Linux: using md to get RAID1

I wanted to give few extra years to our old griffin.linux.hr server (which was replaced with new SMP Celeron box), so I decided to turn it into WiFi AP. However, it came with two disks, one of 4.3Gb (QUANTUM FIREBALL CR4.3A) and one with 6.4Gb (SAMSUNG SV0644A). Since those disks used to be ftp site, I don't trust them much, and hence I decided to put software RAID 1 on them.

For this project I used:

1. pentium server (SMP with P5/200)
Debian Woody (3.0r0 and latest unstable)
LILO 22.2
linux kernel 2.4.18

• make a floppy using http://people.debian.org/~blade/install/#lvm

  Hint: You will have to have correct boot cd for that floppies. In time of writing, floppies are for boot CD 2.4.18-bf2.4 on same location. Don't try with wrong boot disk, it won't work!

• boot from boot CD and press ALT+F2 to get console after initial question about language of installation
• use command extdisk to add raid and lvm modules from floppy

  Hint: Have floppy in drive before you press enter on extdisk!

• make partitions. Set the type of partitions to "FD" and copy partition table from first disk to second (both partition tables should be the same)

  Hint: Switch units to sectors in fdisk to make partitions exact size. You can partition rest of bigger disk normally (as 83 - Linux).

  Hint: You really want to have /boot partition on both disks which are in mirror below 1024^th cylinder (I will repeat this again later).

• write /etc/raidtab using nano-tiny (yuk!). Second disk must be set as failed-disk not raid-disk

  Hint: You can write template for this file (/tmp/rt) and then multiply it by number of partitions that you have with cat /tmp/rt /tmp/rt > /etc/raidtab.
  Also, it makes sense to call md* devices the same as partitions.

  raiddev /dev/md1
  	raid-level	1
  	nr-raid-disks	2
  	nr-spare-disks	0
  	chunk-size	64
  	persistent-superblock	1
  	
  	device	/dev/hda1
  	raid-disk	0
  	device	/dev/hdc2
  	failed-disk	1
  

  Hint: It makes good sense to put disks on separate IDE channels. Just take a look at speed comparison at the end.

• make raid (mkraid)
  (use mkraid --force /dev/md* if mkraid "protests")

  Hint: If you get error whey you try to run mkraid like this:
  # cd /tmp/lvmext/bin
  # ./mkraid /dev/md*

• mkfs on /dev/md*

  Hint: mke2fs -j /dev/md1 to make it ext3, of course :-)

• mount the recently created raid disks (/target, /target/var ...)
  mount /dev/hda1 /target - if hda1 is /
  mount /dev/hda2 /target/home - if hda2 is /home
• continue the installation (alt+F1)
• at the end of installation DO NOT forget to copy /etc/raidtab to /target/etc
• make sure that (for now) you have 'normal' (/dev/hda1) devices in /etc/fstab; not /dev/md1 devices. Install LILO as usual in /dev/hda1. For now.
• restart the computer. modprobe raid1 if needed after new installation boot.
• apt-get install kernel-image

  Hint: Alternative is to recompile a new kernel with raid support built in. However, on slower machines it would take much more time.

  Hint: You might want to add do_initrd = Yes to /etc/kernel-img.conf to avoid warnings...

• modify /etc/mkinitrd/modules to include md and raid1 modules.
• apt-get install raidtools2
• change root=/dev/hda1 in /etc/lilo.conf to root=/dev/md1 and add initrd=/initrd.img
• change /etc/fstab so that is uses /dev/md* instead of /dev/hd* devices.
• mkinitrd -o initrd.img-2.4.18-586tsc /lib/modules/2.4.18-586tsc/

  Hint: If you installed kernel which is not 2.4.18-586tsc, change that in mkinitrd line.

• lilo -v

  Hint: You should see someting like:

  Boot image: /vmlinuz -> boot/vmlinuz-2.4.18-586tsc
  Mapping RAM disk /initrd.img -> /boot/initrd.img-2.4.18-586tsc
  Added Linux *
  

  If there isn't 'Mapping RAM disk' part, you might want to change initrd line in lilo.conf

• restart the computer. It should boot new kernel (not the one from boot CD).
• Change lilo.conf so that the boot device is set to /dev/md1. Add raid-extra-boot="/dev/hda,/dev/hdc" and run lilo -v so you will see that it wrote to correct disks.

  Hint: /boot partition (with your kernel) and lilo boot= option must point to same partition (/dev/md1) for this to work!
  /boot partition must be below 1024 cylinder on both disks in mirror for raid-extra-boot to work correctly!

• with raidhotadd /dev/md1 /dev/hdc1 add the second drive to raid. After that change /etc/raidtab from failed-disk to raid-disk

  Hint: You will probably want to do apt-get install hdparm and hdparm -d 1 /dev/hda; hdparm -d 1 /dev/hdc before adding disk to mirror. Speed comparison will show you why.

• cat /proc/mdstat will tell you that it finished rebuilding.
• reboot :-)

  Hint: I had various problems with lilo during usage. At one moment, I swapped disks on first and second IDE channel, so Quantum end up on second IDE channel while Samsung moved to first one. I haven't been able to write correct boot block on /dev/md1 device because I didn't have /boot partition below 1024 cylinder on both disks (because they had different geometry). You have been warned -- take extra care about 1024 cylinder limit, even with modern lilo.