corrupt dinode but xfs_check and xfs_repair are bit detecting any problem and are fixing nothing.

[Zoran Cvetkovic]

Hi Fellows,

I am running a 2.6.26 kernel on debian lenny having a 6 TB data 
partition under xfs, 2.6 TB are in use now.
xfs fs is exported via nfs from that box.

after 12 weeks of popper service suddenly nfs performance was a disaster 
and the box was stuttering, system load was about 10.
There was no indication  for a  hardware problem. HW Raid status was *OK*

/var/log/messages lists several thousand lines like:

Aug 6 21:12:03 nfsserver kernel: [7660448.840139] Filesystem "sdb1": 
corrupt dinode 4806258, (btree extents). Unmount and run xfs_repair.

(allways the same dinode #)

after stopping nfsd and unmounting the fs

both xfs_check and xfs_repaier did not detect any corruption!

back in service i got the same problem and log lines.

So what is wrong here?

There might be a fs issue and the fs utilitys are not able to detect it ?

I did a metadata dump of that fs and could  post it, would this help to 
determine what is going wrong here?

Regards,
Zoran Cvetkovic