Software-Raid1 Root in Woody
Hi all.
This document briefly describes the steps needed to install a Debian
Woody GNU/Linux system with root on a software raid1 device.
It took me a fair amount of trials and errors before I got it right, so
I would like to share my current knowledge.
I'm not subscribed to this list, so if you want to send me
corrections/improvements/whatever, please CC my email address.
NO WARRANTY:
This is the procedure that worked for me.
Despite my efforts to be accurate, there is no warranty
that it will work for you, or even that it will not damage your system.
Be warned that unless you know what you are doing and you are extremely
careful, you seriously risk to loose all the contents of your hard
disks. So I strongly advice you not to play with PCs that contain valuable
information.
Most important: whatever damage you may do by applying the following
procedure, it's YOUR FAULT, and I take no responsibility.
If you do not agree, stop reading now.
Well, now revert to the real stuff:
Since this is a test setup, the environment is going to be small, and
supported by the debian vanilla kernel:
- Pentium III
- SCSI disk 0:0, sda, 160 Mb
- SCSI disk 0:1, sdb, 160 Mb
I used SCSI disks because I had them easily available, but the procedure
should work for IDE disks on (e.g.) hda and hdc as well.
1. Install Debian Woody.
I started from floppies (vanilla kernel) + network, and partitioned
the disks as follows:
# fdisk -l /dev/sda
Disk /dev/sda: 64 heads, 32 sectors, 160 cylinders
Units = cylinders of 2048 * 512 bytes
Device Boot Start End Blocks Id System
/dev/sda1 * 1 160 163824 83 Linux
# fdisk -l /dev/sdb
Disk /dev/sdb: 64 heads, 32 sectors, 160 cylinders
Units = cylinders of 2048 * 512 bytes
Device Boot Start End Blocks Id System
/dev/sdb1 * 1 160 163824 fd Linux raid autodetect
Then I made an ext2 filesystem and installed a minimal system on /dev/sda1.
Better avoid installing further packages until the raid is set up, while
it might be a good idea to update the system with
# apt-get update
# apt-get upgrade
2. Switch to kernel 2.4.
I prefer to use raidtools2 with a stock 2.4 series kernel.
Moreover, despite later versions of LILO being easier to setup, I
decided to stick on GRUB. So the next steps were:
# dpkg -P lilo
# apt-get install grub kernel-image-2.4.18-686
# grub-install '(hd0)'
Notice that for processors different from Pentium III you should pick a
flavour different from '-686'.
Create /boot/grub/menu.lst like this:
# cat /boot/grub/menu.lst
title Debian
root (hd0,0)
kernel /vmlinuz root=/dev/sda1 ro
initrd /initrd.img
WARNING: Don't install devfsd. Device paths change and initrd-tools
get confused.
Reboot.
3. Create and populate the Raid 1 structure in degraded mode.
Install raidtools2:
# modprobe -k raid1 # needed to prevent preinst from complaining
# apt-get install raidtools2
Create /etc/raidtab like this:
# cat /etc/raidtab
raiddev /dev/md0
nr-raid-disks 2
raid-level 1
persistent-superblock 1
chunk-size 4
device /dev/sdb1
raid-disk 0
device /dev/sda1
raid-disk 1
failed-disk 1
Notice the failed-disk directive at the end of the file. It specifies
that the current root disk, /dev/sda1, is to be ignored for the moment,
and marked `failed' in the raid superblock.
Now make up the raid:
# mkraid /dev/md0
If this is your n-th trial with n>1, you might be forced to
use the '-f' option to mkraid. Read carefully the warning before
proceeding.
You can check the status by looking at /proc/mdstatus
Build an ext2 filesystem on /dev/md0 and copy the whole system on it
(that's why it's useful to keep it minimal)
# mke2fs -O sparse_super,filetype /dev/md0
# mount /dev/md0 /mnt
# find / -xdev -depth | cpio -pmdu /mnt
Modify /mnt/etc/fstab and /mnt/boot/grub/menu.lst to mention md0 or sdb in
place of sda1:
# diff /etc/fstab /mnt/etc/fstab
4c4
< /dev/sda1 / ext2 errors=remount-ro 0 1
---
> /dev/md0 / ext2 errors=remount-ro 0 1
# diff /boot/grub/menu.lst /mnt/boot/grub/menu.lst
2,3c2,3
< root (hd0,0)
< kernel /vmlinuz root=/dev/sda1 ro
---
> root (hd1,0)
> kernel /vmlinuz root=/dev/md0 ro
Now build an initrd for the new setup:
# mkinitrd -o /mnt/boot/initrd.img-2.4.18-"your flavour" -k -r /dev/md0
The option '-k' instructs mkinitrd to keep the expanded image under
/tmp/mkinitrd.XXX/initrd instead of deleting it.
Notice that /mnt/initrd.img -> /boot/initrd.img-YYYY, so don't try to be
too smart.
4. Reboot with root on /dev/md0 and synchronize the disks.
Reboot. So far, the system is still configured to boot from the first
disk and leave the second alone.
Now we have to manually interrupt grub's boot sequence in order to
launch the second disk instead of the first:
Ask grub for a command line (press 'c') and at the 'grub>' prompt
issue the command:
grub> configfile (hd1,0)/boot/grub/menu.lst
You should get the same menu item as before, but meaning a different sequence
(use 'e' to examine it, then ESC to return to the menu). Press RETURN
to boot.
You should end up with linux running and /dev/md0 mounted as /.
Now it's time to change the partition type of the old root disk from 83 to
FD:
# fdisk -l /dev/sda
Disk /dev/sda: 64 heads, 32 sectors, 160 cylinders
Units = cylinders of 2048 * 512 bytes
Device Boot Start End Blocks Id System
/dev/sda1 * 1 160 163824 fd Linux raid autodetect
Now modify /etc/raidtab to remove the last line ('failed-disk...'), and
attach the old root partition as a plex of the raid1 structure:
# raidhotadd /dev/md0 /dev/sda1
The md driver starts synchronization, that can be checked by looking at
/proc/mdstat.
Rebuild the initial ramdisk with a clean raidtab (paranoid mode on):
# mkinitrd -k -o /initrd.img
Setup grub on both disks:
# (echo 'root (hd1,0)'; echo 'setup (hd1)'; echo quit) | grub
# (echo 'root (hd0,0)'; echo 'setup (hd0)'; echo quit) | grub
Done.
Now you can reboot (just to test the setup works), play with dselect
to your pleasure, etc.
Again: if you happen to install devfsd you might end up with an unbootable
system, and no way to recover.
As an exercise to the reader, try disabling one of the disks and test
your ability to recover... If you find a way out of the Kernel panic,
I'll be glad to know. ;-)
best regards
-- gb
Reply to: