[rhn-users] strange problem with NFS and/or automount and/or autofs

[Brian C. Hill]

	Hi Red Hatters,

	Sorry, this is a little long, but maybe this problem will jump
out at you.  This NFS problem is hard to describe, since it seems to
bounce around, but I will try. First, note that all systems involved
are completely up2date.

Basic architecture:

* 1 'server' system with 2 IP addresses running NFS with 2 exported
filesystems (at the same level - /data/a and /data/b)

* 4 'client' systems, each with 1 IP address using autofs via /net to
mount filesystems from the 'server' system running NFS

* all run Red Hat Enterprise ES 3.0

Problem:

We see a variety of problems after rebooting the NFS server that are
resolved in in one or more ways including:

* restarting autofs on the clients
* running exportfs -av on the NFS server
* restarting NFS on the server (even though it has just been rebooted)
* rebooting one or more (though seldom all) client systems

Since the steps necessary to restore access after a reboot vary from
episode to episode, the problem seems to be a moving target and is
really hard to debug. The only error message that seems to appear is
this one on the NFS server:

getfh failed: Operation not permitted

The only other odd symptoms that often presents themselves are:

* sometimes, when the system cannot be accessed via /net, it is still
possible to do a mount command by hand (though not always)

* sometimes, access to only one of the servers IP addresses fails via
/net while both of them are usable with a mount command

* sometimes, both /net and manual mount fail with a 'permission
denied', which is, again, sometimes fixed by an exportfs -a or -r on
the NFS server and sometimes not.

* sometimes /data/a can be accessed, but not /data/b and vice versa.

It is also strange that this usually only seems to affect some of the
clients (usually 2 out of 4). It does generally seem to be the same 2
that fail, but again, it is not always those 2 or only those 2.

I have exhausted many hours trying to figure out this problem (via Red
Hat faqs, google, etc.).

Clues?

Thanks. :)

Brian