[PATCH] io_uring: fix stuck problem caused by potential memory alloc failure

[PATCH] io_uring: fix stuck problem caused by potential memory alloc failure

[Jackie Liu]

when io_req_defer alloc memory failed, it will be return -EAGAIN.
But io_submit_sqe cannot be returned immediately because the reference
count for req is still hold. we should be clean it.

Fixes: de0617e46717 ("io_uring: add support for marking commands as draining")
Cc: linux-block@vger.kernel.org
Signed-off-by: Jackie Liu <liuyun01@kylinos.cn>
---
 fs/io_uring.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 4ef62a45045d..1c388533cdc8 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -1843,8 +1843,8 @@ static int io_submit_sqe(struct io_ring_ctx *ctx, struct sqe_submit *s,
 	ret = io_req_defer(ctx, req, s->sqe);
 	if (ret) {
 		if (ret == -EIOCBQUEUED)
-			ret = 0;
-		return ret;
+			return 0;
+		goto out;
 	}
 
 	ret = __io_submit_sqe(ctx, req, s, true);
-- 
2.22.0

Re: [PATCH] io_uring: fix stuck problem caused by potential memory alloc failure

[Jens Axboe]

On 7/10/19 2:35 AM, Jackie Liu wrote:
> when io_req_defer alloc memory failed, it will be return -EAGAIN.
> But io_submit_sqe cannot be returned immediately because the reference
> count for req is still hold. we should be clean it.

Looks like this actually got fixed in my for-5.3/io_uring branch which
I haven't pushed out yet. Once that's in, I'd suggest we send yours to
stable separately. Probably change the wording to:

When io_req_defer alloc memory fails, it will be -EAGAIN. But
io_submit_sqe cannot return immediately because the reference count for
req is still held. Ensure that we free it.

-- 
Jens Axboe

Re: [PATCH] io_uring: fix stuck problem caused by potential memory alloc failure

[JackieLiu]

Thank you for pointing out. I haven't had time to pay attention to the new 
features of sqe links. When I was debugging another stuck problem, I found 
that there was no free the reference count, so I sent the patch about this.

Anyway, if necessary, I will resend the patch again, with your rewritten
comments.
 

> 在 2019年7月10日，21:57，Jens Axboe <axboe@kernel.dk> 写道：
> 
> On 7/10/19 2:35 AM, Jackie Liu wrote:
>> when io_req_defer alloc memory failed, it will be return -EAGAIN.
>> But io_submit_sqe cannot be returned immediately because the reference
>> count for req is still hold. we should be clean it.
> 
> Looks like this actually got fixed in my for-5.3/io_uring branch which
> I haven't pushed out yet. Once that's in, I'd suggest we send yours to
> stable separately. Probably change the wording to:
> 
> When io_req_defer alloc memory fails, it will be -EAGAIN. But
> io_submit_sqe cannot return immediately because the reference count for
> req is still held. Ensure that we free it.
> 
> -- 
> Jens Axboe
> 
>