On 9/8/20 7:11 AM, Jens Axboe wrote:
> On 9/8/20 6:31 AM, Greg KH wrote:
>> On Fri, Sep 04, 2020 at 03:06:28PM -0600, Jens Axboe wrote:
>>> Hi,
>>>
>>> Linus just pulled 3 fixes from me - 1+2 should apply directly, here's
>>> the 3rd one which will need some love for 5.8-stable. I'm including it
>>> below to preempt the failed to apply message :-)
>>>
>>>
>>> commit fb8d4046d50f77a26570101e5b8a7a026320a610
>>> Author: Jens Axboe <axboe@xxxxxxxxx>
>>> Date: Wed Sep 2 10:19:04 2020 -0600
>>>
>>> io_uring: no read/write-retry on -EAGAIN error and O_NONBLOCK marked file
>>>
>>> Actually two things that need fixing up here:
>>>
>>> - The io_rw_reissue() -EAGAIN retry is explicit to block devices and
>>> regular files, so don't ever attempt to do that on other types of
>>> files.
>>>
>>> - If we hit -EAGAIN on a nonblock marked file, don't arm poll handler for
>>> it. It should just complete with -EAGAIN.
>>>
>>> Cc: stable@xxxxxxxxxxxxxxx
>>> Reported-by: Norman Maurer <norman.maurer@xxxxxxxxxxxxxx>
>>> Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
>>>
>>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>>> index 82e15020d9a8..96be21ace79a 100644
>>> --- a/fs/io_uring.c
>>> +++ b/fs/io_uring.c
>>> @@ -2726,6 +2726,12 @@ static int io_read(struct io_kiocb *req, bool force_nonblock)
>>> ret = ret2;
>>> goto done;
>>> }
>>> + /* no retry on NONBLOCK marked file */
>>> + if (req->file->f_flags & O_NONBLOCK) {
>>> + ret = ret2;
>>> + goto done;
>>> + }
>>> +
>>> /* some cases will consume bytes even on error returns */
>>> iov_iter_revert(iter, iov_count - iov_iter_count(iter));
>>> ret2 = 0;
>>> @@ -2869,9 +2875,15 @@ static int io_write(struct io_kiocb *req, bool force_nonblock)
>>> */
>>> if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT))
>>> ret2 = -EAGAIN;
>>> + /* no retry on NONBLOCK marked file */
>>> + if (ret2 == -EAGAIN && (req->file->f_flags & O_NONBLOCK)) {
>>> + ret = 0;
>>> + goto done;
>>> + }
>>> if (!force_nonblock || ret2 != -EAGAIN) {
>>> if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN)
>>> goto copy_iov;
>>> +done:
>>> kiocb_done(kiocb, ret2);
>>> } else {
>>> copy_iov:
>>>
>>> --
>>> Jens Axboe
>>
>>
>> Thanks for the backport, but this didn't apply at all to the 5.8.y tree.
>> What one did you make it against?
>
> Oh, might have been because I have a pile of pending 5.8 stable patches...
> Let me apply to pristine stable, test, and then I'll send it to you.
Here it is:
commit d0ea3f3d17cf891244f17f8ceb43d4988c170471
Author: Jens Axboe <axboe@xxxxxxxxx>
Date: Tue Sep 8 07:16:12 2020 -0600
io_uring: no read/write-retry on -EAGAIN error and O_NONBLOCK marked file
Actually two things that need fixing up here:
- The io_rw_reissue() -EAGAIN retry is explicit to block devices and
regular files, so don't ever attempt to do that on other types of
files.
- If we hit -EAGAIN on a nonblock marked file, don't arm poll handler for
it. It should just complete with -EAGAIN.
Cc: stable@xxxxxxxxxxxxxxx
Reported-by: Norman Maurer <norman.maurer@xxxxxxxxxxxxxx>
Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
diff --git a/fs/io_uring.c b/fs/io_uring.c
index 4115bfedf15d..fe114511d6d6 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -2697,8 +2697,15 @@ static int io_read(struct io_kiocb *req, bool force_nonblock)
else
ret2 = -EINVAL;
+ /* no retry on NONBLOCK marked file */
+ if (ret2 == -EAGAIN && (req->file->f_flags & O_NONBLOCK)) {
+ ret = 0;
+ goto done;
+ }
+
/* Catch -EAGAIN return for forced non-blocking submission */
if (!force_nonblock || ret2 != -EAGAIN) {
+ done:
kiocb_done(kiocb, ret2);
} else {
copy_iov:
@@ -2823,7 +2830,13 @@ static int io_write(struct io_kiocb *req, bool force_nonblock)
*/
if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT))
ret2 = -EAGAIN;
+ /* no retry on NONBLOCK marked file */
+ if (ret2 == -EAGAIN && (req->file->f_flags & O_NONBLOCK)) {
+ ret = 0;
+ goto done;
+ }
if (!force_nonblock || ret2 != -EAGAIN) {
+done:
kiocb_done(kiocb, ret2);
} else {
copy_iov:
--
Jens Axboe
