[CentOS] Quota (and disk usage) is incorrectly reported on nfs client mounting XFS filesystem

Sat Jun 25 00:17:12 UTC 2011
Julie Ashworth <julie.ashworth at berkeley.edu>

hi all,
I hope you can help me with a strange quota/nfs/XFS behavior...
 
I'm using Centos 5.6 on both a nfs server and client. Both are 
64-bit, and using a recent kernel:

NFS server:
Linux fs2.priv 2.6.18-238.12.1.el5 #1 SMP Tue May 31 13:22:04 EDT 2011 x86_64 x86_64 x86_64 GNU/Linux

NFS client:
Linux nx8.priv 2.6.18-238.12.1.el5 #1 SMP Tue May 31 13:22:04 EDT 2011 x86_64 x86_64 x86_64 GNU/Linux

The NFS server is exporting a XFS filesystem:
Filesystem            Size  Used Avail Use% Mounted on
/dev/mapper/vgXX-lvXX  24T   16T  8.2T  66% /export



User foo (for anonymity) added ~3TB of data in the last 2-3 days. 
On the NFS server, her quota is reported as ~5.8 TB:

Disk quotas for user foo (uid 1314): 
     Filesystem  blocks   quota   limit   grace   files   quota   limit   grace
/dev/mapper/vgXXX-lvXXX
                6129681348  7490000000       0         6991914       0       0    


On a NFS client, her quota is reported as ~1.8TB:

Disk quotas for user foo (uid 1314): 
     Filesystem  blocks   quota   limit   grace   files   quota   limit   grace
    fs2:/XXXXX   1831504512  3195032704       0         6988049       0       0       


When I modify her quota on the NFS sever, it's not reflected 
on the NFS client.

This problem is specific user foo. There are about 20 other 
users with quotas on this NFS server, and if I change one of 
their quotas, the quota/usage is correctly (and instantly) 
reflected on the NFS client.

rpc.quotad is configured to use port 875, and I can see this
is working with strace. And tcpdump shows an established 
connection:
16:50:04.480757 IP nx8.priv.pssc > fs2.priv.875: UDP, length 116
16:50:04.481015 IP fs2.priv.875 > nx8.priv.pssc: UDP, length 68

Any ideas?

Thanks!
Best,
Julie


-- 
Julie Ashworth <julie.ashworth at berkeley.edu>
http://www.neuro.berkeley.edu
PGP Key ID: 0x17F013D2