NFS serives running via udp / tcp part 2

[Aslam Ghumra_2]

Hi All

In my earlier request to find a solution to
Why my server was running NFS under UDP and not TCP, I was given the following responses.

1. add NFS_TCP=1 in the nfsconf file
2. add -o proto=tcp in MOUNTD_OPTIONS within the nfsconf file.

I have now had chance to reboot my server, however starting nfs.server started but failed(!).

In rc.log the following message is posted.

Start NFS server subsystem
Output from "/sbin/rc3.d/S100nfs.server start":
----------------------------
starting NFS SERVER networking

starting up the rpcbind daemon
rpcbind already started, using pid: 917
Reading in /etc/exports
usage: /usr/sbin/rpc.mountd [-l log_file] [-t #] [-X #] [-p]
-t 1 Only log errors (default)
-t 2 Log errors and mount requests
-t 3 Internal tracing
-X 0 [HP-only] Disable pathname exclusion
FAILURE CODE: 10
starting up the NFS daemons
/usr/sbin/nfsd 30
starting up the Status Monitor daemon
rpc.statd already started, using pid: 1035
starting up the Lock Manager daemon
rpc.lockd already started, using pid: 1041
"/sbin/rc3.d/S100nfs.server start" FAILED

The server is still running NFS under UDP.

So my question is how do I get my server to mount NFS filesystems under TCP and not UDP.

Any help will be much appreciated.

regards

Aslam

[Muthukumar_5 · Accepted solution]

Did you add those settings in /etc/rc.config.d/nfsconf file? check that.

In the system terminal,

# /etc/init.d/nfs.server start
# /etc/init.d/nfs.core start

Is server started or having errors. check /var/adm/syslog/syslog.log file.

--
Muthu

[Steven E. Protter]

Shalom Aslam,

Errors are in /var/amd/syslog/syslog.log

Seeing them would be helpful.

You hae changed the mount options for this server. You now need to mount nfs from a remote system and see if errors are generated in the local log and on the nfs client.

You may need to disable the options in the configuration file that refer to UDP. It would be helpful to see the whole config file.

SEP

[Aslam Ghumra_2]

Hi,

thanks for responding.

nfs.server is running.

Unfortunatle I dont know where or which file defines the fact that this server uses udp rather than tcp.

Another machine which mounts off this server can see the filesystems and mount it via tcp.

No errors shown in syslog or on the other machine's syslog.

Aslam

[Muthukumar_5]

Remove this -o proto=tcp from MOUNTD_OPTIONS in nfsconf file. It will work. You have to use this when you are doing mount. Restart the nfs.server and nfs.core as said in my previous reply.

--
Muthu

[Aslam Ghumra_2]

Hi Muthu

I have amneded the file and restarted but it still runs NFS under UDP.....

Aslam

[Muthukumar_5]

what are you getting for this:

# netstat -na | grep 2049

Is it tcp or udp. It has to be tcp.

--
Muthu

[Aslam Ghumra_2]

Hi

I am getting the following output

# netstat -na | grep 2049
tcp 0 0 192.168.1.1.2049 192.168.1.5.872 ESTABLISHED
tcp 0 0 *.2049 *.* LISTEN
tcp 0 0 192.168.1.1.2049 192.168.1.3.957 ESTABLISHED
tcp 0 0 192.168.1.1.2049 192.168.1.2.708 ESTABLISHED
tcp 0 0 192.168.1.1.1023 192.168.1.20.2049 ESTABLISHED
udp 0 0 *.2049 *.*
udp 0 0 *.2049 *.*

Aslam

[Muthukumar_5]

NFS server is running in tcp and udp as well. tcp 0 0 *.2049 *.* LISTEN says it is started. It is working.

Hope you have to remove -o proto=tcp in MOUNTD_OPTIONS for not making to get errors like,

usage: /usr/sbin/rpc.mountd [-l log_file] [-t #] [-X #] [-p]
-t 1 Only log errors (default)
....

"/sbin/rc3.d/S100nfs.server start" FAILED

in system reboot.

--
Muthu

[Aslam Ghumra_2]

Thanks Muthu

I dont get the errors.

When running nfsstat -m, I still get my filsystem running under UDP.

Is there any other way apart from manually mounting them using the "-o proto=tcp" option under mount ?

Aslam

[Muthukumar_5]

You can put proto=tcp in /etc/fstab file in respect to the file system.

NFS.Server is working fine. You have to change this setting only.

--
Muthu

[Aslam Ghumra_2]

Hi Muthu

Forgive my ignorance but where in the fstab file would I need to put this in?
The example below is part of the fstab

nas9000_private:/met /home/met nfs rw,suid 0 0

Cheers,

Aslam

[Muthukumar_5]

Change this,

nas9000_private:/met /home/met nfs rw,suid 0 0

to

nas9000_private:/met /home/met nfs rw,suid,proto=tcp 0 0

--
Muthu

[Muthukumar_5]

Forgot,

# mount -a

after /etc/fstab setting,

will read /etc/fstab to set the new mount with tcp protocol.

--
Muthu

[Aslam Ghumra_2]

Hi

I have changed it, and to test I have umounted and mounted that specific filesystem.

In fstab

nas9000_private:/den /home/den nfs rw,suid,proto=tcp 0 0

I then mount /home/den and get the following :

nfs mount: getaddr_nfs: nas9000_private: NFS service not responding(retry delay=5s)
nfs mount: retry: retrying(1) for: /home/den after 5 seconds
nfs mount: retry: giving up on: /home/den

Aslam

[Muthukumar_5]

Did you restart nfs.server and nfs.core in NFS server side. Again, mount -a in nfs.client side. Are you seeing the same error?

--
Muthu

[Aslam Ghumra_2]

Hi Muthu

No I didn;t restart the nfs.core or nfs.server.

All I did was to umount /home/den
then did the same for mount /home/den.

Would it make a difference ?

Aslam

[Aslam Ghumra_2]

All

When I go into SAm and look at the nfs mounted file systems, it does report that it is using UDP.

HOwever if you modify it, it wont change it , if you remove and read the mount, changing the protocol to NFS results in ... yes you guessed it protocol udp.

I'm completely baffled

[Zinky]

Peace Aslam,

Let's start from the beginning. What exactly is your environment? Is your issue having to do with your HP-UX environment always having to mount NFS via UDP from your NFS server?

1. What is your NFS server? Is it a NAS or another UNIX Server?

2. What is your /etc/fstab entry on your server that pertains to the mount from your NFS server?

3. If your NFS server is a NAS - is it possible it's NFS service is restricted to UDP transport? This may very well explain why TCP does not work...


Hope this helps..

[Aslam Ghumra_2]

Hi

This is hp-ux 11.11 v1 server where it serves out filesystem, to other hp-ux servers. However it also mounts filesystems from a windows NAS9000 servers. (Services for unix v3.5).

This server is similar to the other hp-ux servers but will only mount NFS filesystem using UDP. the other nodes all mount NFS under TCP ( strange eh! ).

The entry from the other server is :
nas9000_private:/met /home/met nfs rw,suid 0 0
which will mount via TCP.

The NFS NAS server, has an option to have TCP support which is selected, my other servers mount using tcp fine.

I hope that I have answered your queries.

Aslam

[Dave Olker]

Hi Aslam,

I think I understand your situation. You have a number of identical NFS clients all mounting a filesystem from the NAS9000 server. All but one of these clients mounts the filesystem using TCP but one of them mounts via UDP. Is this correct?

The one system that mounts via UDP - is it able to mount other filesystems via TCP from other servers? Is it able to mount other filesystems exported by the NAS9000 using TCP, but fails on the specific filesystem you've mentioned?

I'm trying to understand if this is a problem with your client not being able to mount ANY filesystem via TCP or if this is the server not allowing TCP access to this client, or if it is the server not allowing TCP access to this one specific filesystem.

So again:

1) Can the NFS client mount using TCP from any other server?

2) Can the NFS client mount any other filesystems from this NAS9000 using TCP?


Regards,

Dave

[Aslam Ghumra_2]

Dave, et all

I think I understand your situation. You have a number of identical NFS clients all mounting a filesystem from the NAS9000 server. All but one of these clients mounts the filesystem using TCP but one of them mounts via UDP. Is this correct?

>>> Yes this is correct.

The one system that mounts via UDP - is it able to mount other filesystems via TCP from other servers?

>>> Yes, we have a NAS8000 server (linux) which is mounted via TCP.

Is it able to mount other filesystems exported by the NAS9000 using TCP, but fails on the specific filesystem you've mentioned?

>>> No it does not seem able to mount using TCP from the NAS9000, as I have tried it on a number of filesystems.

1) Can the NFS client mount using TCP from any other server?

>>> Yes it can from another NAS8000(linux).

2) Can the NFS client mount any other filesystems from this NAS9000 using TCP?

>>>As yet unable to, however there are other systems similar, same nfsconf file, which can mount using TCP.

My immediate pointer would be to the NAS9000 windows NFS server, however the only config file I know is the nfsconf file, which is now the same on all the 6 hp-ux servers. Yet this is the only server which mounts all but one filesystem under TCP.

Cheers, Aslam

[Dave Olker]

Hi Aslam,

The next piece of data I'd collect would be two nettl traces - one trace collected on the failing HP-UX client when it tries to mount the TCP filesystem from the Windows-based NAS9000, the other trace collected on a different HP-UX client when it successfully mounts the TCP filesystem from the NAS9000 server.

Comparing the failing trace with the working one might give some clues as to why the TCP mount is failing. If you need help collecting the traces or interpretting them let me know.

Regards,

Dave

[Aslam Ghumra_2]

Hi

Yup i'll need some advice on how to collect these traces.

Aslam

[Dave Olker]

The nettl syntax I usually use to start and stop a trace on the HP-UX system is:

START# /etc/nettl -tn pduin pduout loopback -e ns_ls_ip -f

Reproduce the problem

STOP# /etc/nettl -tf -e all

This will create at least one, possibly more than one file in the directory with whatever name you assigned to the trace file (via the -f option) and an extension of .TRC000/1. These are the raw trace files.

If you need help interpretting them, please upload these raw trace files, from both the working and failing clients, here along with the IP addresses of the clients so that we can filter out any non-relevant packets in the traces.

Regards,

Dave