Re: [PATCH 1/7] ocfs2: test reflinking to inline data files

From: "Darrick J. Wong" <darrick.wong@oracle.com>
To: Eryu Guan <eguan@redhat.com>
Cc: ocfs2-devel@oss.oracle.com, fstests@vger.kernel.org
Subject: Re: [PATCH 1/7] ocfs2: test reflinking to inline data files
Date: Tue, 13 Dec 2016 13:35:41 -0800	[thread overview]
Message-ID: <20161213213540.GC6366@birch.djwong.org> (raw)
In-Reply-To: <20161213071136.GB6366@birch.djwong.org>

On Mon, Dec 12, 2016 at 11:11:36PM -0800, Darrick J. Wong wrote:
> On Tue, Dec 13, 2016 at 11:20:57AM +0800, Eryu Guan wrote:
> > On Mon, Dec 12, 2016 at 10:09:22AM -0800, Darrick J. Wong wrote:
> > > On Mon, Dec 12, 2016 at 05:01:20PM +0800, Eryu Guan wrote:
> > > > On Sun, Dec 11, 2016 at 01:52:51PM -0800, Darrick J. Wong wrote:
> > > > > Make sure that we can handle reflinking from and to inline-data files.
> > > > > 
> > > > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > > > ---
> > > > >  common/reflink       |    2 +
> > > > >  tests/ocfs2/001      |   88 ++++++++++++++++++++++++++++++++++++++++++++++++++
> > > > >  tests/ocfs2/001.out  |   14 ++++++++
> > > > >  tests/ocfs2/Makefile |   20 +++++++++++
> > > > >  tests/ocfs2/group    |    1 +
> > > > >  5 files changed, 124 insertions(+), 1 deletion(-)
> > > > >  create mode 100755 tests/ocfs2/001
> > > > >  create mode 100644 tests/ocfs2/001.out
> > > > >  create mode 100644 tests/ocfs2/Makefile
> > > > >  create mode 100644 tests/ocfs2/group
> > > > > 
> > > > > 
> > > > > diff --git a/common/reflink b/common/reflink
> > > > > index 9d51729..d048045 100644
> > > > > --- a/common/reflink
> > > > > +++ b/common/reflink
> > > > > @@ -197,7 +197,7 @@ _cp_reflink() {
> > > > >  	file1="$1"
> > > > >  	file2="$2"
> > > > >  
> > > > > -	cp --reflink=always -p "$file1" "$file2"
> > > > > +	cp --reflink=always -p -f "$file1" "$file2"
> > > > 
> > > > I'm still seeing "File exists" error with this patch, tested with your
> > > > ocfs2-vfs-reflink-6 branch, compiled on openSUSE Tumbleweed.
> > > > 
> > > > FSTYP         -- ocfs2
> > > > PLATFORM      -- Linux/x86_64 bootp-73-5-234 4.9.0-rc8.djwong-ocfs2+
> > > > MKFS_OPTIONS  -- --fs-features=local /dev/sda5
> > > > MOUNT_OPTIONS -- /dev/sda5 /mnt/testarea/scratch
> > > > 
> > > > ocfs2/001        - output mismatch (see /root/xfstests/results//ocfs2/ocfs2/001.out.bad)
> > > >     --- tests/ocfs2/001.out     2016-12-12 13:51:38.053909486 +0800
> > > >     +++ /root/xfstests/results//ocfs2/ocfs2/001.out.bad 2016-12-12 16:49:35.038882697 +0800
> > > >     @@ -1,14 +1,18 @@
> > > >      QA output created by 001
> > > >     +mkfs.ocfs2 1.8.4
> > > 
> > > I guess I'll have to neuter this on newer ocfs2-tools.  (Still running 1.6.4
> > > here).
> > 
> > I have a patch pending on review that could fix & filter this, FYI
> > 
> > fstests: teach _scratch_mkfs to handle mkfs option conflicts
> > https://www.spinics.net/lists/fstests/msg04550.html
> 
> <nod> I actually meant the printing of the mkfs version string.
> 
> > 
> > > 
> > > >      Format and mount
> > > >      Create the original files
> > > >      reflink into the start of file2
> > > >     +cp: failed to reflink '/mnt/testarea/scratch/test-001/file2' from '/mnt/testarea/scratch/test-001/file1': File exists
> > > >      reflink past the stuff in file3
> > > >     ...
> > > >     (Run 'diff -u tests/ocfs2/001.out /root/xfstests/results//ocfs2/ocfs2/001.out.bad'  to see the entire diff)
> > > > 
> > > > --- tests/ocfs2/001.out 2016-12-12 13:51:38.053909486 +0800
> > > > +++ /root/xfstests/results//ocfs2/ocfs2/001.out.bad     2016-12-12 16:44:31.991943842 +0800
> > > > @@ -1,14 +1,18 @@
> > > >  QA output created by 001
> > > > +mkfs.ocfs2 1.8.4
> > > >  Format and mount
> > > >  Create the original files
> > > >  reflink into the start of file2
> > > > +cp: failed to reflink '/mnt/testarea/scratch/test-001/file2' from '/mnt/testarea/scratch/test-001/file1': File exists
> > > >  reflink past the stuff in file3
> > > >  reflink an inline-data file to a regular one
> > > > +cp: failed to reflink '/mnt/testarea/scratch/test-001/file5' from '/mnt/testarea/scratch/test-001/file4': File exists
> > > >  reflink an inline-data file to another inline-data file
> > > > +cp: failed to reflink '/mnt/testarea/scratch/test-001/file6' from '/mnt/testarea/scratch/test-001/file4': File exists
> > > >  Verify the whole mess
> > > >  2d61aa54b58c2e94403fb092c3dbc027  SCRATCH_MNT/test-001/file1
> > > > -2d61aa54b58c2e94403fb092c3dbc027  SCRATCH_MNT/test-001/file2
> > > > +401b30e3b8b5d629635a5c613cdb7919  SCRATCH_MNT/test-001/file2
> > > >  4e68a2e24b6b0f386ab39d01d902293d  SCRATCH_MNT/test-001/file3
> > > >  009520053b00386d1173f3988c55d192  SCRATCH_MNT/test-001/file4
> > > > -009520053b00386d1173f3988c55d192  SCRATCH_MNT/test-001/file5
> > > > -009520053b00386d1173f3988c55d192  SCRATCH_MNT/test-001/file6
> > > > +2d61aa54b58c2e94403fb092c3dbc027  SCRATCH_MNT/test-001/file5
> > > > +60b725f10c9c85c70d97880dfe8191b3  SCRATCH_MNT/test-001/file6
> > > > 
> > > > I did the "reflink into the start of file2" test manually, and strace
> > > > showed that it is file1 reports EEXIST.
> > > > 
> > > > open("/mnt/ocfs2/testfile1", O_RDONLY)  = 3
> > > > fstat(3, {st_mode=S_IFREG|0600, st_size=65536, ...}) = 0
> > > > ioctl(3, _IOC(_IOC_WRITE, 0x6f, 0x04, 0x18), 0x7ffc8daf09f0) = -1 EEXIST (File exists)
> > > 
> > > Wait, that's the ocfs2 reflink ioctl, not FICLONE.  Well that explains
> > > why the functional parts of the tests fail.  I looked at latest coreutils
> > > source, which does not use OCFS2_IOC_REFLINK, so I guess this must be an
> > > opensuse thing?
> > 
> > Ok, I'll try a newer coreutils.
> 
> coreutils 8.26 (latest) doesn't have any OCFS2_IOC_REFLINK support; I
> suspect a distro patch in the srpm or something....

Confirmed.  SuSE's coreutils package has a patch that tries the ocfs2
reflink ioctl and bails out of cp without trying the btrfs/vfs clone
ioctl if the ocfs2 ioctl returns EEXIST.  I suppose that'll have to get
fixed in their coreutils package, but until then it'll just be broken.
:/

FWIW I also confirm that none of (upstream, RHEL, OL, Debian, or Ubuntu)
have this patch, so this shouldn't be a problem on any of them.

--D

> 
> --D
> 
> > 
> > Thanks,
> > Eryu
> > --
> > To unsubscribe from this list: send the line "unsubscribe fstests" in
> > the body of a message to majordomo@vger.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> --
> To unsubscribe from this list: send the line "unsubscribe fstests" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html