RPC/NFS problems

[Richard Munn]

I have a C3600 running HP-UX 10.20 which is persistantly getting errors when doing NFS mounts during bootup.

When the NFS client starts the following is recorded in rc.log:

mounting remote NFS file systems ...
:: RPC_PMAP_FAILURE - RPC_TIMED_OUT

FAILURE CODE: 1
> Mount failed! Possible local filesystem mount.
> If local filesystem, then nfs.server will mount it.

After the system has finished booting however the NFS fielsystem can be mounted without error. The NFS server is a netapps box on a different subnet. The box is connected to a CISCO switch and we found that if PORTFAST is set on the error is avoided some of the time. The only thing PORTFAST does is if the controller drop carrier is prevents a spanning tree rebuild in the switch which takes about 50 seconds. But even so, this does not stop the error all the time. However, the network folks are not keen to have PORTFAST set on.

If the delay caused by the spaning tree rebuild is causing the RPC errors, the RPC timeout must be very short or am I missing something here?

[Steven E. Protter]

1) check /etc/fstab for consistency and correctness of mounts.

2) Make sure you are up to date on NFS patches.

3) A word on Cisco switches and HP boxes.

HP NIC's don't like Cisco switches set to auto negotiate. All kinds of nasty things happen and the connections are not reliable.

Have the Cisco admin set the port to the HP NIC's expected settings. If thats 100 BaseT Full Duplex, have him/her do it explicitly.

After thats done, boot the HP box and check this command.

lanadmin -x 0

Replace the 0 with the actual number of the NIC in question. See if that matches spec. If not, hard code the expectations in one of the following files, depending on what kind of card you have.

/etc/rc.config.d/hpbtlanconf (not likely).
/etc/rc.config.d/hpgslanconf (not sure of the spelling but its documented).

Interesting situation you have, make sure you post the solution when you get there.

SEP

[Richard Munn]

fstab is OK, after the boot a mount -a will fill in the missing mounts (and not all NFS mounts go missing). The ports on the CISCO not autonegotiate and match the HP box. Unless there is something new all the right patches should be on.

[James A. Donovan]

Since you're HP box is 10.20, you're using UDP protocol for NFS. You're router may be having issues with routing the UDP datagrams.

Also, and this may be more likely, you should check the OS version on the NetApps box, and check with NetApps to see if there are any issues.

Finally, as a possible workaround, try pinging the NetApps box just prior to running the NFS mount scripts.