Re: [PATCH 5.4 103/142] Revert "block: end bio with BLK_STS_AGAIN in case of non-mq devs and REQ_NOWAIT"

[Andres Freund]

Hi,

On 2020-06-01 19:54:21 +0200, Greg Kroah-Hartman wrote:
> From: Jens Axboe <axboe@xxxxxxxxx>
>
> [ Upstream commit b0beb28097fa04177b3769f4bb7a0d0d9c4ae76e ]
>
> This reverts commit c58c1f83436b501d45d4050fd1296d71a9760bcb.
>
> io_uring does do the right thing for this case, and we're still returning
> -EAGAIN to userspace for the cases we don't support. Revert this change
> to avoid doing endless spins of resubmits.
>
> Cc: stable@xxxxxxxxxxxxxxx # v5.6
> Reported-by: Bijan Mottahedeh <bijan.mottahedeh@xxxxxxxxxx>
> Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
> Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
> ---
>  block/blk-core.c | 11 ++++-------
>  1 file changed, 4 insertions(+), 7 deletions(-)
>

This broke io_uring direct-io on ext4 over md.

fallocate -l $((1024*1024*1024)) /srv/part1
fallocate -l $((1024*1024*1024)) /srv/part2
losetup -f /srv/part1
losetup -f /srv/part2
losetup -a # assuming these were loop0/1
mdadm --create -n2 -l stripe  -N fast-striped /dev/md/fast-striped /dev/loop0 /dev/loop1
mkfs.ext4 /dev/md/fast-striped
mount /dev/md/fast-striped /mnt/t2
fio --directory=/mnt/t2 --ioengine io_uring --rw write --filesize 1MB --overwrite=1 --name=test --direct=1 --bs=4k

On v5.4.43-101-gbba91cdba612 this fails with
fio: io_u error on file /mnt/t2/test.0.0: Input/output error: write offset=0, buflen=4096
fio: pid=734, err=5/file:io_u.c:1834, func=io_u error, error=Input/output error

whereas previously it worked. libaio still works...

I haven't checked which major kernel version fixed this again, but I did
verify that it's still broken in 5.4.94 and that 5.10.9 works.

I would suspect it's

commit 4503b7676a2e0abe69c2f2c0d8b03aec53f2f048
Author: Jens Axboe <axboe@xxxxxxxxx>
Date:   2020-06-01 10:00:27 -0600

    io_uring: catch -EIO from buffered issue request failure

    -EIO bubbles up like -EAGAIN if we fail to allocate a request at the
    lower level. Play it safe and treat it like -EAGAIN in terms of sync
    retry, to avoid passing back an errant -EIO.

    Catch some of these early for block based file, as non-mq devices
    generally do not support NOWAIT. That saves us some overhead by
    not first trying, then retrying from async context. We can go straight
    to async punt instead.

    Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>


which isn't in stable/linux-5.4.y

Greetings,

Andres Freund