Re: [xfstests generic/617] fsx io_uring dio starts to fail on overlayfs since v6.6-rc1

[Zorro Lang <zlang@redhat.com>]

On Sun, Sep 24, 2023 at 11:52:12AM -0600, Jens Axboe wrote:
> On 9/24/23 8:29 AM, Zorro Lang wrote:
> > Hi,
> > 
> > The generic/617 of fstests is a test case does IO_URING soak direct-IO
> > fsx test, but recently (about from v6.6-rc1 to now) it always fails on
> > overlayfs as [1], no matter the underlying fs is ext4 or xfs. But it
> > never failed on overlay before, likes [2].
> > 
> > So I thought it might be a regression of overlay or io-uring on current v6.6.
> > Please help to review, it's easy to reproduce. My system is Fedora-rawhide/RHEL-9,
> > with upstream mainline linux HEAD=dc912ba91b7e2fa74650a0fc22cccf0e0d50f371.
> > The generic/617.full output as [3].
> 
> It works without overlayfs - would be great if you could include how to
> reproduce this with overlayfs.

Sorry I didn't provide the test steps:

1) git clone git://git.kernel.org/pub/scm/fs/xfs/xfstests-dev.git
2) cd xfstests-dev
3) make -j8  # refer to README, about how to build xfstests
4) vi local.config
# The FSTYP can be other fs, e.g. ext4.
# The TEST_DEV and SCRATCH_DEV can be any local devices you have (even loopdev)
# TEST_DIR and SCRATCH_MNT can be changed too
export FSTYP=xfs
export TEST_DEV=/dev/sda5
export TEST_DIR=/mnt/test
export SCRATCH_DEV=/dev/sda3
export SCRATCH_MNT=/mnt/scratch

5) mkfs.xfs -f /dev/sda5 && mkfs.xfs -f /dev/sda3
6) mkdir -p /mnt/test && mkdir -p /mnt/scratch
7) ./check -overlay generic/617

The step#6 will be failed.

Thanks,
Zorro

> 
> -- 
> Jens Axboe
>