On Wed, Oct 02, 2002 at 01:50:32PM -0400, Javier Gostling wrote:

> Also, is this bug reported in bugzilla?

Forget I asked. It has been reported tons of times. I found plenty of
info on the subject on the different bugs, in case anyone needs help on
this:

- Use lilo instead of grub. Lilo stores the major and minor in the boot
sector and has no problem with this issue.

- As previously stated, changing the grub.conf kernel line to include
the major/minor solves the issue too.

- There is a patch (applied to more recent kernels) which includes the
definitions for the RAID devices mentioned (/dev/rd/*, /dev/ida/*, etc.)
You can find it in http://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=65207

For those who want a detailed explanation of this issue, keep on
reading.

The kernel has a table which can be used to lookup device names and
transform them into major/minor device numbers. This table is hard
coded, and must reflect the entries in /dev to be useful. Grub send the
kernel an option telling it where the root fs is (/dev/ida/c0d0p3 in my
case). Since the initrd does not have this (and many other) entry in
/dev, it must use this table to mount the right device with the real
root fs. Now, if the string passed by grub does not match any of the
device strings in the table, the kernel is unable to mount the new root
fs. Lilo does not suffer from this, because the boot sector itself has
the major/minor combo stored in it, so the kernel is passed the numbers,
not the device name.

So, since I'm not into recompiling kernels if I can avoid it, I'm back
to lilo. At least until we have had time to test 7.3 enough to migrate
production servers.

Cheers,
-- 
Javier Gostling
Ingeniero de Sistemas
Virtualia S.A.
[EMAIL PROTECTED]
Fono: +56 (2) 202-6264 x 130
Fax: +56 (2) 342-8763

Av. Kennedy 5757, of 1502
Las Condes
Santiago
Chile

Attachment: msg89909/pgp00000.pgp
Description: PGP signature

Reply via email to