Advice on FS corruption

Hi

I've been running btrfs in various VMs for a while, and periodically I've experienced corruption in the filesystems being used. None of the data is important, but I'd like to track down how the corruption occurred in the first place.

Trying to mount any of the corrupt filesystems fails with an error of this form:

[   47.805146] device label baserock devid 1 transid 90 /dev/sdb1
[   47.810073] btrfs: disk space caching is enabled
[   47.817261] parent transid verify failed on 1636728832 wanted 76 found 95
[   47.818081] parent transid verify failed on 1636728832 wanted 76 found 95
[   47.818522] Failed to read block groups: -5
[   47.826103] btrfs: open_ctree failed

This is with Linux master as of 29/Aug/2012, so including the latest 'for-linus' branch from the btrfs tree. Attempts to run btrfs-debug-tree on the disk images fail with the same error, and btrfsck segfaults.

I've not yet been able to reliably reproduce the cause of the corruption, but I know that in at least one case the VM was compiling code and then had a forced power-off. However, in at least one case the corruption appeared after a clean shut down. I have a suspicion that it may be linked with suspending the host machine and thus causing weird things to happen to time in the VM's universe, since btrfs has been working fine with the same kernel in a VM on a machine that is never suspended or powered off, but I've not yet managed to prove anything.

I'll keep trying to reproduce the issue, in the mean time I'm interested how common this sort of issue is and if anyone has any tips for repairing the image, or if there is work in progress to prevent or fix the corruption.

Thanks
Sam

Try mounting with -o recovery.

[I'm resending this mail to list instead of as a personal reply, sorry]

Thanks, this gets more interesting!

For two of the FS's I got the exact same error message.

For a much larger (40GB) filesystem the recovery silently succeeded. At this point I ran 'find' in the root directory, which gave frequent:

find: ./foo: Input/output error

messages for various small files. I aborted and found all this in dmesg:

[   29.498581] device fsid 7aaaea86-e354-46f7-aa9e-2278c858170a devid 1 transid 35 /dev/sdb1
[   42.937330] parent transid verify failed on 31920128 wanted 9 found 26
[   42.961755] parent transid verify failed on 31920128 wanted 9 found 26
[   42.999560] parent transid verify failed on 31875072 wanted 9 found 26
[   43.035490] parent transid verify failed on 31875072 wanted 9 found 26
[   43.078782] parent transid verify failed on 31907840 wanted 9 found 26
[   43.079767] parent transid verify failed on 31907840 wanted 9 found 26
[   43.081685] parent transid verify failed on 31920128 wanted 9 found 26
[   43.082478] parent transid verify failed on 31920128 wanted 9 found 26
[   43.110576] parent transid verify failed on 31952896 wanted 9 found 27
[   43.112616] parent transid verify failed on 31952896 wanted 9 found 27

So, it seems to have improved matters, but am I correct in thinking this FS would now only be suitable for extracting as much of the data as possible and then discarding the whole thing? Or is the intention that an FS in such a state should be recovered to the point of being usable again?

Thanks
Sam