Re: reiserfs broken in 2.6.32 was Re: [GIT PULL] reiserfs fixes

[Frederic Weisbecker <fweisbec@gmail.com>]

On Sat, Jan 02, 2010 at 10:27:58PM -0500, tytso@mit.edu wrote:
> On Sat, Jan 02, 2010 at 06:05:13PM -0800, Christian Kujau wrote:
> > On Sun, 3 Jan 2010 at 02:52, Frederic Weisbecker wrote:
> > > [: 53: ==: unexpected operator
> > > common.rc: Error: $TEST_DEV (/dev/sda3) is not a MOUNTED xfs filesystem
> > 
> > Yeah, I'm playing around with xfstests as well, but apparently they're 
> > assuming !/bin/sh will be run under /bin/bash, which is not always the 
> > case. A short fix is to link /bin/sh to /bin/bash, but perhaps some of the 
> > tests can be tweaked to run under /bin/sh as well.
> > 
> > > I'm not sure how I can run these tests on a non-xfs partitions.
> > > I must be missing something.
> > 
> > I haven't found an easy way to do this yet without rewriting a few 
> > routines (mkfs, mount, etc...). As Ted is already using xfstests for 
> > btrfs, ext4, maybe he wants to share his magic? :-)
> 
> I'm using it for ext4.  It looks like someone has already tried using
> the xfstests with reiserfs; take a look at common.rc; you'll see case
> statements for xfs, udf, nfs, ext2/3/4, reiserfs, and gfs2.  Someone
> who wants to use xfstests for some other file system may need to
> further edit common.rc.  I thought the btrfs folks were using it as
> well, but at least the kernel.org git tree for xfstests doesn't seem
> to show any btrfs references in common.rc, so perhaps I'm wrong about
> btrfs developers using xfstests (or they haven't sent their patches
> back upstream).  I'm not sure how well tested the reiserfs support is,
> so you may need to edit common.rc as necessary.
> 
> In any case, the README file has pretty much what you need.  I
> personally run my test kernels using KVM, and I have a run-test script
> which invokes check as follows:
> 
> #!/bin/sh
> export TEST_DEV=/dev/sdb
> export TEST_DIR=/test
> export SCRATCH_DEV=/dev/sdc1
> export SCRATCH_MNT=/scratch
> export EXT_MOUNT_OPTIONS="-o block_validity"
> exec ./check -ext4 $*
> 
> /dev/sdb is an ext4-formated filesystem, which you're supposed to not
> reformat from run to run, so that some testing can be done with an
> "aged" file system.  The scratch filesystem will be reformatted for
> various tests, so you shouldn't keep anything valueable on it.
> 
> I also have a 1k block file system on /dev/sdd, so I invoke check as
> follows to check to make sure things work when blocksize != pagesize:
> 
> #!/bin/sh
> export TEST_DEV=/dev/sdd
> export TEST_DIR=/test-1k
> export SCRATCH_DEV=/dev/sdc1
> export SCRATCH_MNT=/scratch
> export MKFS_OPTIONS="-b 1024"
> exec ./check -ext4 $*



Ok. Thanks a lot for all these details. It seems to be working
now that I have edited some shebang and replaced /bin/sh by
/bin/bash, like Christopher suggested.


 
> As far as the inconsistency between TEST_DIR versus SCRATCH_MNT ---
> all I can say is, the XFS engineers who threw together the xfstests
> scripts may have been very good file system engineers, but they
> obviously weren't very well used as UI/User Experience designers.  :-)



Agreed :)



> (The documentation isn't all that great either, but patches sent to
> xfs@oss.sgi.com do tend to be accepted and merged into the kernel.org
> tree if they are clean.)
> 
> Hope this helps,


Yeah, thanks!