Coda File System

Re: volutil restore

From: Dr A V Le Blanc <LeBlanc_at_mcc.ac.uk>
Date: Tue, 4 Apr 2000 10:52:28 +0100
On Mon, Apr 03, 2000 at 04:06:14PM -0400, Jan Harkes wrote:
> The restored volumes are actually restored backup volumes. i.e.
> non-replicated and possibly read-only. That is why the recovery
> instructions are to mount them in a temporary place and copy the stuff
> out into newly created replicated volumes.

Ah, but copying volumes is awfully difficult, since (so far as I know)
there are no programs that copy coda mount points and ACLs correctly.
This means that restoring a large system accurately could take weeks
of very tedious work.  (Find all mount points, record and remove,
tar and untar in new locations, then for each directory use 'cfs la'
to list ACLs, and manually use 'cfs sa' to set corresponding ACLs
on the other directory (since coda's cfs doesn't support 'cfs ca'),
then restore all the mount points in the recovered volume and repeat
the same process).

> In most cases when a volume is replicated and only one server dies, it
> is trivial to recover. Create an empty rw-replica with the correct
> replica-id and force the date from the other replica back by doing an
> 'ls -lR' from some client over the volume.

But the problem comes when restoring an unreplicated volume, or one
which has lost all its replicas.  Also, since coda does not support
something like the AFS 'vos move', as far as I can see, moving a
volume from one file server to another becomes a major operation.

> It might be possible to flip
> a few flags in the restored volume header to turn these non-replicated
> volumes into real rw-replica's, in which case recreating empty replicas
> on the other servers and editing and regenerating the VRlist would be
> sufficient to facilitate full recovery using server-server resolution.
> However we haven't had time to look at this.

But the net result of all of this appears to be that coda has no
facilities at present for restoring volumes from backups, though it
can produce a readonly copy of what was backed up, from which an
administrator can manually reconstruct the data.  Is this correct?
And there is no easy way to move a volume, rename a volume, convert
a volume with all its mount points and ACLs from non-replicated to
replicated, or conversely?

     -- Owen
     LeBlanc_at_mcc.ac.uk
Received on 2000-04-04 05:54:24