hi Lars,
On Mar 6, 9:16pm, Lars Kellogg-Stedman wrote:
> Subject: More raid5 issues: Kernel panic running xfs_db
> I got a kernel panic trying to run xfs_admin earlier today to label some
> filesytems. Further experiments revealed:
>
> (a) The panic is apparently caused when xfs_admin calls xfs_db, and
> (b) The panic only occurs on the soft raid5 volume. Other filesystems
> presented no problems.
>
> The kernel panic looked like:
>
> Mar 6 20:08:40 flowers kernel: Kernel panic: Invalid blocksize passed to
> set_blocksize
>
> Which looks like it's coming from fs/buffer.c.
>
> The xfs device that triggered this was /dev/md/0, which is a soft raid5
> volume over 4 scsi disks.
>
Were you were running xfs_db (via xfs_admin script) on a
mounted filesystem? That shouldn't work at all - would be a
bug in xfs_db if it tries to write to a filesystem while its
mounted.
Hmm - just tried it - seems to do the right thing...
# sh -x xfs_admin -L test /dev/hdb10
+ OPTS=
+ USAGE=Usage: xfs_admin [-flu] [-L label] [-U uuid] special
+ getopts fluL:U: c
+ OPTS= -c 'label test'
+ getopts fluL:U: c
+ set -- extra -L test /dev/hdb10
+ shift 3
+ eval xfs_db -x -p xfs_admin -c ''\''label' 'test'\''' /dev/hdb10
++ xfs_db -x -p xfs_admin -c 'label test' /dev/hdb10
xfs_admin: /dev/hdb10 contains a mounted filesystem
fatal error -- couldn't initialize XFS library
+ status=1
+ exit 1
So I guess you're not using a mounted filesystem - in which
case, xfs_db is doing writes to the raw device, not via xfs.
Perhaps the use of xfs_db/xfs_admin is a red herring - do
you see the panic while the filesystem is mounted? (after
you've run xfs_db & mount). Are you using Russell/Martin's
md patch?
cheers.
--
Nathan
|