Re: kernel BUG at /data/lemmy/linux.trees.git/fs/nfs/idmap.c:681!

[Bryan Schumaker <bjschuma@netapp.com>]

On 08/07/2012 11:14 AM, Myklebust, Trond wrote:
> On Tue, 2012-08-07 at 10:36 -0400, Bryan Schumaker wrote:
>> On 08/07/2012 10:27 AM, Joerg Roedel wrote:
>>> On Tue, Aug 07, 2012 at 10:17:33AM -0400, Bryan Schumaker wrote:
>>>> On 08/07/2012 10:15 AM, Joerg Roedel wrote:
>>>>> Yes, it reproduces pretty reliable here with Ubuntu 11.10 Server on an
>>>>> Intel box with an NFSv3 directory mounted at boot. This is the only box
>>>>> I have seen this so far, probably it depends on the config. I attach the
>>>>> config of the failing box.
>>>>
>>>> Interesting.  Are you mounting v4, too?  This code shouldn't be
>>>> running for v3... maybe that's why I haven't been able to hit it.
>>>
>>> No, I am not using NFSv4 on the box where the BUG happens. I have
>>> another box mounting the same directory where the BUG does not trigger
>>> with v3.6-rc1. A difference I spotted between the kernels is, that on
>>> the failing box NFS is compiled as a module whereas it is compiled into
>>> the kernel on the box that works fine. Not sure if that has anything to
>>> do with the problem...
>>>
>>
>> Your stack trace is showing v4 calls on the failing box, those definitely shouldn't be happening if you're using v3.  Can you double check /etc/fstab and /proc/mounts on a working kernel to be sure?
>>
>> My VM has nfs as a module, so I don't think that's the issue... I just started compiling your config to test on my own.
> 
> Joerg,
> 
> The stack trace definitely shows that the NFS client is attempting an
> NFSv4 mount. Are you supplying a 'vers=3' mount option? If not, then
> note that recent versions of nfs-utils can be configured to try NFSv4 as
> the default mount option, so I'd guess this is why you are hitting an
> NFSv4 path.
> 
> Bryan,
> 
> That said, when looking at the legacy upcall, it seems that if
> rpc_queue_upcall fails, then we don't do anything to clear
> idmap->idmap_key_cons. Ditto if the call times out, or if the pipe is
> closed before the downcall.
> 

Ah!  I didn't think about the upcall failing, thanks Trond!  I'll work on a patch.

- Bryan