Re: [PATCH 5/9] common/dmthin: make this work with external log devices

From: "Theodore Ts'o" <tytso@mit.edu>
To: Eryu Guan <guan@eryu.me>
Cc: "Darrick J. Wong" <djwong@kernel.org>,
	guaneryu@gmail.com, linux-xfs@vger.kernel.org,
	fstests@vger.kernel.org
Subject: Re: [PATCH 5/9] common/dmthin: make this work with external log devices
Date: Sun, 18 Apr 2021 10:44:08 -0400	[thread overview]
Message-ID: <YHxFuFvbAiiIrbIo@mit.edu> (raw)
In-Reply-To: <YHwlTMySYgKuaw6Y@desktop>

On Sun, Apr 18, 2021 at 08:25:48PM +0800, Eryu Guan wrote:
> > diff --git a/tests/generic/223 b/tests/generic/223
> > index 1f85efe5..a5ace82f 100755
> > --- a/tests/generic/223
> > +++ b/tests/generic/223
> > @@ -43,6 +43,9 @@ for SUNIT_K in 8 16 32 64 128; do
> >  	_scratch_mkfs_geom $SUNIT_BYTES 4 $BLOCKSIZE >> $seqres.full 2>&1
> >  	_scratch_mount
> >  
> > +	# Make sure everything is on the data device
> > +	$XFS_IO_PROG -c 'chattr -t' $SCRATCH_MNT
> 
> What does this do for non-xfs filesystems? Do we need a FSTYP check and
> do chattr only on XFS?

This clears the FS_NOTAIL_FL flag, which prevents tail merging, on the
root directory of the mounted scratch file system.  That should be
harmless on non-xfs file systems; in fact, the only file system that
even uses NOTAIL_FL flag is reiserfs, and the NOTAIL_FL flag is not
set by default on the root directory of a newly created reiserfs file
system.

However, by default reiserfs does not support the
FS_IOC_{GET,SET}FLAGS ioctl unless the mount option "attrs" is given.
Why, I have no idea:

root@kvm-xfstests:~# mount -t reiserfs /vtmp/foo.img /mnt
root@kvm-xfstests:~# xfs_io -c 'chattr -t' /mnt
xfs_io: cannot get flags on /mnt: Inappropriate ioctl for device

So it might be a good idea to redirect stderr for the xfs_io
invocation to /dev/null, for those file systems which do not support
the FS_IOC_{GET,SET}FLAGS ioctls.

I also have no idea why this helps for xfs --- I would think it's a
no-op, but I'm guessing there's some magical side-effect which is
taking place when FS_IOC_SETFLAGS ioctl is processed?  Maybe it would
be worth a comment explaining what is going on --- and whether this is
going to make any difference if the patch series which unifies
FS_IOC_{GETSET}FLAGS handling is merged?

						- Ted