RAID-1 in a hurry with grub and mdadm

From CLUG Wiki

This is for people who are in a hurry, not novices

The instructions are for people who don't build monolithic kernels, or use standard pre-compiled kernels. If you do build monolithic kernels, you can skip all the initrd stuff - just be sure md is compiled into your kernel.

Let's assume your existing installation is on hda, and you've just added hdd. We also assume that hdd is the same size as (or bigger than) hda. You have two partitions: / and /boot on hda currently.

Install the mdadm tools

apt-get install mdadm  <---- Enable autostart when asked

Make sure that you don't have mdadm version 1.8.1* - they are development releases and don't produce random UUIDs! If you have a raid array created with one of these mdadms, it will give error messages on boot, and may accidentaly merge it with another array!!!!

I don't use /etc/mdadm/mdadm.conf so just delete it or it'll mess you around

Set up the partition tables

sfdisk -d /dev/hda | sfdisk /dev/hdd -O hdd-partition-sectors.save
fdisk /dev/hdd -> set all partition types (except extended partition) to FD

Create the (degraded) RAID devices

We'll use md0 for /boot, md1 for the root filesystem and md2 for swap.

mdadm -C /dev/md2 --level=raid1 --raid-devices=2 missing /dev/hdd6
mdadm -C /dev/md1 --level=raid1 --raid-devices=2 missing /dev/hdd5
mdadm -C /dev/md0 --level=raid1 --raid-devices=2 missing /dev/hdd1

You may have to use the following instead if you get the following error during the next boot:

md: Autodetecting RAID arrays.
md: invalid raid superblock magic on hda1

mdadm -C /dev/md2 -e 0.90 --level=raid1 --raid-devices=2 missing /dev/hdd6
mdadm -C /dev/md1 -e 0.90 --level=raid1 --raid-devices=2 missing /dev/hdd5
mdadm -C /dev/md0 -e 0.90 --level=raid1 --raid-devices=2 missing /dev/hdd1

This is because mdadm version 2.5 uses a version 1.0 superblock which does not support in-kernel auto-detection. To verify this, just cat /proc/mdstat | grep super, and you will see 'super 1.0'

NOTE: depending on you installation your partitions might be called differently, check the partition table!

Have a look at the results

cat /proc/mdstat

Create a swap partition on one of the RAID devices

mkswap /dev/md2
swapon /dev/md2
swapon -s

NOTE: the latter two steps are merely to test if you were succesful in creating swap space

Create EXT3 partitions on the other two RAID devices

mkfs -t ext3 /dev/md0
mkfs -t ext3 /dev/md1

Copy all your files into the two new RAID devices

First the / partition

mount /dev/md1 /mnt
cd /
find . -xdev -print0 | cpio -0pdvum --sparse /mnt
umount /mnt

Then the /boot partition

mount /dev/md0 /mnt
cd /boot
find . -xdev -print0 | cpio -0pdvum --sparse /mnt
cd /
umount /mnt

Now mount both new partitions and set up fstab

mount /dev/md1 /mnt
mount /dev/md0 /mnt/boot
vim /mnt/etc/fstab          <--- hda1 => md0, hda5 => md1, hda6 => md2

Change the grub config:

vim /mnt/boot/grub/menu.lst

First copy and paste all lines of your current bootoption. Then change the first occurrence of the two identical bootoptions. You need to replace hda1 with md0 and hda5 with md1. Further change the name of the initrd-image so it ends in -raid. Don't forget the change the title so it reflects this is the raid option.

NOTE: the unaltered copy of your current boot options is merely for safety sake.

Install the grub loader:

grub>
 device (hd0) /dev/hdd
 root (hd0,0)
 setup (hd0)
 quit

Now create a new initrd for the degraded RAID so you can reboot using the RAID modules.

cd /mnt
mount -o bind /proc ./proc
# check if you see anything below /mnt/dev, if not:
mount -o bind /dev ./dev
chroot /mnt
# mkinitrd has more flavours, use one of the following two syntaxis (check manpage)
# syntax 1
mkinitrd -r /dev/md1 -o <the initrd-image-name from the grub config>-raid
# syntax 2 (a.o. Fedora FC3)
mkinitrd <the initrd-image-name from the grub config>-raid <kernel-version>

Do yourself a favour and check if the image contains the necessary modules. Smart way to do this:

gunzip -c <the initrd-image-name from the grub config>-raid | cpio -it

and check for modules md (or dm-) and raid1

Reboot (into hdd not hda) and test if you can boot directly into the RAID array. You have two ways to go about:

(1) Copy the lines from the raid bootoption in the grub config on the raided device and paste those into the grub config on the non-raided device. Be sure to change the root line so it says

root            (hd1,0)

(2) Boot as you would normally do, interrupt the boot process and go into grub (c)ommandline mode. Now type

root (hd1,0)
configfile /grub/menu.lst

You'll see the "new" bootmenu. Now choose the new raid boot option. Choose (e)dit, highlight the line root (hd0,0), again press (e)dit and change it into root (hd1,0). Now (b)oot!

Make the hda partition types "RAID auto-start" when the previous step works

fdisk /dev/hda -> set all partition types (except extended partition) to FD

Put hda into the RAID array (this is where hda gets clobbered)

mdadm /dev/md2 -a /dev/hda6
mdadm /dev/md0 -a /dev/hda1
mdadm /dev/md1 -a /dev/hda5

Make some coffee while hda gets overwritten

watch cat /proc/mdstat

Now you must create a non-degraded RAID initrd

mkinitrd -r /dev/md1 -o /boot/initrd.img-2.4.24-1-586tsc-raid

Change the grub config

vim /boot/grub/menu.lst

Remove all lines of the old bootoption (with hda)

Re-install the grub loader

grub>
 device (hd0) /dev/hda
 root (hd0,0)
 setup (hd0)
 quit

Reboot check that the RAID array does not come up degraded

Some commands that might come in handy (RTFM for details):

mdadm --zero-superblock /dev/hde5
mdadm -D /dev/md0
mdadm --stop /dev/md0
mdadm -S /dev/md0
less -S /proc/partitions
dpkg-reconfigure mdadm --> to de/re-activate autostart
mdadm --detail --scan >> /etc/mdadm/mdadm.conf