NFS umount failed :"device busy"

[Rambo_1]

I can't umount the NFS even if after I used the
#fuser -kcu /NFS_mount_point.

someone told me the only way to fix it is to reboot the server . is it true ?
I'm looking for another way to fix the NFS umount problme .can anyone tell me ?

Thanks & B.R
rambo

[Rafael Casero]

Seems that someone is cd into the mount point, have you tried forcing the umount ?

umount -f

[Warren_9]

Hi,

try the "lsof" and it may be able to locate the open file/process on the mount point.

http://hpux.cs.utah.edu/hppd/hpux/Sysadmin/lsof-4.77/

GOOD LUCK!

Warren

[Andrew Young_2]

Hi Rambo

You will probably find that using lsof that it is opened by the kernel, this is the NFS daemon doing this. You may need to stop all the NFS daemons being NFS server, NFS client and NFS core in that order. If that doesn't work there is the option of restarting Rpcd however I would not do that unless you really are feeling brave as many other daemons use this too. Rebooting may be easier if it is an option.

HTH

Andrew Y

[Paul Senior_1]

1)is this the nfs server or client (or both)?

2)is there any 3rd party s/w which uses nfs which needs to be stopped (e.g. file CIFS sharing s/w on another machine - e.g. a windows server

3) is it a hard nfs mount (is the nfs server up?)

4)very obvious, but worth stating: are you in the nfs mount directory or do you have a current telnet session open in it? whatabout on the console?

[Richard Tesar]

If you have mounted NFS volume from any NFS server and this one for any reason cannot be reachable
you may have some problem during unmounting that shared followed this message "Device is busy".
In this case have two choice. Wait for NFS server or disconnect by force. In principle is
concerned it oust of nfs lock. The trick is to send a SIGKILL to the 'lockd' kernel thread.
This tells lockd to release all of its locks.

good luck
rich

[Dennis Handly]

>rich: The trick is to send a SIGKILL to the 'lockd' kernel thread. This tells lockd to release all of its locks.

I can't comment whether this is true, no mentioned of special signals lockd(1m) handles. But one thing is certain, if you send SIGKILL, that processs can't do anything. And if you kill lockd, you must restart it manually.

[Timothy Nibbe]

You could do a kill -HUP on the lockd process

[Dennis Handly]

>Timothy: You could do a kill -HUP on the lockd process

I doubt it. It isn't documented and the sources don't show this.

[MSufiyan]

or try to mount on temp dir

for eg: mount /NFS_mount_point /tmp/test_ms

If you get success , then

rmdir the original mount point

create it one more time

mkdir -m 755 mount_point ( the one you have removed )

umount /tmp/test_ms

mount /NFS_mount_point mount_point

[Todd McDaniel_1]

Is this mount point defined in fstab or manual mount?

I have seen cases where you can blow away the mnttab and immediately attempt a umount. (and/or comment out the entry in fstab)

I'd suggest executing the umount command, which will fail. then rm mnttab, then arrow up and run umount again.

Regards,

Todd McDaniel