[linux-lvm] Snapshot weirdness

[kelsey hudson <khudson@tsss.org>]

Hello.

I'm building a disk-to-disk-to-tape backup appliance here, and decided 
that for maximum flexibility I'd use LVM2 (mainly because of its 
snapshot feature and the ability to hot-add disks and extend volumes 
seamlessly. Good stuff.)

Anyhow, I have a 600GB primary physical volume configured with a single 
logical volume utilizing 99% of the extents. I have the system set to 
take a snapshot every night so there's always a live copy of the data 
available for backup. Three such snapshots are used in rotation (the 
oldest snapshot is deleted and recreated as the newest); each occupy 25 
extents. The problem is, after some time, I'll have a bunch of errors 
regarding the snapshot volumes spewed to the system logs and console. If 
I subsequently try and read from the filesystem, the kernel shuts the 
filesystem down (XFS feature).

This makes it rather inconvenient to back up a snapshot -- if I can't 
read it, it doesn't do me much good to have it. I'm basically using the 
snapshots read-only, and the filesystems are mounted as such, as well.

So, can anyone shed some insight on why I have self-corrupting 
filesystems on my snapshot volumes?

Thanks in advance.
-Kelsey