I'm having trouble with NFS on a Fedora 3 server. The client machines are a mixture of HP Unix and Solaris (also samba is used for Windows clients). Most of the time the NFS export point runs perfectly fine, a client have it mounted all the time, and files and directories can be listed as expected. All users have write access.

However, occasionally (about once a day) the files in various directories disappear from the client machines. If I go to the server and cd to the troublesome directories, they immediately become available to the clients again.

The only error I get in the logs is: Jun 13 07:39:58 kelvin rpc.statd[2248]: Received erroneous SM_UNMON request from kelvin.test for 172.16.44.23 but this indicates a write access problem with /var/lib/nfs/statd/sm and sm.d. But I have checked that the rpc users has r/w access to these.

I tried a tcpdump and got the following:

13:27:36.751186 IP (tos 0x0, ttl 64, id 52539, offset 0, flags [DF], proto 6, length: 148) tescat2.680050 > kelvin.test.nfs:108 lookup [|nfs] 13:27:36.751223 IP (tos 0x0, ttl 64, id 14271, offset 0, flags [DF], proto 6, length: 160) kelvin.test.nfs > tescat2.680050:reply ok 120 lookup ERROR: No such file or directory post dattr: 13:27:36.751646 IP (tos 0x0, ttl 64, id 52540, offset 0, flags [DF], proto 6, length: 132) tescat2.680051 > kelvin.test.nfs:92 getattr [|nfs]

Based on this I have tried turning off client side attribute caching, to no effect.

I have exported the directories as such:

/data 172.16.44.0/24 (rw,async,insecure,insecure_locks,no_subtree_check)

I have also tried sync instead of async, and removing the no_subtree_check flag, also to no effect. Other things I have tried: reducing the MTU on the network to 1000 from 1500

Any ideas would be appreciated.

Many thanks,

Stuart.