[Lxc-users] Running an NFS server within LXC

[Rob Landley]

On 04/15/2013 05:24:20 AM, Papp Tamas wrote:
> On 04/15/2013 12:17 PM, David Parks wrote:
> > Could this be an LXC specific issue? I’m trying to run an NFS  
> server within a container.
> >
> > davidparks21 at hadoop-fsimage-bkup1:~$ sudo service nfs-kernel-server  
> restart
> >
> > * Stopping NFS kernel daemon
> >
> >     ...done.
> >
> > * Unexporting directories for NFS kernel daemon...
> >
> >     ...done.
> >
> > *FATAL: Could not load /lib/modules/3.5.0-27-generic/modules.dep:  
> No such file or directory*
> >
> > ** Not starting NFS kernel daemon: no support in current kernel.*
> >
> > This is an ubuntu 12.10 container (built from Ubuntu cloud server  
> images) in a vanilla 12.10 host
> > installed from CD.
> >
> > I’m having a lot of troubles tracking down a solution on google and  
> am wondering if this is
> > LXC-related. I’ve tried enabling this option in LXC:
> >
> > lxc.aa_profile = unconfined
> >
> > Or is this somehow related to my host installation? I am trying to  
> avoid **any** unnecessary
> > configuration in the host, so I haven’t tried installing NFS server  
> to the host OS.
> 
> 
> You cannot make it.
> Try glusterfs as a workaround.

I fixed CIFS to work, and at one point had 9p working (although I don't  
remember if the patch went upstream, it was pretty trivial though).

But NFS is a crawling horror that does superblock merging and has so  
many different layers of passing data to itself, and reimplements half  
of the VFS, and has asynchronous RPC layers of sun crap, and has  
parallel implementations of several verions of the same code (v2! v3!  
v4! v4.1 is a different protocol than 4!)

It's... it's evil.

9p (The "diod" server plus mount -t v9fs) is a very nice network  
filesystem which, unlike NFS, is _NOT_CRAZY_.

Rob