RE: oops umounting full LVM snapshots

["Stephenson, Dale" <dstephenson@snapserver.com>]

Eric,

I followed your steps and did not get the oops.  It seems that the mount
options on the original xfs source volume matter.  I was able to recreate
the oops with the same steps by mounting with quota support.

mount -o quota /dev/volgroup/logicalvol /mnt/lvm/logicalvol
  or
mount -o usrquota,grpquota,quota /dev/volgroup/logicalvol
/mnt/lvm/logicalvol

I also found (accidentally) that after the first snapshot umount goes
through xfs_force_shutdown(), an oops can be triggered by umounting the
original source volume--not just umounting its other snapshot. It doesn't
xfs_force_shutdown the source volume, but oops in the same spot just the
same.

Dale Stephenson
steph@snapserver.com 

Eric Sandeen wrote:
> Hi Dale - a quick try at reproducing this here was unsuccessful.  This
> is what I did:
> 
> Create volume group & logical volume:
> 
> vgcreate volgroup /dev/sda2 /dev/sda3
> lvcreate -L 100M -n logicalvol volgroup
> 
> mkfs and mount it:
> 
> mkfs.xfs -f /dev/volgroup/logicalvol 
> mount /dev/volgroup/logicalvol /mnt/lvm/logicalvol
> 
> Make a couple snapshots and mount them:
> 
> lvcreate --size 16m --snapshot --name snap1 /dev/volgroup/logicalvol 
> lvcreate --size 16m --snapshot --name snap2 /dev/volgroup/logicalvol 
> mount -o nouuid,ro,norecovery /dev/volgroup/snap1 /mnt/lvm/snap1
> mount -o nouuid,ro,norecovery /dev/volgroup/snap2 /mnt/lvm/snap2
> 
> Overflow the snapshots:
> 
> cp -vaR /usr/src/linux-2.4.9-13SGI_XFS_1.0.2/ .
> 
> Unmount the snapshots:
> 
> umount /mnt/lvm/snap1/
> umount /mnt/lvm/snap2/
> 
> I got no I/O errors or oopses...  This is with a fresh CVS 
> checkout from
> today.
> 
> -Eric
> 
> 
> On Mon, 2002-02-25 at 10:44, Stephenson, Dale wrote:
> > At Keith Owens' suggestion, I've rerun the oops output 
> using the System Map
> > and a more recent ksymoops.
> > This oops was generated by filling two mounted LVM 
> snapshots of an XFS
> > filesystem, and then umounting them both.  Both umounts 
> resulted in I/O
> > errors and xfs_force_shutdown calls, and the second umount 
> oopsed.  The
> > snapshots are a read only device, and they were mounted
> > ro,nouuid,norecovery.
> 
> 
> -- 
> Eric Sandeen      XFS for Linux     http://oss.sgi.com/projects/xfs
> sandeen@sgi.com   SGI, Inc.
>