NFS error: Client kills mountd on server!

[J Andersen]

I have just been help excellently by Dave Olker with a similar problem, so I am cautiously optimistic about this one.

Here goes: I have a newly installed rx3600 running hp-ux 11.31. It is meant to mount two NFS filesystems from a Linux box (Redhat AS 3), but as soon as I try, rpc.mountd dies on the NFS server. Wireshark shows that it only gets as far as contacting mountd and send some parameters, then it dies.

I would have thought this was a server side problem, were it not because we have >20 UNIXes of all descriptions mounting the same directories.

[Dave Olker · Accepted solution]

Hi Jan,

As last time, I'll want to know everything about these systems so I can try to duplicate the problem on my systems.

First, what version of the kernel is the Redhat server running? What version of the nfs-utils RPM package is installed on the Redhat server? How exactly (exact syntax) is the RHAS3 system exporting the directory? How exactly (exact syntax) is the 11.31 system trying to mount the filesystem?

I would also want to see the Wireshark trace of the mount request that causes rpc.mountd to die on the server. Additionally, I would like you to collect a second Wireshark trace of another NFS client successfully mounting the same directory from the same server.

If you can provide the exact steps from the client, server, and the two traces (working and failing) it will go a long way towards helping me get to the bottom of the problem.

Regards,

Dave

[J Andersen]

Hi again Dave!

The Redhat server: kernel 2.4.21-4.ELsmp, nfs-utils nfs-utils-1.0.5-3.

/etc/exports:

/dtl *(rw,insecure,sync,no_root_squash)
/home1 *(rw,insecure,sync,no_root_squash)

These are the entries in /etc/fstab on the client (bldsrv is the Redhat server):

bldsrv:/home1 /home1 nfs defaults 0 0
bldsrv:/dtl /dtl nfs defaults 0 0

I have tried different options, but the result is the same.

Another server (PA-RISC, hp-ux 11.23) that can mount the filesystems has the following entries in /etc/fstab:

bldsrv:/home1 /home1 nfs rw,suid 0 0
bldsrv:/dtl /dtl nfs rw,suid 0 0

I have run a trace for the mount that fails (bad.trc) and one for the successful one (good.trc), which I'll try to attach to this entry as a tar file.

[Dave Olker]

Hi Jan,

Have you tried loading the latest nfs-utils bundle on the Linux server to see if this is a known problem fixed on the Linux side? I tried duplicating this problem on my RH3 system but I was unable to - possibly because I'm running the latest nfs-utils bundle on my system.

I looked at the traces and I do see a couple differences, but one thing I noticed is the failing mount is for /home1 and the working mount you traced is for /dtl. Would it be possible for you to collect a new set of traces where each system attempts to mount the same filesystem from this server?

Ideally I'd like to compare two traces where *everything* is identical except for the client sending the mount request.

Regards,

Dave

[Dave Olker]

Hi Jan,

I checked the SourceForge NFS database and found multiple cases where rpc.mountd dies when servicing mount requests. These defects have been addressed in later versions of the nfs-utils package.

I have no way of knowing if the bug you're seeing is the same as the resolved ones but I'd strongly recommend getting the latest nfs-utils on your system and see if that resolves the problem.

Regards,

Dave

[J Andersen]

That sounds promising, I'll give that a try, and if it doesn't help, I'll go and make some new traces.