nfs mount problem

[Dave Olker]

Sorry Rita! I typically look at ITRC threads after I get the daily summary email at the end of the day. :)

Feel free to ping me if you need me to look at something urgently: dave.olker@hp.com.

Cheers,

Dave

[Rita C Workman]

I thought I should re-open and update this thread.

Turns out the mount options used above do work and the mount command does appear to work for NFS. But when I went out to my 11.31 NFS client and ran 'll' it hangs, never to shows anything as mounted.

The interesting thing is...the NFS mount is actually there! I ran (from the 11.31 client side) to install from the NFS host to the 11.31 NFS client and lo and behold, it installed just fine on the client. No delay, no hangs.
But you simply can not see the NFS filesystem at all on the client side.

Now Mr. Olker (our NFS Guru...) was able to recreate the problem and came up with the following suggestion:

The quick workaround for you would be to mount the filesystem using the â readdirâ mount option. This disables READDIRPLUS for this mount point and falls back to the previous behavior of sending READDIR and LOOKUP calls instead.

He says he will see about sending this up to the lab to address.

So, Mr. Olker, just stop by and add any additional information for folks, should they run into this....and the 10 points are yours!!

Many Thanks Dave,
Regards,
Rita

[Rita C Workman]

THREAD CLOSED...

[Dave Olker]

Here's what I've found so far...

The reason why the "ls" command stops responding is because the 11i v3 client gets into a loop calling the same READDIRPLUS call over and over again. It retrieves the first portion of the remote directory contents but never advances its pointer to request the next chunk of the directory. Instead a network trace shows the same request for the first part of the directory contents over and over again.

What's interesting is this same problem happens with Solaris 10 clients. I used the very latest Solaris 10 bits to build a client and I get the identical behavior from that box.

When I use an 11i v3 client running an older version of ONCplus (11.31.04) I don't see this problem, so this makes me think we inherited this problem when we ported the ONC 2.5 client code as part of ONCplus 11.31.06. Sure enough, when I install ONCplus 11.31.06 on my test system the ls command stops responding.

I've submitted a CR for this problem so we'll look into this further. In the mean time, the best solution for this problem is to mount the NFS filesystem with the "readdir" option, which effectively disables READDIRPLUS for that filesystem.

Dave

[Dave Olker]

Update: this problem has been root-caused to being a defect in the CDFS code. The problem is fixed in the 11.31 patch PHKL_40531:

( QX:QXCR1000940411 )
The following symptoms are seen on a NFS client:
1. ls(1) or ll(1) command stops responding on a NFS mount point
exported from a CDFS filesystem.

Unfortunately this doesn't help Rita because her server is running 11.11. I've got a call into the CDFS lab to find out if they have plans to roll this fix back to 11.11 and 11.23. I'll update here when I hear something.

Regards,

Dave