From: Jeff Layton <jlayton@redhat.com>
Subject: Re: Stale NFS handles on LVM2
Date: Wed, 30 May 2007 08:28:20 -0400
Message-ID: <20070530082820.2cd26b89.jlayton@redhat.com>
References: <465D229E.5080809@bio.ifi.lmu.de>
	<20070530082312.GA11045@janus>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Cc: nfs@lists.sourceforge.net, linux-lvm@redhat.com
To: Frank van Maarseveen <frankvm@frankvm.com>
In-Reply-To: <20070530082312.GA11045@janus>
Sender: nfs-bounces@lists.sourceforge.net
Errors-To: nfs-bounces@lists.sourceforge.net

On Wed, 30 May 2007 10:23:12 +0200
Frank van Maarseveen <frankvm@frankvm.com> wrote:

> On Wed, May 30, 2007 at 09:07:10AM +0200, Frank Steiner wrote:
> > Hi,
> > 
> > I'm not sure if this is a NFS or LVM issue, so I cross-post this.
> > We get stale NFS handles on a LVM2 volume in the following scenario:
> > 
> > - The NFS server is running SLES10/kernel 2.6.16, but it happens
> >   with 2.6.21, too
> > 
> > - the server has /dev/sda, a built-in SCSI disk, with
> >   /dev/sda7 a LVM2 volume, say "exportlvm"
> > 
> > - due to module order in the initrd sda is always sda
> > 
> > - exportlvm is exported to the NFS clients
> > 
> > - the server has 4 fibre-channel raids connected, those have
> >   several partitions and two of the raids have LVM2 filesystems.
> > 
> > - the server doesn't mount or export any of the FC filesystems
> > 
> > Now, when we shutdown the fc port of the server or remove the fc cable, so 
> > that all the fc filesystems disappear, and then reboot the server,
> > the NFS client gets stale NFS handles on the "exportlvm". When we reboot
> > the client and then reconnect the FC cable or port and reboot the server
> > again, the stale NFS handles show up again.
> > 
> > When we change /dev/sda7 from a LVM2 to a normal fs, there are no stale 
> > NFS handles when rebooting the server in this way. 
> > 
> > Looks like rebooting the server after dis-/connecting the FC raids with
> > their LVMs somehow changes sth. on the LVM2 on /dev/sda7, like the minor
> > numbers etc. Is that possible? Is it a bug?
> 
> This is the behavior I would expect and is not a bug. Major/minor numbers
> depend on the initialization order and number of block level devices
> finally available for mounting the filesystems (disks/partitions/logical
> volumes, etc). When pulling a disk or (in your case) volume in the middle
> some numbers may shift and that affects the filehandles.
> 
> Use the fsid= option in /etc/exports to avoid this problem (clients need
> to remount because this changes filehandles again).
> 

There are also some options with lvchange that can give LVM volumes a
persistent minor number. But using the fsid option (like frankvm said)
is probably your safest bet.

-- 
Jeff Layton <jlayton@redhat.com>

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs