Bug#353679: linux-image-2.6.15-1-686 can't boot from partitioned RAID1 (partitions not discovered)
Package: linux-image-2.6.15-1-686
Version: 2.6.15-4
Severity: normal
Problem:
For a file server, we want to have everything on sw RAID1.
Having /home on a raid1 made from /dev/sda and /dev/sdb works fine.
�Note that I used the whole devices not partitions. So no
"raid autodetect" partition type)
The machine in question also have /dev/hda and /dev/hdb on
which I want to install linux on raid1, with several partitions.
The cumbersome way is to use partitions and create one md device for /,
one for /usr, one for /var, one for swap, and one for /usr/src.
I know this will work though.
The ideal way is to create a partitionable md device from /dev/hda+/dev/hdb,
and then partition this md device into several components.
The problem with this, is that the initrd does not sufficiently recognize
such a setup, so I can't have root on the partitioned md device.
While testing, I have been using a plain root on /dev/hda1, and a
raid1 in degraded mode on /dev/hdb. I have noticed that the
initrd seems to detect the raid1 on hdb, but does not run mdadm
in the way required to detect the partitions on the md device.
Therefore, I can't have root on /dev/md1p1 as I hope.
The problem will probably not be solved in time for this server,
which will use a more cumbersome setup lots of little md devices instead.
But I am interested in working with debian developers to improve this
for the future. I have a home machine where I can experiment with
partitionable md device and even try booting from them. (That machine
will have to use a amd64 kernel though.)
I could use some advice on changing an initrd - the simple and
hurried approach of unpacking the existing one with gzip+cpio,
adding an mdadm command and repacking with cpio+gzip didn't work.
Anyway, a general solution for this problem probably shouldn't
hardcode an mdadm command either. Ideally, the partitionable
md device(s) should be detected and have the partitions recognized
because they exist. I am interested in looking at this,
unless someone else already is busy doing it. Is there
some documentation on the proper way to change the initrd,
so that the changes might get merged back for the benefit
of others?
Helge Hafting
-- System Information:
Removed, for we don't run email on the server in question. Reportbug
ran on a different machine. The server has kernel 2.6.15-1-686
from debian, and runs debian testing. The risk of running 'testing'
on this server is ok.
Reply to: