On Sat, Dec 21, 2002 at 08:28:09AM +1100, Daniel Carosone wrote:
> On Fri, Dec 20, 2002 at 10:09:13AM +0000, David Laight wrote:
> > I would have a file that contained what you found on the last
> > boot, and what you called it.
> 
> This is all well and good, until you don't have that file.  Many
> an inexperienced Solaris sysadmin has zorched the wrong disk when
> they've booted from CD to attempt a recovery and found the disks
> and controllers numbered differently than they expected.

But if you always use 'probe' order then a change could result
in the wrong filesystem being detected as /tmp and having all
its contents removed at boot time.
(This must have happened to someone...)
 
> By all means lets minimise the effects (such as by reserving a full
> 16 or whatever slots for a label type), but not go overboard.

The mbr extended partition is a particular problem because it is
a linked list [1] and not an array, 

Certainly making use of volume name / serial numbers to track
disks is likely to be advantageous.  Naming them - and having
/dev entries with the corresponding names so that /etc/fstab
can refer to the disks and/or partitions by name might also
be advantageous.
This would probably stop the CD boot getting it completely
wrong.

Unfortunately you can't rely on the 'unique' serial number
in the ATA/SCSI identify response being unique.  I checked
a few CF cards, those from the same batch were indistinguishable.

On i386 there is space in sector zero of the partition for
a filesystem name.

	David

[1] in spite of the appearance of being a tree structure.

-- 
David Laight: david@l8s.co.uk