Isn't there a way to save a burning NFS ?

[Fred Ruffet]

Hi !

I never liked NFS particularly, but since this morning, it became worse...

I have two problems with it.

One is a configuration problem I cannot help resolving : on server A, I can mount an exported Windows share with CIFS. On server B, I can't. Problem is that configurations are identical (NIS in action). Error is :

Bserv:/>mount -F cifs NTServ:share /mount_point
Bserv:/>cd /mount_point
NFS access failed for server NTServ: RPC: Remote system error
sh: share: The specified directory is not valid.

(On server A it runs perfectly)

Second problem is a not unmounted share on a poweroff server...

Bserv:/>mount
...
/mount_point2 on localhost:\\AnotherNT\share soft,noac,retrans=3,timeo=205,acregmin=0,acregmax=0,acdirmin=0,acdirmax=0,NFSv3 on Wed Dec 3 11:47:47 2003
...
Bserv:/>umount /mount_point2
Unmounting Share: Path not mounted!
Bserv:/>bdf /mount_point2
NFS getattr failed for server AnotherNT: RPC: Timed out
bdf: /mount_point2: Connection timed out

I believe I could resolve first problem by restarting CIFS/NFS/RPC, but second problem always keeps up nfskd and prevents from restarting clean NFS.

So, finally, the question : Is there a way to kill a crashed CIFS/NFS mount ?

All of these happening after a server crash on a cluster, CIFS mounts being a part of the node start...

Regards,

Fred

[Pete Randall]

Fred,

This sounds similar to the "stale NFS find handle" that I get on clients when an NFS server goes down. The latest NFS/CIFS patches will probably address this issue. In my case, it was PHNE_29883 - but I don't have CIFS in the mix.


Pete

[Pete Randall]

Fred,

Here's another thought along the patches line.
Have you got the mountall cumulative patch, PHCO_27958 installed?


Pete

[Fred Ruffet]

Pete,

I have none of those 2 patches, but I imagine I won't be able to install a patch within this situation (need reboot ?).

Another point I didn't mention : HP-UX 11.00.

Thanks,

Fred

[Pete Randall]

Fred,

For 11.0, then, searching the patch database for "CIFS", I came up with PHSS_28953, PHSS_30575, and PHSS_29299 (recommended) or PHSS_30687 (most recent). The NFS patch becomes PHNE_20371 (recommended) or PHNE_29882 (most recent).

I would make sure that I checked out all those, none of the PHSS patches require a reboot - the PHNE patches do.


Pete

[Fred Ruffet]

Pete,

I don't know wether I should focus on CIFS or NFS. Anyway, if I patch CIFS or NFS, in order to run the patched version, I will need to restart the whole chain and this will be done only by killing this nfskd process... and I still don't know how to do but reboot.

Nevermind, part of my problems will end up in a few minutes : I will switch my cluster package to Aserv :) Will only stay my dead NFS connection...

Thanks,

Fred

[Dave Olker]

Hi Fred,

Regarding the second problem, we will hopefully be providing a solution in 11i v2 update 2 (aka 11i v2 0409) where we will be adding support for a "forcible" option to the unmount command. This will allow you to forcibly unmount a hung or stale NFS filesystem without having to reboot the client.

In the mean time, I've written a white paper that discusses ways of getting hung or stale NFS filesystems to unmount. If your interested it's available at: http://docs.hp.com/hpux/onlinedocs/3929/ForciblyUnmountingNFSFilesystems.pdf.

As for the first problem, it could be an NFS issue or a CIFS problem. As you likely know, our current HP CIFS client product leverages the NFS client to send RPC requests, so if either one is having problems it could affect this. If you still want to troubleshoot this problem further I can try to offer some advice. However, if you're switching packages to another server then this may not be a concern any longer.

Best of luck,

Dave

[Mohanasundaram_1]

Hi Dave,

The Whitepaper is not available on the URL you provided. Can you check this and confirm?
It should be a very useful whitepaper to refer.

Appreciate your time on this.

Cheers,
Mohan.

[Dave Olker]

Hi Mohan,

The above link is correct, you just have to strip off the trailing '.', so the real link is:

http://docs.hp.com/hpux/onlinedocs/3929/ForciblyUnmountingNFSFilesystems.pdf

Regards,

Dave

[Fred Ruffet]

Dave,

Breaking an NFS mount will be a great improvement. Of course, I can not upgrade now, but it will probably help one day.

Switching the package make me come back to a better situation on Aserver, not for the broken NFS on Bserver. I planned a reboot this evening, as that's the only way I can found. It will be clean after that.

Thanks,

Fred