[linux-lvm] Snapshotting concatenated devices.

[Jay Weber]

Hi folks,

Here's an odd one for you.  When using concatenated logical volumes (an
LV that consists of multiple PVs in the VG) snapshots don't appears to be
proper after they are taken.  It doesn't generate any errors when the
snapshot is performed.  But, if you dumpe2fs or attempt to mount the
snapshot it doesn't see the filesystem superblock, etc.

To expand on this a bit further, the above isn't always true.  It only
seems to occur when I use a larger volume which uses larger extent sizes.
The larger volume in total consists of about 400G in the VG.  So, for
example, I create an LV of 350G and leave myself 50G for snapshot
purposes.

It does work properly for me using my memorystick, which is a total of 16M
and I use an extent size of 1M.

In both cases I just make two partitions on the underlying device and
concat them into the one VG.

Lastly, in the case where snapshots aren't readable after being created,
there is one step of events I can perform to make the snapshot usable
again.  It consists of:

	umount the LV
	vgchange -an VG
	vghcnage -ay VG
	mount the snapshot of the LV first
	then mount the LV

If I mount the LV first though, I end up with the unusable snapshot again.
In the above example after the snapshot is mounted, it appears to look
okay and appears to be functioning (COW, etc) properly.

Any ideas?

I haven't seen this occur on non concattenated devices.  We also seem to
have this issue when using striped devices.