Re: [PATCH] xfs/298: Add 100ms sleep before scratch_umount

[Dave Chinner <david@fromorbit.com>]

On Mon, May 09, 2022 at 02:56:57PM +0800, Yang Xu wrote:
> When testing this case on my machine, it reports the following error:
> umount: /mnt/xfstests/scratch: target is busy.
> xfs_db: /dev/sda11 contains a mounted filesystem
> 
> scratch_unmount failed, so  _scratch_xfs_db reports scratch_dev is a
> mounted filesystem. It seems filesystem has something to be doing.
> 
> To avoid this, just add a 100ms sleep before scratch_umount.
> 
> Signed-off-by: Yang Xu <xuyang2018.jy@fujitsu.com>
> ---
>  tests/xfs/298 | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/tests/xfs/298 b/tests/xfs/298
> index b0153ebf..17510379 100755
> --- a/tests/xfs/298
> +++ b/tests/xfs/298
> @@ -51,6 +51,7 @@ while [ $SIZE -lt 1024 ];do
>  	rm $SYMLINK_FILE
>  # umount and check the number of extents on the inode. Should be 0.
>  	cd /
> +	sleep 0.1
>  	_scratch_unmount >/dev/null 2>&1

What? No.

Never put random undocument sleeps in tests to hide failures - they
are almost always covering up a problem that needs to be understood
and fixed.

First thing to do here is understand why the filesystem is busy and
can't unmount. What is holding a reference that prevents unmount?

Is systemd or udev doing something stupid on your system and so
racing with unmount? Or is something else going on?

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com