[PATCH] io_uring: fixup a few spots where link failure isn't flagged

[PATCH] io_uring: fixup a few spots where link failure isn't flagged

[Jens Axboe]

If a request fails, we need to ensure we set REQ_F_FAIL_LINK on it if
REQ_F_LINK is set. Any failure in the chain should break the chain.

We were missing a few spots where this should be done. It might be nice
to generalize this somewhat at some point, as long as we factor in the
fact that failure looks different for each request type.

Signed-off-by: Jens Axboe <axboe@kernel.dk>

--


diff --git a/fs/io_uring.c b/fs/io_uring.c
index bda27b52fd5b..4edc94aab17e 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -1672,6 +1672,8 @@ static int io_send_recvmsg(struct io_kiocb *req, const struct io_uring_sqe *sqe,
 	}
 
 	io_cqring_add_event(req->ctx, sqe->user_data, ret);
+	if (ret < 0 && (req->flags & REQ_F_LINK))
+		req->flags |= REQ_F_FAIL_LINK;
 	io_put_req(req, nxt);
 	return 0;
 }
@@ -1787,6 +1789,8 @@ static int io_poll_remove(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	spin_unlock_irq(&ctx->completion_lock);
 
 	io_cqring_add_event(req->ctx, sqe->user_data, ret);
+	if (ret < 0 && (req->flags & REQ_F_LINK))
+		req->flags |= REQ_F_FAIL_LINK;
 	io_put_req(req, NULL);
 	return 0;
 }
@@ -1994,6 +1998,8 @@ static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
 	spin_unlock_irqrestore(&ctx->completion_lock, flags);
 
 	io_cqring_ev_posted(ctx);
+	if (req->flags & REQ_F_LINK)
+		req->flags |= REQ_F_FAIL_LINK;
 	io_put_req(req, NULL);
 	return HRTIMER_NORESTART;
 }
@@ -2035,6 +2041,8 @@ static int io_timeout_remove(struct io_kiocb *req,
 		io_commit_cqring(ctx);
 		spin_unlock_irq(&ctx->completion_lock);
 		io_cqring_ev_posted(ctx);
+		if (req->flags & REQ_F_LINK)
+			req->flags |= REQ_F_FAIL_LINK;
 		io_put_req(req, NULL);
 		return 0;
 	}
@@ -2328,6 +2336,8 @@ static void io_wq_submit_work(struct io_wq_work **workptr)
 	io_put_req(req, NULL);
 
 	if (ret) {
+		if (req->flags & REQ_F_LINK)
+			req->flags |= REQ_F_FAIL_LINK;
 		io_cqring_add_event(ctx, sqe->user_data, ret);
 		io_put_req(req, NULL);
 	}

-- 
Jens Axboe

Re: [PATCH] io_uring: fixup a few spots where link failure isn't flagged

[Pavel Begunkov]

[-- Attachment #1.1: Type: text/plain, Size: 2554 bytes --]

On 06/11/2019 06:32, Jens Axboe wrote:
> If a request fails, we need to ensure we set REQ_F_FAIL_LINK on it if
> REQ_F_LINK is set. Any failure in the chain should break the chain.
> 
> We were missing a few spots where this should be done. It might be nice
> to generalize this somewhat at some point, as long as we factor in the
> fact that failure looks different for each request type.
> 

The completion path also starts to get complicated, especially
non-uniform handling of links there.

i.e. io_put_req() -> io_put_req_find_next() ->
	io_free_req() -> __io_free_req()
Plus, io_free_req_many(), which can be used only in some cases.

My compiler didn't even inlined it, so there are 4 CALLs.
Though, still in TODO list.


> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> 
> --
> 
> 
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index bda27b52fd5b..4edc94aab17e 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -1672,6 +1672,8 @@ static int io_send_recvmsg(struct io_kiocb *req, const struct io_uring_sqe *sqe,
>  	}
>  
>  	io_cqring_add_event(req->ctx, sqe->user_data, ret);
> +	if (ret < 0 && (req->flags & REQ_F_LINK))
> +		req->flags |= REQ_F_FAIL_LINK;
>  	io_put_req(req, nxt);
>  	return 0;
>  }
> @@ -1787,6 +1789,8 @@ static int io_poll_remove(struct io_kiocb *req, const struct io_uring_sqe *sqe)
>  	spin_unlock_irq(&ctx->completion_lock);
>  
>  	io_cqring_add_event(req->ctx, sqe->user_data, ret);
> +	if (ret < 0 && (req->flags & REQ_F_LINK))
> +		req->flags |= REQ_F_FAIL_LINK;
>  	io_put_req(req, NULL);
>  	return 0;
>  }
> @@ -1994,6 +1998,8 @@ static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
>  	spin_unlock_irqrestore(&ctx->completion_lock, flags);
>  
>  	io_cqring_ev_posted(ctx);
> +	if (req->flags & REQ_F_LINK)
> +		req->flags |= REQ_F_FAIL_LINK;
>  	io_put_req(req, NULL);
>  	return HRTIMER_NORESTART;
>  }
> @@ -2035,6 +2041,8 @@ static int io_timeout_remove(struct io_kiocb *req,
>  		io_commit_cqring(ctx);
>  		spin_unlock_irq(&ctx->completion_lock);
>  		io_cqring_ev_posted(ctx);
> +		if (req->flags & REQ_F_LINK)
> +			req->flags |= REQ_F_FAIL_LINK;
>  		io_put_req(req, NULL);
>  		return 0;
>  	}
> @@ -2328,6 +2336,8 @@ static void io_wq_submit_work(struct io_wq_work **workptr)
>  	io_put_req(req, NULL);
>  
>  	if (ret) {
> +		if (req->flags & REQ_F_LINK)
> +			req->flags |= REQ_F_FAIL_LINK;
>  		io_cqring_add_event(ctx, sqe->user_data, ret);
>  		io_put_req(req, NULL);
>  	}
> 

-- 
Pavel Begunkov


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH] io_uring: fixup a few spots where link failure isn't flagged

[Jens Axboe]

On 11/6/19 11:44 AM, Pavel Begunkov wrote:
> On 06/11/2019 06:32, Jens Axboe wrote:
>> If a request fails, we need to ensure we set REQ_F_FAIL_LINK on it if
>> REQ_F_LINK is set. Any failure in the chain should break the chain.
>>
>> We were missing a few spots where this should be done. It might be nice
>> to generalize this somewhat at some point, as long as we factor in the
>> fact that failure looks different for each request type.
>>
> 
> The completion path also starts to get complicated, especially
> non-uniform handling of links there.
> 
> i.e. io_put_req() -> io_put_req_find_next() ->
> 	io_free_req() -> __io_free_req()
> Plus, io_free_req_many(), which can be used only in some cases.

The many case is just for polling, I actually think that one is pretty
clean and self explanatory. But I agree on the proliferation of them
otherwise, at least the patches today cleaned it up a little.
 
> My compiler didn't even inlined it, so there are 4 CALLs.
> Though, still in TODO list.

Huh, that's odd!

-- 
Jens Axboe