Re: nfs umount error : Bad MNT RPC: RPC: Timed out

[Jan Kryl <loo%bazmek.net@localhost>]

On 28/06/13 11:49 +0200, Nicolas Joly wrote:
> On Fri, Jun 28, 2013 at 01:57:23AM +0000, Christos Zoulas wrote:
> > In article <20130627161325.GA7478%lynche.sis.pasteur.fr@localhost>,
> > >
> > >Looking into umount(8) the code, show that lastest communication with
> > >the server is unconditionnaly made over UDP (from sbin/umount/umount.c) :
> 
> Is there a specific reason we do server notification in umount(8)
> instead of sys/nfs/nfs_vfsops.c:nfs_unmount() where we, if i
> understand things correctly, have all needed stuff to make it right
> without too much hassle.
> 
mount protocol has been always handled in userland (except
diskless boot). Moving just unmount RPC call to kernel wouldn't
be systematic. There should be an interface to obtain mount protocol
transport (and other NFS options) from kernel and then umount could
use this information to properly unmount filesystem. Linux seems to
get this information from /etc/mtab. FreeBSD and Solaris suffer from
the same problem as NetBSD.

cheers
-jan


> > >Hacking it to use "tcp" instead instead of "udp" makes the problem
> > >vanish.
> > >
> > >While i can ask network admins to open a few more UDP ports to make it
> > >work ... I wonder if this should be adjusted to use the protocol which
> > >was requested at mount ?
> > 
> > Or at least a flag to specify the protocol....
> 
> This won't work when unmounting all filesystems if we have nfs mount
> with both protocols.
>
> > This is a bug, we should fix it.
> 
> Thanks for confirming.
> 
> -- 
> Nicolas Joly
> 
> Biology IT Center
> Institut Pasteur, Paris.