Re: scsi disk detection and /dev names, RE: My $.02 on the raging de vfs debate

Guest section DW (dwguest@win.tue.nl)
Tue, 12 Oct 1999 00:16:11 +0200 (MET DST)


From paul@clubi.ie Mon Oct 11 22:53:34 1999

On Mon, 11 Oct 1999, Guest section DW wrote:

> So devfs is a quick workaround of this limitation rather than being a real
> cure (for this particular prob). If anyone with the coding skills is
> listening, it'd be great if static /dev could support location based disk
> names too.

Oh, but those people with coding skills did their work already years ago.
Not many people use this work, but it exists. Eric Youngdale wrote
scsidev. From man scsidev:

i know about scsidev, but how robust is it? what happens in between
the time that the kernel boots and the time the script runs? (eg
RAID0145 or LVM kernel autostartup). I'd rather use the old way, and
stay on my toes than use a script which might fail (or operator
failure).

All you can be sure of with scsidev is that 8,1 = con0/bus0/id3 at
the point the script was run. With real location-based /dev nodes
there's never any ambiguity. Which i want to see supported. If it's
going to take massive hackery to the scsi disk driver then maybe
devfs? It solves it pretty easily.. (no no.. don't go there!!) :)

Are you trying to get me into this idiotic discussion?

Look: If you like devfs - fine, so use it.
If you dislike devfs - that is OK as well. There are several
other ways to get stable device names. I have pointed out three:
(i) Use UUID - works today (for ext2), but requires some initrd hackery
if you want to use it also to find the root device.
(ii) Use scsidev - works today (for scsi), but requires some initrd hackery
if you want to use it also to find the root device.
(iii) Assign device IDs to devices that encode SCSI ID and all the rest.
Easy, if one has large device numbers. Easy to do. I did it a few times.
Maybe some official kernel will have them one of these centuries.

People who want especially to address the problem of having the kernel boot
from the right SCSI disk, say, can probably easily write some small patch
where the desired device name is given on the kernel command line so that
after the SCSI detection phase this name is parsed and the correct root
device found. Also after introduction of large device numbers such a boot
parameter remains meaningful: large numbers are for computers, but humans
want to type some easy symbolic representation. Something for you?

Andries

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/