Introduction to Linux Software RAID1

This is a HTML-ised version of a presentation I gave to Gloucestershire Linux User Group on 17th April 2007. It is also available as a PDF.

RAID Basics

RAID = Redundant Array of Inexpensive Disks

Many drives to create a single volume

Different levels, eg:

RAID0 = "Striping" - Large volume data is chunked ("striped") across drives (faster read/write but no redundancy)
RAID1 = "Mirroring" - Small volume copied identically to several drives (slower read/write but can survive drive loss)
RAID5 = "Block-Level Striping" - Large volume made up of many drives, data is arranged so that any one drive can fail and data can be rebuilt from blocks on surviving drives

There are many other variations, I'm only going to talk about RAID1

Support

Used to require dedicated hardware card

Now also on some motherboards

Hardware RAID is set up via extension to BIOS - you get an extra "Press F3 to set up RAID" screen or similar upon boot

Linux has supported RAID in OS ("software RAID") since 2.4

No need for hardware card or fancy motherboard in Linux!

You can use any IDE, SCSI or SATA drives

With RAID1 "Mirroring", you will see a volume equal only to the smallest drive size in the array (eg. 60gb & 40gb = 40gb)

Easy choices

I'm only going to talk about RAID1 "Mirroring"

Create a single volume, identical copy on each of two disks

Either disk can fail and we will keep our data

Ideal for:

Surviving drive hardware failure
Confidence when using second-hand hardware
Keeping a moment-by-moment "backup" of data
Typical Samba fileserver, home mail server, low/medium use webserver

Not ideal for:

Gaming and other intensive disk IO applications
Both reads and writes are marginally slower
NOT A SUBSTITUTE FOR OFFLINE BACKUPS
If you delete the data, it is just as gone as normal!

Avoid Legacy Tools

Software RAID HOWTO is woefully out of date

raidtools are deprecated as of kernel 2.4

fd RAID volume type (eg. in fdisk) is also deprecated

mdadm is what you want to use, certainly on distros <5 years old

mdadm rarely uses config file - don't bother with config file, it will almost always autodetect and ignore it anyway

Assumptions

You have a single root partition on /dev/hda1

You have a spare drive /dev/hdc that is the same size or bigger

You want to create a RAID1 mirror of your root filesystem

You want to boot from root partition using GRUB (if not, this demo will still be interesting, but ignore GRUB bits)

You have kernel >=2.4 and mdadm installed

You have recent verified backups of all important files

Get Started

Create a partition on /dev/hdc1 that is the same size or larger than /dev/hda1

Create a new RAID device made of /dev/hdc1 plus a special virtual device called "missing" - we will fill this with hda1 later

# mdadm --create /dev/md0 --level=1 --raid-devices=2 /dev/hdc1 missing

Format /dev/md0

# mkfs.ext3 /dev/md0

Mount new /dev/md0

# mkdir /newroot
# mount /dev/md0 /newroot

Copy existing root filesystem onto /dev/md0

# cp -axv / /newroot

Reboot into RAID

Change grub (or lilo) to mount root as /dev/md0

title           Ubuntu, kernel 2.6.15-28-k7
root            (hd0,0)
kernel          /boot/vmlinuz-2.6.15-28-k7 root=/dev/md0 ro quiet splash
initrd          /boot/initrd.img-2.6.15-28-k7
savedefault
boot

Don't forget to copy /boot/grub/menu.lst to /newroot

# cp /boot/grub/menu.lst /newroot/boot/grub/menu.lst

Setup GRUB

grub> root (hd0,0)
grub> setup (hd0,0)
grub> root (hd2,0)
grub> setup (hd2,0)
grub> quit

Check everything

Reboot. If it fails to boot, use GRUB menu to go back to previous config.

Otherwise you should now be on /dev/md0

# df
# cat /proc/mdstat

Should show root filesystem is /dev/md0

Should show /dev/md0 has one device and one missing [U_]

MAKE SURE YOUR SYSTEM WORKS NOW. We are about to add /dev/hda1 to the array, overwriting it - point of no return!

# mdadm --manage /dev/md0 --add /dev/hda1
# cat /proc/mdstat

That's All Folks

You're done! Reboot and check everything is okay.

If something's gone wrong, you can use rescue disk to boot from /dev/hda1 or /dev/hdc1 - you can boot from either without needing RAID, they are normal ext3 filesystems! (Note: This only applies to RAID1! Very useful feature!)

You can observe RAID status with:

# cat /proc/mdstat
Personalities : [raid1] 
md0 : active raid1 hdc1[0] hda1[1]
      38074560 blocks [2/2] [UU]
unused devices:

mdmonitor can email you when a drive fails or other event action

Gotchas

mdadm will ignore it's own config file when booting from RAID (well, duh, how could it read the config file from something it's only just about to boot from?)

If you have multiple RAID arrays, mdadm boot autodetect will arrange arrays in order of drive letter: eg. two RAID volumes hda1 & hdc1=md0 , hdb1 & hdd1=md1

You can't change it so that hda1 & hdc1=md1 and vice versa - unless you don't boot from RAID

Personally I don't see much point having RAID that isn't bootable, if the boot partition fails then the system is dead

You can do really cool stuff by taking one drive out of a RAID1 array and putting it in another system! Instant system copy!

Public Domain - Andrew Oakley - 2007-04-18

Top - More Computing Articles - Article Index - aoakley.com