Re: [PATCH] io_uring: don't recursively hold ctx->uring_lock in io_wq_submit_work()

[Pavel Begunkov <asml.silence@gmail.com>]

On 18/02/2021 17:16, Hao Xu wrote:
> 在 2021/1/25 下午12:31, Jens Axboe 写道:
>> On 1/23/21 2:40 AM, Hao Xu wrote:
>>> Abaci reported the following warning:
>>>
>>> [   97.862205] ============================================
>>> [   97.863400] WARNING: possible recursive locking detected
>>> [   97.864640] 5.11.0-rc4+ #12 Not tainted
>>> [   97.865537] --------------------------------------------
>>> [   97.866748] a.out/2890 is trying to acquire lock:
>>> [   97.867829] ffff8881046763e8 (&ctx->uring_lock){+.+.}-{3:3}, at:
>>> io_wq_submit_work+0x155/0x240
>>> [   97.869735]
>>> [   97.869735] but task is already holding lock:
>>> [   97.871033] ffff88810dfe0be8 (&ctx->uring_lock){+.+.}-{3:3}, at:
>>> __x64_sys_io_uring_enter+0x3f0/0x5b0
>>> [   97.873074]
>>> [   97.873074] other info that might help us debug this:
>>> [   97.874520]  Possible unsafe locking scenario:
>>> [   97.874520]
>>> [   97.875845]        CPU0
>>> [   97.876440]        ----
>>> [   97.877048]   lock(&ctx->uring_lock);
>>> [   97.877961]   lock(&ctx->uring_lock);
>>> [   97.878881]
>>> [   97.878881]  *** DEADLOCK ***
>>> [   97.878881]
>>> [   97.880341]  May be due to missing lock nesting notation
>>> [   97.880341]
>>> [   97.881952] 1 lock held by a.out/2890:
>>> [   97.882873]  #0: ffff88810dfe0be8 (&ctx->uring_lock){+.+.}-{3:3}, at:
>>> __x64_sys_io_uring_enter+0x3f0/0x5b0
>>> [   97.885108]
>>> [   97.885108] stack backtrace:
>>> [   97.886209] CPU: 0 PID: 2890 Comm: a.out Not tainted 5.11.0-rc4+ #12
>>> [   97.887683] Hardware name: Alibaba Cloud Alibaba Cloud ECS, BIOS
>>> rel-1.7.5-0-ge51488c-20140602_164612-nilsson.home.kraxel.org 04/01/2014
>>> [   97.890457] Call Trace:
>>> [   97.891121]  dump_stack+0xac/0xe3
>>> [   97.891972]  __lock_acquire+0xab6/0x13a0
>>> [   97.892940]  lock_acquire+0x2c3/0x390
>>> [   97.893853]  ? io_wq_submit_work+0x155/0x240
>>> [   97.894894]  __mutex_lock+0xae/0x9f0
>>> [   97.895785]  ? io_wq_submit_work+0x155/0x240
>>> [   97.896816]  ? __lock_acquire+0x782/0x13a0
>>> [   97.897817]  ? io_wq_submit_work+0x155/0x240
>>> [   97.898867]  ? io_wq_submit_work+0x155/0x240
>>> [   97.899916]  ? _raw_spin_unlock_irqrestore+0x2d/0x40
>>> [   97.901101]  io_wq_submit_work+0x155/0x240
>>> [   97.902112]  io_wq_cancel_cb+0x162/0x490
>>> [   97.903084]  ? io_uring_get_socket+0x40/0x40
>>> [   97.904126]  io_async_find_and_cancel+0x3b/0x140
>>> [   97.905247]  io_issue_sqe+0x86d/0x13e0
>>> [   97.906186]  ? __lock_acquire+0x782/0x13a0
>>> [   97.907195]  ? __io_queue_sqe+0x10b/0x550
>>> [   97.908175]  ? lock_acquire+0x2c3/0x390
>>> [   97.909122]  __io_queue_sqe+0x10b/0x550
>>> [   97.910080]  ? io_req_prep+0xd8/0x1090
>>> [   97.911044]  ? mark_held_locks+0x5a/0x80
>>> [   97.912042]  ? mark_held_locks+0x5a/0x80
>>> [   97.913014]  ? io_queue_sqe+0x235/0x470
>>> [   97.913971]  io_queue_sqe+0x235/0x470
>>> [   97.914894]  io_submit_sqes+0xcce/0xf10
>>> [   97.915842]  ? xa_store+0x3b/0x50
>>> [   97.916683]  ? __x64_sys_io_uring_enter+0x3f0/0x5b0
>>> [   97.917872]  __x64_sys_io_uring_enter+0x3fb/0x5b0
>>> [   97.918995]  ? lockdep_hardirqs_on_prepare+0xde/0x180
>>> [   97.920204]  ? syscall_enter_from_user_mode+0x26/0x70
>>> [   97.921424]  do_syscall_64+0x2d/0x40
>>> [   97.922329]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
>>> [   97.923538] RIP: 0033:0x7f0b62601239
>>> [   97.924437] 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 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 27 ec 2c 00 f7 d8 64 89 01
>>>        48
>>> [   97.928628] RSP: 002b:00007f0b62cc4d28 EFLAGS: 00000246 ORIG_RAX:
>>> 00000000000001aa
>>> [   97.930422] RAX: ffffffffffffffda RBX: 0000000000000000 RCX:
>>> 00007f0b62601239
>>> [   97.932073] RDX: 0000000000000000 RSI: 0000000000006cf6 RDI:
>>> 0000000000000005
>>> [   97.933710] RBP: 00007f0b62cc4e20 R08: 0000000000000000 R09:
>>> 0000000000000000
>>> [   97.935369] R10: 0000000000000000 R11: 0000000000000246 R12:
>>> 0000000000000000
>>> [   97.937008] R13: 0000000000021000 R14: 0000000000000000 R15:
>>> 00007f0b62cc5700
>>>
>>> This is caused by try to hold uring_lock in io_wq_submit_work() without
>>> checking if we are in io-wq thread context or not. It can be in original
>>> context when io_wq_submit_work() is called from IORING_OP_ASYNC_CANCEL
>>> code path, where we already held uring_lock.
>>
>> Looks like another fallout of the split CLOSE handling. I've got the
>> right fixes pending for 5.12:
>>
>> https://git.kernel.dk/cgit/linux-block/commit/?h=for-5.12/io_uring&id=6bb0079ef3420041886afe1bcd8e7a87e08992e1
>>
>> (and the prep patch before that in the tree). But that won't really
>> help us for 5.11 and earlier, though we probably should just queue
>> those two patches for 5.11 and get them into stable. I really don't
>> like the below patch, though it should fix it. But the root cause
>> is really the weird open cancelation...
>>
> Hi Jens,
> I've repro-ed this issue on branch for-5.12/io_uring-2021-02-17
> which contains the patch you give, the issue still exists.
> I think this one is not an async close specifical problem.
> The rootcause is we try to run an iowq work in the original
> context(queue an iowq work, then async cancel it).
If you mean cancellation executed from task_work or inline (during
submission), then yes, I agree.

Can you try a diff below?

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 2fdfe5fa00b0..8dab07f42b34 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -2337,7 +2337,9 @@ static void io_req_task_cancel(struct callback_head *cb)
 	struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
 	struct io_ring_ctx *ctx = req->ctx;
 
+	mutex_lock(&ctx->uring_lock);
 	__io_req_task_cancel(req, -ECANCELED);
+	mutex_unlock(&ctx->uring_lock);
 	percpu_ref_put(&ctx->refs);
 }
 
@@ -6426,8 +6428,13 @@ static void io_wq_submit_work(struct io_wq_work *work)
 	if (timeout)
 		io_queue_linked_timeout(timeout);
 
-	if (work->flags & IO_WQ_WORK_CANCEL)
-		ret = -ECANCELED;
+	if (work->flags & IO_WQ_WORK_CANCEL) {
+		/* io-wq is going to take down one */
+		refcount_inc(&req->refs);
+		percpu_ref_get(&req->ctx->refs);
+		io_req_task_work_add_fallback(req, io_req_task_cancel);
+		return;
+	}
 
 	if (!ret) {
 		do {
-- 
2.24.0