NFS Mess -- How to get unstuck?

Hi:

I have a Suse 8.1 server and 9.1 client.

I mounted two NFS shares:

mango2:/home/crcarle # mount /mango
mango2:/home/crcarle # mount /mango/home.hda6

Now I have this:

# cat mtab

mango.somewhere.gov:/ /mango nfs rw,addr=xxx.yyy.zz.13 0 0
mango.somewhere.gov:/home.hda6 /mango/home.hda6 nfs
rw,addr=xxx.yyy.zz.13 0 0

(obviously I obscured some details)

I want to unmount these, but I get this:

mango2:/home/crcarle # umount /mango/home.hda6
umount: /mango/home.hda6: device is busy
umount: /mango/home.hda6: device is busy

So I did:

mango2:/home/crcarle # umount -fl /mango/home.hda6

And now this process is hung and I can't even kill it with kill -9.

I turned off the server after that, hoping the client would just give up
eventually.

Why is NFS so prone to getting stuck? How can I get my umount process
to give up? How can I unmount the now non-existent shares?

How does one make NFS deal more "elegantly" with interrupted servers?

There seems to be a problem with Suse 9.1 in which I can never unmount
stuff. It always says it is busy, even if it isn't.

Thanks for input.


--
__________________________________________________ _____________________
Christopher R. Carlen
Principal Laser/Optical Technologist
Sandia National Laboratories CA USA
(E-Mail Removed) -- NOTE: Remove "BOGUS" from email address to reply.

Chris Carlen wrote:
> Hi:
>
> I have a Suse 8.1 server and 9.1 client.
>
> I mounted two NFS shares:
>
> mango2:/home/crcarle # mount /mango
> mango2:/home/crcarle # mount /mango/home.hda6
>
> Now I have this:
>
> # cat mtab
>
> mango.somewhere.gov:/ /mango nfs rw,addr=xxx.yyy.zz.13 0 0
> mango.somewhere.gov:/home.hda6 /mango/home.hda6 nfs
> rw,addr=xxx.yyy.zz.13 0 0
>
> (obviously I obscured some details)
>
> I want to unmount these, but I get this:
>
> mango2:/home/crcarle # umount /mango/home.hda6
> umount: /mango/home.hda6: device is busy
> umount: /mango/home.hda6: device is busy
>
> So I did:
>
> mango2:/home/crcarle # umount -fl /mango/home.hda6
>
> And now this process is hung and I can't even kill it with kill -9.
>
> I turned off the server after that, hoping the client would just give up
> eventually.

Well, turning on the server again allowed my to umount /mango/home.hda6
but umounting the /mango still reported busy. So I umount -l that one,
and it's straightened out.

There's still a problem with reporting busy though.


Good day!




--
__________________________________________________ _____________________
Christopher R. Carlen
Principal Laser/Optical Technologist
Sandia National Laboratories CA USA
(E-Mail Removed) -- NOTE: Remove "BOGUS" from email address to reply.

On Thu, 05 Aug 2004 10:09:56 -0700, Chris Carlen wrote:
> I have a Suse 8.1 server and 9.1 client.
>
> I mounted two NFS shares:
>
> mango2:/home/crcarle # mount /mango
> mango2:/home/crcarle # mount /mango/home.hda6
>
> Now I have this:
>
> # cat mtab
>
> mango.somewhere.gov:/ /mango nfs rw,addr=xxx.yyy.zz.13 0 0
> mango.somewhere.gov:/home.hda6 /mango/home.hda6 nfs
> rw,addr=xxx.yyy.zz.13 0 0
>
> (obviously I obscured some details)
>
> I want to unmount these, but I get this:
>
> mango2:/home/crcarle # umount /mango/home.hda6
> umount: /mango/home.hda6: device is busy
> umount: /mango/home.hda6: device is busy
>
> So I did:
>
> mango2:/home/crcarle # umount -fl /mango/home.hda6
>
> And now this process is hung and I can't even kill it with kill -9.
>
> I turned off the server after that, hoping the client would just give up
> eventually.
>
> Why is NFS so prone to getting stuck? How can I get my umount process
> to give up? How can I unmount the now non-existent shares?
>
> How does one make NFS deal more "elegantly" with interrupted servers?
>
> There seems to be a problem with Suse 9.1 in which I can never unmount
> stuff. It always says it is busy, even if it isn't.

Hmm, I've done dumb stuff like "sitting in a directory" that I've tried to
unmount, but I assume you (or one of your processes) isn't doing that?

In many cases when I've thought NFS was hung/wedged, I have found the
program "lsof" to be useful. I think (but I'm not positive) that it will
report on directories (i.e. NFS mount points) and then you can trace
further from there. I've had several "doh!" moments as a result.

I have been using NFS a LOT, for decades. Usually, my NFS servers are Sun
Solaris machines with Solaris and/or Linux clients. I do have some Linux
servers occasionally. I have heard rumours that "Linux NFS is busted" and
I might have seen some anomalies (server side, never to my knowledge on
the client side), but I'm not sure. The "busted" comments were ISTR
related to network booting: Linux serves stuff in reverse order?

Generally, I've found NFS to be rock solid. Problems have been usually
"fat fingers" (my own). The "stateless" NFS server design is nice, because
(as someone said here) if the server crashes (rare) then NFS will resume
on reboot. I have sometimes done maintenance reboots on NFS servers not
worrying about the NFS clients, because I know they will soon recover.

--
Juhan Leemet
Logicognosis, Inc.

Chris Carlen wrote:
>
> How does one make NFS deal more "elegantly" with interrupted servers?

'man mount' and read the options for the NFS filesystem type. The
combination of "hard" and "intr" is probably what you're after.