On Mon, Apr 09, 2001 at 07:32:43PM -0500, John R. Jackson wrote:
> >Could you try this patch on Tru64.
>
> This is probably a silly question, but did you try the patch on Linux?
> I was playing with this too and my version (functionally the same as
> yours, I think) put Linux back to trying to use /dev/root.
The patch works for me, what's your /proc/mounts and /etc/mtab entry?
> Here's what I think is happening. Amanda is trying to convert the
> disklist entry (e.g. "/") to both a device name and a file system type.
> It wants the device name to pass to dump, even though that's silly
> because every version of dump I've been around already knows how to do
> that (and probably better than Amanda). It wants the file system type
> to know what dump program to run (e.g. advfs -> vdump instead of dump).
>
> The 2.4.2p2 changes effectively caused search_fstab to fail if it could
> not find a match. That, in turn, caused amname_to_devname to return
> the input arg (e.g. "/") as the device name, which dump was perfectly
> happy with.
>
> One thing I cannot remember is why we had a problem in the first place
> (the "LABEL=/xxx" /etc/fstab entries). I'm guessing what happened was
> search_fstab returned a match based on the mount point, but the "device"
> was returned as "LABEL=/xxx", which dump would definitely not have liked.
The original problem I fixed was not related to the LABEL, it was
a selfcheck problem:
Amanda Backup Client Hosts Check
--------------------------------
ERROR: dev: [can not access /dev/root (/): No such file or directory]
>
> Amname_to_fstype is not so forgiving of search_fstab failing and returns
> an empty string. So when we changed search_fstab to fail if the device
> did not exist, we lost the ability to detect device type.
>
> Note that in the advfs branch of sendbackup-dump, it passes the mount
> point, not the device name, to dump. That's why this used to work.
>
> So how about:
>
> * We go back to the original search_fstab code. On Linux and advfs
> systems, it will find a match on the mount point but return a
> bogus device name.
>
> * We change amname_to_devname to stat the device name and if it does
> not exist, return whatever amname_to_dirname() returns. This should
> solve the original Linux problem and put advfs back the way it was.
selfcheck will fail, we can't check permission on amname_to_dirname(), we
should check permission of the device.
> We might also re-consider the use of amname_to_devname in general and
> maybe change to using amname_to_dirname like advfs is already doing.
Jean-Louis
--
Jean-Louis Martineau email: [EMAIL PROTECTED]
Departement IRO, Universite de Montreal
C.P. 6128, Succ. CENTRE-VILLE Tel: (514) 343-6111 ext. 3529
Montreal, Canada, H3C 3J7 Fax: (514) 343-5834
