Mount points hang

[Dieter S. Vener]

Hi,

We have an issue on our HP RP8400. We are running HP-UX 11.11 on this system and are exporting a mount point called /report_area to about 20 different flavors of UNIX. For some reason, all of our Solaris boxes have a problem with this mount point. It works fine for about a day and then the mount point gets hung on all of the Solaris boxes.

I.E on Solaris box

%cd /net/chpr8400/report_area


The only thing that solves the problem is rebooting the RP 8400. Rebooting the Solaris boxes does not work. While rebooting works, I do not feel that rebooting a system is a good solution.

Does anyone have any ideas as to what the problem might be?

Thanks,

Dieter

[Chan 007 · Accepted solution]

As you say it has problems during the night,

check your backup timing, sometimes the backup may take the network and your NFS is having this problem.

007

[Alzhy]

Possibilities:

- what's your Patch Rev for NFS? (show_patches|grep -i nfs"

- your network is busy

- how are you exporting your shares? (cat /etc/exports)

- how do your Solairs machines mount the NFS shares (mount -p|grep -i nfs)

FInally a tip:

On Solaris - you can actually force a umount of an NFS mount. (umount -f /nfsmount)

HTH.

[Henk Geurts]

using a soft mount might fix your problem :
an example:
mount -o soft -o proto=udp hpserver:/whatever /mnt

see also
man mount.

regards.

[Dieter S. Vener]

Hi Guys,

/report_area is mounted on Solaris using autofs.

mount -p | grep -i nfs
chpr8400:/report_area - /net/chpr8400/report_area nfs - no nosuid,xattr


/report_area exported in /etc/exports:

/report_area -access=csun4:csun5:csun6,root=csun4:csun5:csun6

umount -f /net/chpr8400/report_area


I am not sure what show_patches is, I used swlist.

%find / -name show_patches -print

%swlist -l product | grep -i nfs
NFS B.11.11 ONC/NFS; Network-File System, Information Services, Utilities

Thanks,

Dieter

[Alzhy]

Then you may may want to try looking at switching to pure soft NFS mounts and UDP protocol as suggested.

BTW, show_patches is /usr/contrib/bin/show_patches on my 11i systems.

Do check you have the following at least:

# /usr/contrib/bin/show_patches|grep -i nfs
PHKL_25238 11.00 NFS nfsd deadlock
PHKL_25993 thread nostop for NFS, rlimit, Ufalloc fix
PHKL_28185 Tunable;vxportal;vx_maxlink;DMAPI NFS hang
PHKL_29335 vx_nospace on NFS write.
PHKL_30151 NFS binary overwrite hang
PHKL_30920 (u)mount,final close,NFS umount,Busy syncer
PHNE_30380 ONC/NFS General Release/Performance Patch


If thee wisheth to continue using AUTOFS, make sure your Solaris end and HP end have the latest patches. For Solaris - apply the latest recommended patch cluster. For HP-UX 11i .. the latest GoldApps bundle and check via Custom Patch Assenssment at ITRC and specifically check for NFS and AUTOFS patches.

[Dave Olker]

Hi Dieter,

Are all of the other UNIX systems using AutoFS and TCP for their mounts of this filesystem and none of them are experiencing problems?

A couple of things to try:

1) When the mount point becomes hung on the Solaris clients via AutoFS, can you manually mount this same filesystem to a different directory on the Solaris client and get a normal response? In other words:

csun4# mkdir /nfs
csun4# mount chpr8400:/report_area /nfs

Does this work?

2) What does an "nfsstat -m" output on the Solaris systems show?

3) You should contact Sun and find out how to enable debug AutoFS logging to see if this is an AutoFS-specific problem. I know there is a "secret" command that enables this logging but I can't recall what it is, or if it is specific to different flavors of Solaris. A quick call to Sun should help get this information.

4) What does a network trace taken on the HP-UX server show during one of these stops responding? Does the HP-UX system ever see the NFS requests coming from the Solaris systems? Does it reply to the correct IP address?

5) Your swlist output indicates that you don't have any 11.11 ONC/NFS patches installed on the server. You should consider at least installing the latest ONC/NFS bundle patch and any dependencies on the server to eliminate any known/fixed problems that could cause this, like NFS/TCP connection problems.

The latest patch is PHNE_33498 and it's depencencies include:

PHCO_33205 mountall, Dev IDs enabler, iSCSI support
PHKL_30398 KI FSS ID and KI_rfscall
PHKL_33367 CacheFS umount device busy

I recommend installing these 4 patches on your NFS server to eliminate any known problems.


Hope this helps,

Dave

[Dieter S. Vener]

Thanks all. You've given me some stuff to work with.

Thanks,

Dieter

[Steven E. Protter]

Shalom Dieter,

I'd get the latest quarterly patch on HP-UX and a good selection of patches you find searching for NFS and hang.

Also, look into the patch levels of the Solaris Box.

SEP

[DRAXLER, Markus]

Hi Dieter,

it looks like I am running into the same problem.

I am exporting several file systems from a HP-UX 11.11 box and the mount on a linux and another HP box hang. A reboots helps - sometimes for hours sometimes for days.

I found out that "exportfs -au" hangs when I try to stop the NFS server on the box exporting the file systems. Also it is not possibele to kill any of the nfsd processes and there are two nfsd processes belonging to the inetd as father. One of them has "-3" as owner. This is looking very strange. I have the last recommended patch level from HP on this machine.

Did you find out what caused your problem?

Thanks,
Markus.

[DRAXLER, Markus]

Hi Dieter,

it looks like I am running into the same problem.

I am exporting several file systems from a HP-UX 11.11 box and mounts on a linux and another HP box hang. A reboots helps - sometimes for hours sometimes for days.

I found out that "exportfs -au" hangs when I tried to stop the NFS server exporting the file systems. Also it is not possibele to kill any of the nfsd processes and there are two nfsd processes belonging to the inetd as father process. One of them has "-3" as owner. This is looking very strange. I have the last recommended patch level from HP on this machine.

Did you find out what caused your problem?

Thanks,
Markus.