NFS regression on recent STABLE

[Josh Paetzel]

I've been involved with a project at work doing some fun things with NFS.  
Recently due to changes in a available hardware we did a complete refresh of 
the system.  New HBAs, new storage boxes, and due to some internal software 
confusion we updated the OS on the head from 8.1-R with zpool 15 backported to 
STABLE.

Our primary client that we were using against this setup was a ESXi 4.1 
machine.  In a nutshell, it didn't work.  Long description is, ZFS would 
deadlock and any operation on the pool would hang.  The ESXi instance would 
mark the NFS mount as unavailable.  I initially thought this could be due to 
any number of factors, we have new HBAs in the  mix, new storage boxes, a new 
version of zpool, and one test case.

Meanwhile, back at the ranch, I have a somewhat similar setup at home.  
FreeBSD 8.1 NFS server, ESXi 4.1 box mounting an NFS exported ZFS filesystem 
from the FreeBSD box as a datastore.

Last night I pulled that box up to STABLE, rebooted it, and a minute after it 
rebooted the ESXi box marked the NFS datastore as unavailable.  I checked the 
FreeBSD machine and sure enough it hung doing an ls on the zpool.

I ran a few tests, and as soon as the ESXi box mounts the NFS export it hangs 
the ZFS filesystem.  If I don't mount it up, the NFS server does fine.  
Thinking it might be a ZFS problem I moved the mount to a UFS filesystem.  
While this doesn't cause the box to hang on filesystem operations, the mount 
goes unavailable.

The only other client I have on my network is a FreeBSD 8.1 box, and that has 
no issues
 
All of this is with the standard NFS server, I haven't yet tested with the 
experimental server.

-- 
Thanks,

Josh Paetzel
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 488 bytes
Desc: This is a digitally signed message part.
Url : http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20101123/06cf7ba0/attachment.pgp