Otherwise it's buggy.
2. probably same with unregister and submit.
Quoted 45 lines
Reported-by Abaci <[email protected]>Fixes 1ffc54220c44 ("io_uring: fix io_sqe_files_unregister() hangs")Signed-off-by Hao Xu <[email protected]>--- fs/io_uring.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)fs/io_uring.c +13
-6@@ -7362,18 +7362,25 @@ static int io_sqe_files_unregister(struct io_ring_ctx *ctx, bool locked)/* wait for all refs nodes to complete */ flush_delayed_work (&ctx->file_put_work); +if (locked) +mutex_unlock (&ctx->uring_lock);do { ret =wait_for_completion_interruptible (&data->done);if (!ret)break ; ret =io_run_task_work_sig (); -if (ret <0 ) { -percpu_ref_resurrect (&data->refs); -reinit_completion (&data->done); -io_sqe_files_set_node (data, backup_node); -return ret; - } +if (ret <0 ) +break ; }while (1 ); +if (locked) +mutex_lock (&ctx->uring_lock); + +if (ret <0 ) { +percpu_ref_resurrect (&data->refs); +reinit_completion (&data->done); +io_sqe_files_set_node (data, backup_node); +return ret; + }__io_sqe_files_unregister (ctx); nr_tables =DIV_ROUND_UP (ctx->nr_user_files,IORING_MAX_FILES_TABLE );
Signature · 1 line
Pavel Begunkov
Re: [PATCH 2/2] io_uring: don't hold uring_lock when calling io_run_task_work*
Quoted 11 lines · Pavel Begunkov
Quoted 6 lines · Hao Xu
This is caused by calling
io_run_task_work_sig()to do work underuring_lockwhile the callerio_sqe_files_unregister()already helduring_lock.
we need to check ifuring_lockis held by us when doing unlock aroundio_run_task_work_sig()since there are code paths down to that place withouturing_lockheld.1. we don't want to allow parallel io_sqe_files_unregister()s happening, it's synchronised by
uring_lockatm. Otherwise it's buggy.
This one should be simple, alike to
if (percpu_refs_is_dying())
return error; // fail *files_unregister();
Quoted 49 lines
2. probably same with unregister and submit.
Quoted 45 lines
Reported-by Abaci <[email protected]>Fixes 1ffc54220c44 ("io_uring: fix io_sqe_files_unregister() hangs")Signed-off-by Hao Xu <[email protected]>--- fs/io_uring.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)fs/io_uring.c +13
-6@@ -7362,18 +7362,25 @@ static int io_sqe_files_unregister(struct io_ring_ctx *ctx, bool locked)/* wait for all refs nodes to complete */ flush_delayed_work (&ctx->file_put_work); +if (locked) +mutex_unlock (&ctx->uring_lock);do { ret =wait_for_completion_interruptible (&data->done);if (!ret)break ; ret =io_run_task_work_sig (); -if (ret <0 ) { -percpu_ref_resurrect (&data->refs); -reinit_completion (&data->done); -io_sqe_files_set_node (data, backup_node); -return ret; - } +if (ret <0 ) +break ; }while (1 ); +if (locked) +mutex_lock (&ctx->uring_lock); + +if (ret <0 ) { +percpu_ref_resurrect (&data->refs); +reinit_completion (&data->done); +io_sqe_files_set_node (data, backup_node); +return ret; + }__io_sqe_files_unregister (ctx); nr_tables =DIV_ROUND_UP (ctx->nr_user_files,IORING_MAX_FILES_TABLE );
Signature · 1 line
Pavel Begunkov
Re: [PATCH 2/2] io_uring: don't hold uring_lock when calling io_run_task_work*
Quoted 9 lines · Pavel Begunkov
Quoted 8 lines · Pavel Begunkov
Quoted 6 lines · Hao Xu
This is caused by calling
io_run_task_work_sig()to do work underuring_lockwhile the callerio_sqe_files_unregister()already helduring_lock.
we need to check ifuring_lockis held by us when doing unlock aroundio_run_task_work_sig()since there are code paths down to that place withouturing_lockheld.
Quoted 3 lines
Quoted 3 lines
1. we don't want to allow parallel io_sqe_files_unregister()s happening, it's synchronised by
uring_lockatm. Otherwise it's buggy.
Here "since there are code paths down to that place without uring_lock held" I mean code path of io_ring_ctx_free().
Quoted 56 lines
This one should be simple, alike to
if (
percpu_refs_is_dying())return error; // fail *files_unregister();Quoted 49 lines
2. probably same with unregister and submit.
Quoted 45 lines
Reported-by Abaci <[email protected]>Fixes 1ffc54220c44 ("io_uring: fix io_sqe_files_unregister() hangs")Signed-off-by Hao Xu <[email protected]>--- fs/io_uring.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)fs/io_uring.c +13
-6@@ -7362,18 +7362,25 @@ static int io_sqe_files_unregister(struct io_ring_ctx *ctx, bool locked)/* wait for all refs nodes to complete */ flush_delayed_work (&ctx->file_put_work); +if (locked) +mutex_unlock (&ctx->uring_lock);do { ret =wait_for_completion_interruptible (&data->done);if (!ret)break ; ret =io_run_task_work_sig (); -if (ret <0 ) { -percpu_ref_resurrect (&data->refs); -reinit_completion (&data->done); -io_sqe_files_set_node (data, backup_node); -return ret; - } +if (ret <0 ) +break ; }while (1 ); +if (locked) +mutex_lock (&ctx->uring_lock); + +if (ret <0 ) { +percpu_ref_resurrect (&data->refs); +reinit_completion (&data->done); +io_sqe_files_set_node (data, backup_node); +return ret; + }__io_sqe_files_unregister (ctx); nr_tables =DIV_ROUND_UP (ctx->nr_user_files,IORING_MAX_FILES_TABLE );
Re: [PATCH 2/2] io_uring: don't hold uring_lock when calling io_run_task_work*
Quoted 10 lines · Hao Xu
Quoted 9 lines · Pavel Begunkov
Quoted 8 lines · Pavel Begunkov
Quoted 6 lines · Hao Xu
This is caused by calling
io_run_task_work_sig()to do work underuring_lockwhile the callerio_sqe_files_unregister()already helduring_lock.
we need to check ifuring_lockis held by us when doing unlock aroundio_run_task_work_sig()since there are code paths down to that place withouturing_lockheld.
Quoted 4 lines
Quoted 3 lines
Quoted 3 lines
1. we don't want to allow parallel io_sqe_files_unregister()s happening, it's synchronised by
uring_lockatm. Otherwise it's buggy.Here "since there are code paths down to that place without
uring_lockheld" I mean code path ofio_ring_ctx_free().
I guess it's to the 1/2, but let me outline the problem again:
if you have two tasks userspace threads sharing a ring, then they can both and in parallel call syscall:files_unregeister. That's a potential double percpu_ref_kill(&data->refs), or even worse.
Same for 2, but racing for the table and refs.
Quoted 55 lines
Quoted 54 lines
This one should be simple, alike to
if (
percpu_refs_is_dying())
return error; // fail*files_unregister();Quoted 47 lines
2. probably same with unregister and submit.
Quoted 43 lines
Reported-by Abaci <[email protected]>Fixes 1ffc54220c44 ("io_uring: fix io_sqe_files_unregister() hangs")Signed-off-by Hao Xu <[email protected]>---
fs/io_uring.c| 19 +++++++++++++------1 file changed, 13 insertions(+), 6 deletions(-)fs/io_uring.c +13
-6@@ -7362,18 +7362,25 @@ static int io_sqe_files_unregister(struct io_ring_ctx *ctx, bool locked) /* wait for all refs nodes to complete */ flush_delayed_work(&ctx->file_put_work); + if (locked) + mutex_unlock(&ctx->uring_lock); do { ret = wait_for_completion_interruptible(&data->done); if (!ret) break; ret = io_run_task_work_sig(); - if (ret < 0) { - percpu_ref_resurrect(&data->refs); - reinit_completion(&data->done); - io_sqe_files_set_node(data, backup_node); - return ret; - } + if (ret < 0) + break; } while (1); + if (locked) + mutex_lock(&ctx->uring_lock); + + if (ret < 0) { + percpu_ref_resurrect(&data->refs); + reinit_completion(&data->done); + io_sqe_files_set_node(data, backup_node); + return ret; + } __io_sqe_files_unregister(ctx); nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE);
Signature · 1 line
Pavel Begunkov
Re: [PATCH 2/2] io_uring: don't hold uring_lock when calling io_run_task_work*
Quoted 14 lines · Pavel Begunkov
Quoted 13 lines · Hao Xu
Quoted 12 lines · Pavel Begunkov
Quoted 11 lines · Pavel Begunkov
Quoted 6 lines · Hao Xu
This is caused by calling
io_run_task_work_sig()to do work underuring_lockwhile the callerio_sqe_files_unregister()already helduring_lock.
we need to check ifuring_lockis held by us when doing unlock aroundio_run_task_work_sig()since there are code paths down to that place withouturing_lockheld.1. we don't want to allow parallel io_sqe_files_unregister()s happening, it's synchronised by
uring_lockatm. Otherwise it's buggy.
Quoted 8 lines
Quoted 1 line
Here "since there are code paths down to that place without
uring_lockheld" I mean code path ofio_ring_ctx_free().I guess it's to the 1/2, but let me outline the problem again:
if you have two tasks userspace threads sharing a ring, then they can both and in parallel call syscall:files_unregeister. That's a potential doublepercpu_ref_kill(&data->refs), or even worse.Same for 2, but racing for the table and refs.
There is a couple of thoughts for this:
1. I don't like waiting without holding the lock in general, because someone can submit more reqs in-between and so indefinitely postponing the files_unregister.
2. I wouldn't want to add checks for that in submission path.
So, a solution I think about is to wait under the lock, If we need to run task_works -- briefly drop the lock, run task_works and then do all unregister all over again. Keep an eye on refs, e.g. probably need to resurrect it.
Because we current task is busy nobody submits new requests on its behalf, and so there can't be infinite number of in-task_work reqs, and eventually it will just go wait/sleep forever (if not signalled) under the mutex, so we can a kind of upper bound on time.
Quoted 57 lines
Quoted 56 lines
Quoted 55 lines
This one should be simple, alike to
if (
percpu_refs_is_dying())
return error; // fail*files_unregister();Quoted 47 lines
2. probably same with unregister and submit.
Quoted 43 lines
Reported-by Abaci <[email protected]>Fixes 1ffc54220c44 ("io_uring: fix io_sqe_files_unregister() hangs")Signed-off-by Hao Xu <[email protected]>---
fs/io_uring.c| 19 +++++++++++++------1 file changed, 13 insertions(+), 6 deletions(-)fs/io_uring.c +13
-6@@ -7362,18 +7362,25 @@ static int io_sqe_files_unregister(struct io_ring_ctx *ctx, bool locked) /* wait for all refs nodes to complete */ flush_delayed_work(&ctx->file_put_work); + if (locked) + mutex_unlock(&ctx->uring_lock); do { ret = wait_for_completion_interruptible(&data->done); if (!ret) break; ret = io_run_task_work_sig(); - if (ret < 0) { - percpu_ref_resurrect(&data->refs); - reinit_completion(&data->done); - io_sqe_files_set_node(data, backup_node); - return ret; - } + if (ret < 0) + break; } while (1); + if (locked) + mutex_lock(&ctx->uring_lock); + + if (ret < 0) { + percpu_ref_resurrect(&data->refs); + reinit_completion(&data->done); + io_sqe_files_set_node(data, backup_node); + return ret; + } __io_sqe_files_unregister(ctx); nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE);
Signature · 1 line
Pavel Begunkov
Re: [PATCH 2/2] io_uring: don't hold uring_lock when calling io_run_task_work*
Quoted 24 lines · Pavel Begunkov
Quoted 21 lines · Pavel Begunkov
Quoted 10 lines · Hao Xu
Quoted 9 lines · Pavel Begunkov
Quoted 8 lines · Pavel Begunkov
Quoted 6 lines · Hao Xu
This is caused by calling
io_run_task_work_sig()to do work underuring_lockwhile the callerio_sqe_files_unregister()already helduring_lock.
we need to check ifuring_lockis held by us when doing unlock aroundio_run_task_work_sig()since there are code paths down to that place withouturing_lockheld.Quoted 4 lines
Quoted 3 lines
Quoted 3 lines
1. we don't want to allow parallel io_sqe_files_unregister()s happening, it's synchronised by
uring_lockatm. Otherwise it's buggy.Here "since there are code paths down to that place without
uring_lockheld" I mean code path ofio_ring_ctx_free().I guess it's to the 1/2, but let me outline the problem again:
if you have two tasks userspace threads sharing a ring, then they can both and in parallel call syscall:files_unregeister. That's a potential doublepercpu_ref_kill(&data->refs), or even worse.
Quoted 7 lines
Quoted 1 line
Same for 2, but racing for the table and refs.
There is a couple of thoughts for this:
1. I don't like waiting without holding the lock in general, because someone can submit more reqs in-between and so indefinitely postponing the
files_unregister.
Thanks, Pavel.
I thought this issue before, until I saw this in __io_uring_register:
if (io_register_op_must_quiesce(opcode)) {
percpu_ref_kill(&ctx->refs); /*
¦* Drop uring mutex before waiting for references to exit. If
¦* another thread is currently inside io_uring_enter() it might
¦* need to grab the uring_lock to make progress. If we hold it
¦* here across the drain wait, then we can deadlock. It's safe
¦* to drop the mutex here, since no new references will come in
¦* after we've killed the percpu ref.
¦*/
mutex_unlock(&ctx->uring_lock);
do {
ret = wait_for_completion_interruptible(&ctx->ref_comp);
if (!ret)
break;
ret = io_run_task_work_sig();
if (ret < 0)
break;
} while (1);
mutex_lock(&ctx->uring_lock);
if (ret) {
percpu_ref_resurrect(&ctx->refs);
goto out_quiesce;
}}
So now I guess the postponement issue also exits in the above code since there could be another thread submiting reqs to the shared ctx(or we can say uring fd).
Quoted 7 lines
2. I wouldn't want to add checks for that in submission path.
So, a solution I think about is to wait under the lock, If we need to run
task_works-- briefly drop the lock, runtask_worksand then do all unregister all over again. Keep an eye on refs, e.g. probably need to resurrect it.
Quoted 6 lines
Because we current task is busy nobody submits new requests on its behalf, and so there can't be infinite number of in-task_work reqs, and eventually it will just go wait/sleep forever (if not signalled) under the mutex, so we can a kind of upper bound on time.
Do you mean sleeping with timeout rather than just sleeping? I think this works, I'll work on this and think about the detail. But before addressing this issue, Should I first send a patch to just fix the deadlock issue?
Thanks,
Hao
Quoted 58 lines
Quoted 57 lines
Quoted 56 lines
Quoted 55 lines
This one should be simple, alike to
if (
percpu_refs_is_dying())
return error; // fail*files_unregister();Quoted 47 lines
2. probably same with unregister and submit.
Quoted 43 lines
Reported-by Abaci <[email protected]>Fixes 1ffc54220c44 ("io_uring: fix io_sqe_files_unregister() hangs")Signed-off-by Hao Xu <[email protected]>--- fs/io_uring.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)fs/io_uring.c +13
-6@@ -7362,18 +7362,25 @@ static int io_sqe_files_unregister(struct io_ring_ctx *ctx, bool locked) /* wait for all refs nodes to complete */ flush_delayed_work(&ctx->file_put_work); + if (locked) + mutex_unlock(&ctx->uring_lock); do { ret = wait_for_completion_interruptible(&data->done); if (!ret) break; ret = io_run_task_work_sig(); - if (ret < 0) { - percpu_ref_resurrect(&data->refs); - reinit_completion(&data->done); - io_sqe_files_set_node(data, backup_node); - return ret; - } + if (ret < 0) + break; } while (1); + if (locked) + mutex_lock(&ctx->uring_lock); + + if (ret < 0) { + percpu_ref_resurrect(&data->refs); + reinit_completion(&data->done); + io_sqe_files_set_node(data, backup_node); + return ret; + } __io_sqe_files_unregister(ctx); nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE);
Re: [PATCH 2/2] io_uring: don't hold uring_lock when calling io_run_task_work*
Quoted 27 lines · Hao Xu
Quoted 26 lines · Pavel Begunkov
Quoted 14 lines · Pavel Begunkov
Quoted 13 lines · Hao Xu
Quoted 12 lines · Pavel Begunkov
Quoted 11 lines · Pavel Begunkov
Quoted 6 lines · Hao Xu
This is caused by calling
io_run_task_work_sig()to do work underuring_lockwhile the callerio_sqe_files_unregister()already helduring_lock.
we need to check ifuring_lockis held by us when doing unlock aroundio_run_task_work_sig()since there are code paths down to that place withouturing_lockheld.1. we don't want to allow parallel io_sqe_files_unregister()s happening, it's synchronised by
uring_lockatm. Otherwise it's buggy.Quoted 8 lines
Quoted 1 line
Here "since there are code paths down to that place without
uring_lockheld" I mean code path ofio_ring_ctx_free().I guess it's to the 1/2, but let me outline the problem again:
if you have two tasks userspace threads sharing a ring, then they can both and in parallel call syscall:files_unregeister. That's a potential doublepercpu_ref_kill(&data->refs), or even worse.Same for 2, but racing for the table and refs.
There is a couple of thoughts for this:
Quoted 8 lines
Quoted 3 lines
1. I don't like waiting without holding the lock in general, because someone can submit more reqs in-between and so indefinitely postponing the
files_unregister.Thanks, Pavel.
I thought this issue before, until I saw this in__io_uring_register:if (
io_register_op_must_quiesce(opcode)) {
percpu_ref_kill(&ctx->refs);
It is different because of this kill, it will prevent submissions.
Quoted 37 lines
/*
¦* Drop uring mutex before waiting for references to exit. If ¦* another thread is currently insideio_uring_enter()it might ¦* need to grab theuring_lockto make progress. If we hold it ¦* here across the drain wait, then we can deadlock. It's safe ¦* to drop the mutex here, since no new references will come in ¦* after we've killed the percpu ref.
¦*/
mutex_unlock(&ctx->uring_lock);
do {
ret =wait_for_completion_interruptible(&ctx->ref_comp); if (!ret)
break;
ret =io_run_task_work_sig();
if (ret < 0)
break;
} while (1);
mutex_lock(&ctx->uring_lock);if (ret) {
percpu_ref_resurrect(&ctx->refs); gotoout_quiesce;
}
}So now I guess the postponement issue also exits in the above code since there could be another thread submiting reqs to the shared ctx(or we can say uring fd).
Quoted 7 lines
2. I wouldn't want to add checks for that in submission path.
So, a solution I think about is to wait under the lock, If we need to run
task_works-- briefly drop the lock, runtask_worksand then do all unregister all over again. Keep an eye on refs, e.g. probably need to resurrect it.
Quoted 7 lines
Quoted 6 lines
Because we current task is busy nobody submits new requests on its behalf, and so there can't be infinite number of in-task_work reqs, and eventually it will just go wait/sleep forever (if not signalled) under the mutex, so we can a kind of upper bound on time.
Do you mean sleeping with timeout rather than just sleeping? I think this works, I'll work on this and think about the detail.
Without timeout -- it will be awaken when new task_works are coming in, but Jens knows better.
Quoted 1 line
But before addressing this issue, Should I first send a patch to just fix the deadlock issue?
Do you mean the deadlock 2/2 was trying to fix? Or some else? The thread is all about fixing it, but doing it right. Not sure there is a need for faster but incomplete solution, if that's what you meant.
Signature · 1 line
Pavel Begunkov
Re: [PATCH 2/2] io_uring: don't hold uring_lock when calling io_run_task_work*
Quoted 31 lines · Pavel Begunkov
Quoted 30 lines · Hao Xu
Quoted 29 lines · Pavel Begunkov
Quoted 14 lines · Pavel Begunkov
Quoted 13 lines · Hao Xu
Quoted 12 lines · Pavel Begunkov
Quoted 11 lines · Pavel Begunkov
Quoted 6 lines · Hao Xu
This is caused by calling
io_run_task_work_sig()to do work underuring_lockwhile the callerio_sqe_files_unregister()already helduring_lock.
we need to check ifuring_lockis held by us when doing unlock aroundio_run_task_work_sig()since there are code paths down to that place withouturing_lockheld.1. we don't want to allow parallel io_sqe_files_unregister()s happening, it's synchronised by
uring_lockatm. Otherwise it's buggy.Quoted 8 lines
Quoted 1 line
Here "since there are code paths down to that place without
uring_lockheld" I mean code path ofio_ring_ctx_free().I guess it's to the 1/2, but let me outline the problem again:
if you have two tasks userspace threads sharing a ring, then they can both and in parallel call syscall:files_unregeister. That's a potential doublepercpu_ref_kill(&data->refs), or even worse.Same for 2, but racing for the table and refs.
There is a couple of thoughts for this:
1. I don't like waiting without holding the lock in general, because someone can submit more reqs in-between and so indefinitely postponing the
files_unregister.
Quoted 8 lines
Quoted 5 lines
Thanks, Pavel.
I thought this issue before, until I saw this in__io_uring_register:if (
io_register_op_must_quiesce(opcode)) {
percpu_ref_kill(&ctx->refs);It is different because of this kill, it will prevent submissions.
I saw percpu_ref_is_dying(&ctx->refs) check in sq thread but not in io_uring_enter(), so I guess there could be another thread doing io_uring_enter() and submiting sqes.
Quoted 37 lines
Quoted 37 lines
/*
¦* Drop uring mutex before waiting for references to exit. If ¦* another thread is currently insideio_uring_enter()it might ¦* need to grab theuring_lockto make progress. If we hold it ¦* here across the drain wait, then we can deadlock. It's safe ¦* to drop the mutex here, since no new references will come in ¦* after we've killed the percpu ref.
¦*/
mutex_unlock(&ctx->uring_lock);
do {
ret =wait_for_completion_interruptible(&ctx->ref_comp); if (!ret)
break;
ret =io_run_task_work_sig();
if (ret < 0)
break;
} while (1);
mutex_lock(&ctx->uring_lock);if (ret) {
percpu_ref_resurrect(&ctx->refs); gotoout_quiesce;
}
}So now I guess the postponement issue also exits in the above code since there could be another thread submiting reqs to the shared ctx(or we can say uring fd).
Quoted 7 lines
2. I wouldn't want to add checks for that in submission path.
So, a solution I think about is to wait under the lock, If we need to run
task_works-- briefly drop the lock, runtask_worksand then do all unregister all over again. Keep an eye on refs, e.g. probably need to resurrect it.
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Because we current task is busy nobody submits new requests on its behalf, and so there can't be infinite number of in-task_work reqs, and eventually it will just go wait/sleep forever (if not signalled) under the mutex, so we can a kind of upper bound on time.
Sorry Pavel, I don't quiet understand "so we can a kind of upper bound on time". :(
Quoted 5 lines
Quoted 2 lines
Quoted 1 line
Do you mean sleeping with timeout rather than just sleeping? I think this works, I'll work on this and think about the detail.
Without timeout -- it will be awaken when new
task_worksare coming in, but Jens knows better.
So we can just put unlock and lock around io_run_task_work_sig() to address the issue 2?
Quoted 7 lines
Quoted 2 lines
But before addressing this issue, Should I first send a patch to just fix the deadlock issue?
Do you mean the deadlock 2/2 was trying to fix? Or some else? The thread is all about fixing it, but doing it right. Not sure there is a need for faster but incomplete solution, if that's what you meant.
Re: [PATCH 2/2] io_uring: don't hold uring_lock when calling io_run_task_work*
Quoted 3 lines · Hao Xu
I saw
percpu_ref_is_dying(&ctx->refs)check in sq thread but not inio_uring_enter(), so I guess there could be another thread doingio_uring_enter()and submiting sqes.
Ok, looks tryget does it in a different fashion than I recalled.
Quoted 2 lines
So we can just put unlock and lock around
io_run_task_work_sig()to address the issue 2?
io_set_resource_node()
-> percpu_ref_get()
If refs are left dying, it won't be safe for in-parallel enters. tryget or resurrect
Signature · 1 line
Pavel Begunkov
Re: [PATCH 2/2] io_uring: don't hold uring_lock when calling io_run_task_work*
In reply to Pavel BegunkovQuoted 82 lines · Pavel Begunkov
Quoted 27 lines · Hao Xu
Quoted 26 lines · Pavel Begunkov
Quoted 14 lines · Pavel Begunkov
Quoted 13 lines · Hao Xu
Quoted 12 lines · Pavel Begunkov
Quoted 11 lines · Pavel Begunkov
Quoted 6 lines · Hao Xu
This is caused by calling
io_run_task_work_sig()to do work underuring_lockwhile the callerio_sqe_files_unregister()already helduring_lock.
we need to check ifuring_lockis held by us when doing unlock aroundio_run_task_work_sig()since there are code paths down to that place withouturing_lockheld.1. we don't want to allow parallel io_sqe_files_unregister()s happening, it's synchronised by
uring_lockatm. Otherwise it's buggy.Quoted 8 lines
Quoted 1 line
Here "since there are code paths down to that place without
uring_lockheld" I mean code path ofio_ring_ctx_free().I guess it's to the 1/2, but let me outline the problem again:
if you have two tasks userspace threads sharing a ring, then they can both and in parallel call syscall:files_unregeister. That's a potential doublepercpu_ref_kill(&data->refs), or even worse.Same for 2, but racing for the table and refs.
There is a couple of thoughts for this:
Quoted 8 lines
Quoted 3 lines
1. I don't like waiting without holding the lock in general, because someone can submit more reqs in-between and so indefinitely postponing the
files_unregister.Thanks, Pavel.
I thought this issue before, until I saw this in__io_uring_register:if (
io_register_op_must_quiesce(opcode)) {
percpu_ref_kill(&ctx->refs);It is different because of this kill, it will prevent submissions.
Quoted 43 lines
/*
¦* Drop uring mutex before waiting for references to exit. If ¦* another thread is currently insideio_uring_enter()it might ¦* need to grab theuring_lockto make progress. If we hold it ¦* here across the drain wait, then we can deadlock. It's safe ¦* to drop the mutex here, since no new references will come in ¦* after we've killed the percpu ref.
¦*/
mutex_unlock(&ctx->uring_lock);
do {
ret =wait_for_completion_interruptible(&ctx->ref_comp); if (!ret)
break;
ret =io_run_task_work_sig();
if (ret < 0)
break;
} while (1);
mutex_lock(&ctx->uring_lock);if (ret) {
percpu_ref_resurrect(&ctx->refs); gotoout_quiesce;
}
}So now I guess the postponement issue also exits in the above code since there could be another thread submiting reqs to the shared ctx(or we can say uring fd).
Quoted 13 lines
2. I wouldn't want to add checks for that in submission path.
So, a solution I think about is to wait under the lock, If we need to run
task_works-- briefly drop the lock, runtask_worksand then do all unregister all over again. Keep an eye on refs, e.g. probably need to resurrect it.Because we current task is busy nobody submits new requests on its behalf, and so there can't be infinite number of in-task_work reqs, and eventually it will just go wait/sleep forever (if not signalled) under the mutex, so we can a kind of upper bound on time.
Quoted 5 lines
Quoted 1 line
Do you mean sleeping with timeout rather than just sleeping? I think this works, I'll work on this and think about the detail.
Without timeout -- it will be awaken when new
task_worksare coming in, but Jens knows better.
Ah, gotcha, yes, I think it's because inserting a task_work raises a signal.
Quoted 6 lines
Quoted 1 line
But before addressing this issue, Should I first send a patch to just fix the deadlock issue?
Do you mean the deadlock 2/2 was trying to fix? Or some else? The thread is all about fixing it, but doing it right. Not sure there is a need for faster but incomplete solution, if that's what you meant.
Re: [PATCH 2/2] io_uring: don't hold uring_lock when calling io_run_task_work*
In reply to Pavel BegunkovQuoted 82 lines · Pavel Begunkov
Quoted 27 lines · Hao Xu
Quoted 26 lines · Pavel Begunkov
Quoted 14 lines · Pavel Begunkov
Quoted 13 lines · Hao Xu
Quoted 12 lines · Pavel Begunkov
Quoted 11 lines · Pavel Begunkov
Quoted 6 lines · Hao Xu
This is caused by calling
io_run_task_work_sig()to do work underuring_lockwhile the callerio_sqe_files_unregister()already helduring_lock.
we need to check ifuring_lockis held by us when doing unlock aroundio_run_task_work_sig()since there are code paths down to that place withouturing_lockheld.1. we don't want to allow parallel io_sqe_files_unregister()s happening, it's synchronised by
uring_lockatm. Otherwise it's buggy.Quoted 8 lines
Quoted 1 line
Here "since there are code paths down to that place without
uring_lockheld" I mean code path ofio_ring_ctx_free().I guess it's to the 1/2, but let me outline the problem again:
if you have two tasks userspace threads sharing a ring, then they can both and in parallel call syscall:files_unregeister. That's a potential doublepercpu_ref_kill(&data->refs), or even worse.Same for 2, but racing for the table and refs.
There is a couple of thoughts for this:
Quoted 8 lines
Quoted 3 lines
1. I don't like waiting without holding the lock in general, because someone can submit more reqs in-between and so indefinitely postponing the
files_unregister.Thanks, Pavel.
I thought this issue before, until I saw this in__io_uring_register:if (
io_register_op_must_quiesce(opcode)) {
percpu_ref_kill(&ctx->refs);It is different because of this kill, it will prevent submissions.
Quoted 43 lines
/*
¦* Drop uring mutex before waiting for references to exit. If ¦* another thread is currently insideio_uring_enter()it might ¦* need to grab theuring_lockto make progress. If we hold it ¦* here across the drain wait, then we can deadlock. It's safe ¦* to drop the mutex here, since no new references will come in ¦* after we've killed the percpu ref.
¦*/
mutex_unlock(&ctx->uring_lock);
do {
ret =wait_for_completion_interruptible(&ctx->ref_comp); if (!ret)
break;
ret =io_run_task_work_sig();
if (ret < 0)
break;
} while (1);
mutex_lock(&ctx->uring_lock);if (ret) {
percpu_ref_resurrect(&ctx->refs); gotoout_quiesce;
}
}So now I guess the postponement issue also exits in the above code since there could be another thread submiting reqs to the shared ctx(or we can say uring fd).
Quoted 13 lines
2. I wouldn't want to add checks for that in submission path.
So, a solution I think about is to wait under the lock, If we need to run
task_works-- briefly drop the lock, runtask_worksand then do all unregister all over again. Keep an eye on refs, e.g. probably need to resurrect it.Because we current task is busy nobody submits new requests on its behalf, and so there can't be infinite number of in-task_work reqs, and eventually it will just go wait/sleep forever (if not signalled) under the mutex, so we can a kind of upper bound on time.
Quoted 4 lines
Quoted 1 line
Do you mean sleeping with timeout rather than just sleeping? I think this works, I'll work on this and think about the detail.
Without timeout -- it will be awaken when new
task_worksare coming in, but Jens knows better.
Isn't this risky, since there could possible be no more task_works coming in, which causes it sleeps forever?
Quoted 7 lines
Quoted 2 lines
But before addressing this issue, Should I first send a patch to just fix the deadlock issue?
Do you mean the deadlock 2/2 was trying to fix? Or some else? The thread is all about fixing it, but doing it right. Not sure there is a need for faster but incomplete solution, if that's what you meant.
Re: [PATCH 2/2] io_uring: don't hold uring_lock when calling io_run_task_work*
In reply to Pavel BegunkovQuoted 9 lines · Pavel Begunkov
Quoted 8 lines · Pavel Begunkov
Quoted 6 lines · Hao Xu
This is caused by calling
io_run_task_work_sig()to do work underuring_lockwhile the callerio_sqe_files_unregister()already helduring_lock.
we need to check ifuring_lockis held by us when doing unlock aroundio_run_task_work_sig()since there are code paths down to that place withouturing_lockheld.
Quoted 9 lines
Quoted 3 lines
1. we don't want to allow parallel io_sqe_files_unregister()s happening, it's synchronised by
uring_lockatm. Otherwise it's buggy.This one should be simple, alike to
if (
percpu_refs_is_dying())return error; // fail *files_unregister();
Agree.
Quoted 50 lines
Quoted 49 lines
2. probably same with unregister and submit.
Quoted 45 lines
Reported-by Abaci <[email protected]>Fixes 1ffc54220c44 ("io_uring: fix io_sqe_files_unregister() hangs")Signed-off-by Hao Xu <[email protected]>--- fs/io_uring.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)fs/io_uring.c +13
-6@@ -7362,18 +7362,25 @@ static int io_sqe_files_unregister(struct io_ring_ctx *ctx, bool locked)/* wait for all refs nodes to complete */ flush_delayed_work (&ctx->file_put_work); +if (locked) +mutex_unlock (&ctx->uring_lock);do { ret =wait_for_completion_interruptible (&data->done);if (!ret)break ; ret =io_run_task_work_sig (); -if (ret <0 ) { -percpu_ref_resurrect (&data->refs); -reinit_completion (&data->done); -io_sqe_files_set_node (data, backup_node); -return ret; - } +if (ret <0 ) +break ; }while (1 ); +if (locked) +mutex_lock (&ctx->uring_lock); + +if (ret <0 ) { +percpu_ref_resurrect (&data->refs); +reinit_completion (&data->done); +io_sqe_files_set_node (data, backup_node); +return ret; + }__io_sqe_files_unregister (ctx); nr_tables =DIV_ROUND_UP (ctx->nr_user_files,IORING_MAX_FILES_TABLE );
Re: [PATCH 2/2] io_uring: don't hold uring_lock when calling io_run_task_work*
In reply to Hao XuQuoted 6 lines · Hao Xu
This is caused by calling
io_run_task_work_sig()to do work underuring_lockwhile the callerio_sqe_files_unregister()already helduring_lock.
we need to check ifuring_lockis held by us when doing unlock aroundio_run_task_work_sig()since there are code paths down to that place withouturing_lockheld.
btw, better to prepare it for-5.12, seems it won't apply
Quoted 2 lines
fs/io_uring.c +0
-0
[...]
fs/io_uring.c static int io_sqe_files_unregister(struct io_ring_ctx *ctx, bool locked) · line 7374
Quoted 9 lines
fs/io_uring.c +2
-6-if (ret <0 ) { -percpu_ref_resurrect (&data->refs); -reinit_completion (&data->done); -io_sqe_files_set_node (data, backup_node); -return ret; - } +if (ret <0 ) +break ; }while (1 );
while (ret >= 0) ?
Signature · 1 line
Pavel Begunkov