[Unionfs] problems with NFS export, v. 1.4

[Klaus Knopper]

Hi,

On Mon, Nov 27, 2006 at 07:15:28PM -0500, Josef Sipek wrote:
> On Wed, Nov 22, 2006 at 11:51:20AM +0100, Michael Creel wrote:
> > Hello to everyone,
> > 
> > I'm adapting the ParallelKnoppix live CD to export home to the compute 
> > nodes via unionfs. This makes things very convenient, since a uniform 
> > set of files is available to all the nodes in the cluster. One the 
> > master node, the home dir is /UNIONFS/home/pk. This is nfs exported to 
> > the nodes using the following entry in /etc/exports:
> > 
> > /UNIONFS/home/pk 
> > 192.168.0.0/255.255.255.0(rw,root_squash,fsid=42,async,no_subtree_check)
> > 
> > 
> > This works fine with a 2.6.17 kernel and unionfs 1.3. However, with a 
> > 2.6.18 kernel and unionfs 1.4, the compute nodes can't mount the 
> > export, and the report the typical "reason given by server: permission 
> > denied" error. So I'm looking for pointers as to how to get this going 
> > with unionfs 1.4 (or I can switch to a 2.6.19 kernel and unionfs CVS, 
> > if that would help).
>  
> If you don't mind runing latest kernels, using CVS snapshots is better. In
> this case, there haven't been any changes since the 1.4 release (mostly due
> to the fact that it was a rather late release.)
> 
> Besides that, it is really odd that it doesn't work with 2.6.18. Looking at
> the Unionfs changelog I don't see anything that would make it not work like
> that. We'll try to reproduce it here in the lab...

Could you look into dmesg at the server (!) side? If you see a kernel
oops caused by nfsd when accessing /var/lib/nfs/v4recovery, this may be
the reason why your clients are not able to access the (defective)
server anymore.

Solution, until this is fixed in unionfs: mount -t tmpfs none
/var/lib/nfs/v4recovery before starting nfsd.

Regards
-Klaus Knopper