NFS lockd/statd lock up network connection

[Doug Rabson]

On 2 May 2009, at 17:21, Tom McLaughlin wrote:

> Doug Rabson wrote, On 04/08/2009 03:20 PM:
>> On 5 Apr 2009, at 07:38, Tom McLaughlin wrote:
>>> Hey, I have a recent -CURRENT box which has a mount exported from an
>>> OpenBSD NFS server.  Recently I enabled lockd and statd on the  
>>> machine
>>> but this has started to cause the network connection on the  
>>> machine  to lockup.  I find the following in dmesg:
>>>
>>> nfs server exports:/mnt/raid0/net/home: lockd not responding
>>> nfs server exports:/mnt/raid0/net/home: lockd not responding
>>> nfs server exports:/mnt/raid0/net/home: lockd not responding
>>> nfs server exports:/mnt/raid0/net/home: lockd not responding
>>> nfs server exports:/mnt/raid0/net/home: lockd not responding
>>> nfs server exports:/mnt/raid0/net/home: lockd not responding
>>> nfs server exports:/mnt/raid0/net/home: lockd not responding
>>> nfs server exports:/mnt/raid0/net/home: lockd not responding
>>> NLM: failed to contact remote rpcbind, stat = 5, port = 28416
>>>
>>> Additionally I see this when trying to restart netif:
>>>
>>> em0: Could not setup receive structures
>>>
>>> I've tried building with NFS_LEGACYRPC but that has not changed   
>>> anything.  Additionally I've tested this on 7-STABLE and while  
>>> lockd  still does not work (so, looks like I'll still have to work  
>>> around  my need for NFS locking) the network connection at least  
>>> does not  lock up.  Is what I'm seeing evidence of some further  
>>> problem?
>> It looks as if lockd is not running on the server. The NFS locking   
>> protocol needs it enabled at both ends. Also, NFS_LEGACYRPC won't   
>> affect this - the record locking code always uses the new RPC code.
>
> Hi Doug, lockd is runing on both ends.  The problem appears to be  
> with the system running out of mbuf clusters when using lockd. [1]   
> For now I'm mounting the particular mount with nolockd as an option  
> to get around this.  I've gotten errors with my -STABLE box using  
> this mount with lockd enabled but at least the system didn't run out  
> of mbuf clusters and lose all network connectivity.

Could you please try the attached (unfortunately untested - I'm at  
BSDCan) patch and see if it affects your problem. This patch attempts  
to fix PR 130628 which appears similar to your issue.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: rpc-deadlock.diff
Type: application/octet-stream
Size: 10744 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-current/attachments/20090507/a54c192d/rpc-deadlock.obj
-------------- next part --------------