Current NFS issues (2.5.59)

David Ford (david+powerix@blue-labs.org)
Sun, 09 Feb 2003 15:18:46 -0800


Ok. Here goes. I have two servers that NFS mount from each other and
provide.

Server 1 exports A, B, and C to server 2. Server 2 exports D and E back
to server 1 and exports F and G to two other clients. Each of these
(A-G) are distinctly different filesystem paths and not part of each other.

1. If server 1 is restarted, server 2 will invalidate (make all 'df'
values '1') F and G. This requires an 'exportfs -vra' or similar on
server 2 to fix the client 'df' values. The client doesn't need to do
anything.

2. Repeated nfs system stops and starts (/etc/init.d/nfs restart) will
eventually cause a kernel panic on server 2 (haven't tested on server
1). The number of restarts is variable.

3. Mount point F (/home/david) infrequently loops. ls -la /home/david
will loop forever until all client memory is exhausted and the kernel
kills it via OOM. ls -la /home/david/somefile or /home/david/somedir/
works just fine as well as any sub directory under /home/david.
Restarts of both systems refuse to fix things.

4. Mounts infrequently get "permission denied" messages on the client
with a " rpc.mountd: getfh failed: Operation not permitted" message on
the server. This is fixable by restarting the nfs system on the server.

Server1 is UNI, server 2 is SMP. All servers and clients are stock
2.5.59[1]. NFS is running on top of Reiserfs filesystems on all client
and server machines.

I'll be happy to apply test patches to either clients or servers.

David

[1] One client is 2.5.56 but it rarely accesses the NFS mount unlike the
other machines which use them constantly

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/