problem with dead NFS mounts on AIX 4.3
BogusStrawman
dies, I can't remove the mount on the client machine - it says the
mounted fs is in use. If I try to use that mount the shell (or
whatever) hangs indefinitely. There appears to be no "normal" way to
kill this mount other than to bring the dead server back to life.
I read in this group of cross mounting as a hack to overcome this
problem but details are thin and I don't really understand what cross
mounting is. I tried the following:
server exports /tmp
client nfs mounts server:/tmp as /nfstmp
client exports /nfstmp
server tries to mount client:/nfstmp but gets the error:
vmount: A file, file system or message queue is no longer
available.
One post suggested the cross mount was "internal", so I tried
server exports /tmp
server acts as client and mounts server:/tmp as /nfsmount
client mounts server:/tmp as nfsmount as well
This has no effect on the primary problem (as you would expect).
Can anyone shed light on this? It would be much appreciated.
mark taylor
bos.net.nfs.client and bos.net.nfs.server.
could be IY35344 U483289 NFS connection hang during NFS server
shutdown
HTH
Mark Taylor
Stephane Gassies
sources, only the last mount will be "active".
If you have a dead "/nfstmp" (because od NFS server dead), you can
ADD a mount to /nfstmp from anywhere. E.g. if you export /tmp,
with correct NFS configuration, you can do a
mount localhost:/tmp /nfstmp
Now /nfstmp will not be considered as "dead", as a active "mount"
is on it.
If fact, this corresponds to the normal way of mounting a filesystem
under a UNIX system : The system hides the previous content of the
[mount point] directory, and gives instead the content of another
directory : the one given at mount time.
BUT this will *NOT* allow you to umount the previous acces to /nfstmp,
you are right, this is only a workaroud to avoid your system to seem
"hanged" when the NFS server goes down.
Regards,
Stephane
BogusStrawman a écrit:
BogusStrawman
filesystem dies, the client cannot unmount the dead nfs mount - the
error returned says the filesystem is in use. When you try to access
that mount, the process hangs indefinitely. The only way that I have
found to remove the mount is to bring the dead server back to life or
reboot the client. If the client never uses the mount in the first
place, the unmount request eventually times out and the mount
disappears.
I read in this group that cross mounting is a hack that overcomes this
problem, but I didn't understand the explanation. I tried the
following:
server exports /tmp
client mounts server:/tmp over /mnt/tmp
client exports /mnt/tmp
server mounts client:/mnt/tmp but gets error:
vmount: A file, file system or message queue is no longer available.
another go as a post mentioned the cross mounts were 'internal':
server exports /tmp
server mounts server:/tmp over /mnt/tmp
client mounts server:/tmp over /mnt/tmp
but as one would expect this had no affect.
Can anyone shed some light on this problem? Many thanks in advance.
BogusStrawman
problem are different - the server is not busy at all. Thanks anyway.
m...@talk21.com (mark taylor) wrote in message news:<dee0a7c3.04051...@posting.google.com>...
Bogus Strawman
displayed the first one.
I am now using my own reader.
Sorry for any confusion.
mark taylor
this
>>problem are different - the server is not busy at all. Thanks
anyway.
ok, that was one picked from many many related fixes as you have not
posted any software levels.
can you post your nfso -a and lslpp -l output for the client and
server filesets on both the server and the client ( am assuming they
are both AIX ).
Also, what flags are you using to mount the filesystem ? soft/hard,
retries, timeouts etc...
Mark
JohnM
system and either cd'd out or killed them that an umount -f could be
used to force the umount without waiting for a server ack. Did you try that?
J.
BogusStrawman wrote:
--
Remove the 'NS' from my address to reply.
BogusStrawman
about the cross mount business from the newsgroup, but at least I now
have something to keep the production machines alive ....
thanks again.
Stephane Gassies <sgas...@club-internet.fr> wrote in message news:<40A131D8...@club-internet.fr>...
> On most UNIX systems, if you mount a directory name from many
> sources, only the last mount will be "active".
>
> If you have a dead "/nfstmp" (because od NFS server dead), you can
> ADD a mount to /nfstmp from anywhere. E.g. if you export /tmp,
> with correct NFS configuration, you can do a
>
> mount localhost:/tmp /nfstmp
>
> Now /nfstmp will not be considered as "dead", as a active "mount"
> is on it.
>
> If fact, this corresponds to the normal way of mounting a filesystem
> under a UNIX system : The system hides the previous content of the
> [mount point] directory, and gives instead the content of another
> directory : the one given at mount time.
>
> BUT this will *NOT* allow you to umount the previous acces to /nfstmp,
> you are right, this is only a workaroud to avoid your system to seem
> "hanged" when the NFS server goes down.
>
> Regards,
> Stephane
>
> BogusStrawman a crit:
BogusStrawman
Yes I have tried that (eg: lsof comes up clean) as well as killing the
nfsd and biosd processes.
Thanks anyway.
JohnM <jmadd_NOFR...@austin.rr.com> wrote in message news:<bNypc.115513$Dn1....@fe2.texas.rr.com>...
Jose Pina Coelho
news:e73daf02.04051...@posting.google.com:
> If a client is using an nfs mount and the server of that exported nfs
> filesystem dies, the client cannot unmount the dead nfs mount - the
> error returned says the filesystem is in use. When you try to access
> that mount, the process hangs indefinitely.
On the client:
# ifconfig lo0 alias server-ip-address netmask 255.255.255.255
# umount /nfs-filesystem
# ifconfig lo0 -alias server-ip-address
Works on all versions.
--
Doing AIX support was the most monty-pythonesque
activity available at the time.
Eagerly awaiting my thin chocolat mint.