To get rid of a "not responding" NFS mount

[Shahul]

Hi,

I am never a winner in the game with stale NFS. I have a NFS client with some remote filesystems mounted. But the NFS server is disappeared for some reason. Now if I do a bdf in client, it hangs, if I do a cd to mount point, it hangs. All I want is to unmount or remove the NFS mounted filesystem.

I have found some HP docs, which advise to do these steps.

Stop NFS client
Stop NFS server
rm -r /etc/sm
rm -r /etc/sm.bak
start NFS client
start NFS server

In my system (10.20), it does not have /etc/sm or sm.bak, instead I have /var/statmon/sm and sm.bak.

Is it allright to delete this? Or any other way to tackle this state?

[Ajitkumar Rane]

Shahul,

The derictories are generated by statd
Each file in /var/statmon/sm represents one or more machines to be monitored by the statd daemon.
Each file in /var/statmon/sm.bak represents one or more machines to be notified by the statd daemon upon its recovery.

I would suggest not to delete but to move the files to a different name.

So far I havent heard of such problems resolved without a reboot. remember to comment the lines for nfs client and the remote filesystems b4 u reboot.

BTW if you try the above procedure and it works do not forget to put post the solution.


Rgds,

Ajit

[Jakes Louw]

Nope, a reboot is usually where it ends....
When, oh, when, is HP going to fix this problem? Ever since NFS v 3 was put into HP-UX, we've had this problem. NO OTHER mainline UNIX OS has the same problem!
In fact, Solaris offers "umount -f" to force the dismount!
I've used soft options, and sometimes that helps, but surely this is a HUGE bug in the NFS client code?

[Alessandro Bocchino]

As far as I know, the client tries indefinitely to contact the server to recover the situation.

Stopping services in an NFS stale situation usually doesn't help - it can lead to an unrecoverable situation unless you reboot.

As a trick if you just need to dismount the
filesystem and the remote server is dead or similar you can assing the server IP as a virtual IP to something else, just to have a responding IP, kill all the processes keeping busy the client mount point, then try again the dismount.

Alessandro Bocchino
http://www.risolve.com

[Chan 007]

Hi Shahul,

You would have tried all options by now. But on 10.20, the only option is "reboot".

You can see the writeup for this PHCO_9543.

Some say that you can apply patches, but even when you apply patches, they require reboot, but never seen solving the problem in 10.20

Hope you can have a reboot & this helps.

Also have this thread for your help

http://forums1.itrc.hp.com/service/forums/bizsupport/questionanswer.do?threadId=69647

Chan

[John E.Ophious]

Greetings Shahul,

I support a handful of 10.20 systems and like everyone else here I have to reboot to solve this issue. Like Jakes mentioned, you can force the unmount in a solaris system, but alas 10.20 does not have this wonderful option. I know it sounds too "windowsish", but I'd just reboot.

Off into the sunset!

John E. Ophious

[Steven E. Protter]

Shalom,

There is one way, and one way only with the OS you mentioned to prevent this.

Stop the NFS server from disappearing.

If it can be upgraded to 11i v2 then it can handle NFS v4, which handles connections differently.

If the remote server were a recent Linux distro, the same would be true.

Stale mount is caused by the NFS server going away. That can be due to network issues, issues on the server or the client box (10.20) simply disconnecting because its too busy.

SEP

[Shahul]

Thanks to all....

Yes, it normally end up in a reboot.. that's what I hate to do.

Regards
Shahul

[Olivier Masse]

Dave Olker replied to me once in this forum that a umount -f command was coming. 11.30? Maybe, I don't remember.In the mean time, soft mounts are the only way to be on the safe side, assuming your application can live with it.

[Christian Schulze]

Hi,
some tips which might help to get rid of this:
make a soft mount (mount ... -o soft)

you might be able to umount it after issuing a
#fuser -ku server.domain:/filesystem

Be sure to run the fuser against the server:/filesystem, not against your /local_mountpoint.

There is a Document somewhere around, which describes a lot of NFS-tips and tricks (something like "forcibly unmounting nfs filesystems")
If you cannot find it, i will upload it.

Christian