2020-01-22 20:11:56

by Pavel Begunkov

[permalink] [raw]
Subject: [PATCH 0/2] IOSQE_ASYNC patches

There are 2 problems addressed:
1. it never calls *_prep() when going through IOSQE_ASYNC path.
2. non-head linked reqs ignore IOSQE_ASYNC.

Also, there could be yet another problem, when we bypass io_issue_req()
and going straight to async.

Pavel Begunkov (2):
io_uring: prep req when do IOSQE_ASYNC
io_uring: honor IOSQE_ASYNC for linked reqs

fs/io_uring.c | 8 ++++++++
1 file changed, 8 insertions(+)

--
2.24.0


2020-01-22 20:12:13

by Pavel Begunkov

[permalink] [raw]
Subject: [PATCH 1/2] io_uring: prep req when do IOSQE_ASYNC

Whenever IOSQE_ASYNC is set, requests will be punted to async without
getting into io_issue_req() and without proper preparation done (e.g.
io_req_defer_prep()). Hence they will be left uninitialised.

Prepare them before punting.

Signed-off-by: Pavel Begunkov <[email protected]>
---
fs/io_uring.c | 4 ++++
1 file changed, 4 insertions(+)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 09503d1e9e45..cdbc711ae5fd 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -4558,11 +4558,15 @@ static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe)
ret = io_req_defer(req, sqe);
if (ret) {
if (ret != -EIOCBQUEUED) {
+fail_req:
io_cqring_add_event(req, ret);
req_set_fail_links(req);
io_double_put_req(req);
}
} else if (req->flags & REQ_F_FORCE_ASYNC) {
+ ret = io_req_defer_prep(req, sqe);
+ if (unlikely(ret < 0))
+ goto fail_req;
/*
* Never try inline submit of IOSQE_ASYNC is set, go straight
* to async execution.
--
2.24.0

2020-01-22 20:16:08

by Pavel Begunkov

[permalink] [raw]
Subject: Re: [PATCH 0/2] IOSQE_ASYNC patches

On 22/01/2020 23:09, Pavel Begunkov wrote:
> There are 2 problems addressed:
> 1. it never calls *_prep() when going through IOSQE_ASYNC path.
> 2. non-head linked reqs ignore IOSQE_ASYNC.

Those two are intentionally short for quick fix up. I'll prepare something
prettier on top of that a bit later.

>
> Also, there could be yet another problem, when we bypass io_issue_req()
> and going straight to async.
>
> Pavel Begunkov (2):
> io_uring: prep req when do IOSQE_ASYNC
> io_uring: honor IOSQE_ASYNC for linked reqs
>
> fs/io_uring.c | 8 ++++++++
> 1 file changed, 8 insertions(+)
>

--
Pavel Begunkov


Attachments:
signature.asc (849.00 B)
OpenPGP digital signature

2020-01-22 20:59:50

by Jens Axboe

[permalink] [raw]
Subject: Re: [PATCH 0/2] IOSQE_ASYNC patches

On 1/22/20 1:09 PM, Pavel Begunkov wrote:
> There are 2 problems addressed:
> 1. it never calls *_prep() when going through IOSQE_ASYNC path.
> 2. non-head linked reqs ignore IOSQE_ASYNC.
>
> Also, there could be yet another problem, when we bypass io_issue_req()
> and going straight to async.

Thanks, applied.

--
Jens Axboe