Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH 1/1] io_uring: don't wait when under-submitting
From:   Jens Axboe <axboe@kernel.dk>
To:     Pavel Begunkov <asml.silence@gmail.com>, io-uring@vger.kernel.org,
        linux-kernel@vger.kernel.org
References: <5caa38be87f069eb4cc921d58ee1a98ff5d53978.1576223348.git.asml.silence@gmail.com>
 <21ca72b0-c35d-96b7-399f-d4034d976c27@kernel.dk>
Message-ID: <9fbb03f4-6444-04a6-4cfb-ee4b3aa0bcd1@kernel.dk>
Date:   Fri, 13 Dec 2019 11:32:38 -0700
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.2.2
MIME-Version: 1.0
In-Reply-To: <21ca72b0-c35d-96b7-399f-d4034d976c27@kernel.dk>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 12/13/19 11:22 AM, Jens Axboe wrote:
> On 12/13/19 12:51 AM, Pavel Begunkov wrote:
>> There is no reliable way to submit and wait in a single syscall, as
>> io_submit_sqes() may under-consume sqes (in case of an early error).
>> Then it will wait for not-yet-submitted requests, deadlocking the user
>> in most cases.
> 
> Why not just cap the wait_nr? If someone does to_submit = 8, wait_nr = 8,
> and we only submit 4, just wait for 4? Ala:
> 
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index 81219a631a6d..4a76ccbb7856 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -5272,6 +5272,10 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
>  		submitted = io_submit_sqes(ctx, to_submit, f.file, fd,
>  					   &cur_mm, false);
>  		mutex_unlock(&ctx->uring_lock);
> +		if (submitted <= 0)
> +			goto done;
> +		if (submitted != to_submit && min_complete > submitted)
> +			min_complete = submitted;
>  	}
>  	if (flags & IORING_ENTER_GETEVENTS) {
>  		unsigned nr_events = 0;
> @@ -5284,7 +5288,7 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
>  			ret = io_cqring_wait(ctx, min_complete, sig, sigsz);
>  		}
>  	}
> -
> +done:
>  	percpu_ref_put(&ctx->refs);
>  out_fput:
>  	fdput(f);
> 

This is probably a bit cleaner, since it only adjusts if we're going to
wait.


diff --git a/fs/io_uring.c b/fs/io_uring.c
index 81219a631a6d..e262549a2601 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -5272,11 +5272,15 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
 		submitted = io_submit_sqes(ctx, to_submit, f.file, fd,
 					   &cur_mm, false);
 		mutex_unlock(&ctx->uring_lock);
+		if (submitted <= 0)
+			goto done;
 	}
 	if (flags & IORING_ENTER_GETEVENTS) {
 		unsigned nr_events = 0;
 
 		min_complete = min(min_complete, ctx->cq_entries);
+		if (submitted != to_submit && min_complete > submitted)
+			min_complete = submitted;
 
 		if (ctx->flags & IORING_SETUP_IOPOLL) {
 			ret = io_iopoll_check(ctx, &nr_events, min_complete);
@@ -5284,7 +5288,7 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
 			ret = io_cqring_wait(ctx, min_complete, sig, sigsz);
 		}
 	}
-
+done:
 	percpu_ref_put(&ctx->refs);
 out_fput:
 	fdput(f);

-- 
Jens Axboe