Linux/mdadm

mdadm is a Linux utility used to manage software RAID devices.

The name is derived from the md (multiple device) device nodes it administers or manages, and it replaced a previous utility mdctl. The original name was "Mirror Disk", but was changed as the functionality increased.

It is free software licensed under version 2 or later of the GNU General Public License - maintained and copyrighted to Neil Brown of Suse.

Types of physical device
mdadm can handle anything which presents to the kernel as a block device. This can encompass whole disks (/dev/sda), partitions (/dev/sda1) and USB flash drives.

RAID Configurations

 * RAID 0 - Block level striping. MD can handle devices of different lengths, the extra space on the larger device is then not striped.
 * RAID 1 - Mirror.
 * RAID 4 - Like RAID 0, but with an extra device for parity.
 * RAID 5 - Like RAID 4, but with the parity distributed across all devices.
 * RAID 6 - Like RAID 5, but with two parity segments per stripe.
 * RAID 10 - Take a number of RAID 1 mirrorsets and stripe across them RAID 0 style.

Non-RAID Configurations

 * LINEAR - Concatenate a number of devices into a single large MD device.
 * MULTIPATH - Provide multiple paths with failover to a single device.
 * FAULTY - A single device which emulates a number of disk fault scenarios for testing and development.
 * CONTAINER - A group of devices managed as one, in which RAID systems can be built.

Types of MD device
The original (standard) form was /dev/mdn where n is a number between 0 and 99. More recent kernels have supported the use of names such as /dev/md/Home. Under kernel 2.4 and earlier these two were the only options. Both of them are non-partitionable.

From kernel 2.6 a new type of MD device was introduced, a partitionable array. The device names were modified by changing md to md_d. The partitions were identified by adding pn; thus /dev/md/md_d2p3 for example.

From kernel 2.6.28 non-partitionable arrays can be partitioned, the partitions being referred to in the same way as for partitionable arrays: /dev/md/md1p2.

Booting
Since support for MD is found in the kernel, there is an issue with using it before the kernel is running. Specifically it will not be present if the boot loader is either (e)LiLo or GRUB legacy. It may not be present for GRUB 2. In order to circumvent this problem a /boot filesystem must be used either without md support, or else with RAID1. In the latter case the system will boot by treating the RAID1 device as a normal filesystem, and once the system is running it can be remounted as md and the second disk added to it. This will result in a catch-up, but /boot filesystems ought to be small.

Create an array
mdadm --create /dev/md0 --level=mirror --raid-devices=2 /dev/sda1 /dev/sdb1

Create a RAID 1 (mirror) array from two partitions. If the partitions differ in size, the array is the size of the smallest partition. You can create a RAID 1 array with more than two devices. This gives you multiple copies. Whilst there is little extra safety in this, it makes sense when you are creating a RAID 5 array for most of your disk space and using RAID 1 only for a small /boot partition. Using the same partitioning for all member drives keeps things simple.

mdadm --create /dev/md1 --level=5 --raid-devices=3 /dev/sda2 /dev/sdb2 /dev/sdc2

Create a RAID 5 volume from three partitions. If the partitions used in your RAID array are not the same size, mdadm will use the size of the smallest from each partition. If you receive an error, such as: "mdadm: RUN_ARRAY failed: Invalid argument", make sure your kernel supports (either via a module or by being directly compiled in) the raid mode you are trying to use. Most modern kernels do, but you never know...

It is possible to create a degraded mirror, with one half missing by replacing a drive name with "missing": mdadm --create /dev/md1 --level=1 --raid-devices=2 /dev/sdb1 missing

The other half mirror is added to the set thus: mdadm --manage /dev/md1 --add /dev/sda1

This is useful when you are adding a disk to a computer which currently isn't mirrored. The new drive is... The computer is then booted off the secondary drive (or a rescue disk), the now idle original disk can be repartitioned if required (no need to format), and then the primary drive submirrors are added.
 * partitioned to match the first (unless you are also repartitioning too)
 * turned into a set of "half-mirrors"
 * formatted with appropriate file system
 * mounted
 * the data is copied over,
 * made bootable
 * its grub config and  mounts changed

Note that the partition types should be changed to 0xFD with fdisk to indicate that they are mirrored devices.

Replace failing disk
First mark the failing disk failed....

mdadm --manage /dev/md127 --fail /dev/sdc1

Then remove it from the array

mdadm --manage /dev/md127 --remove /dev/sdc1

After which it is safe to power down and physically replace the disk.

Then create partition as needed, and add new disk to the array;

mdadm --manage /dev/md127 --add /dev/sdc1

Then check /proc/mdadm for the sync of the device

Recording the array
mdadm --detail /dev/md0

View the status of the multi disk array md0.

mdadm -Es | grep md0 >>/etc/mdadm/mdadm.conf -E examine -s scan

This adds md0 to the configuration file so that it is recognised next time you boot.

You may wish to keep a safe copy of /proc/mdstat. The information will allow you to restart the array manually if mdadm fails to do so.

Growing an array by adding devices
mdadm --add /dev/md1 /dev/sdd1 mdadm --grow /dev/md1 --raid-devices=4

This adds the new device to the array then grows the array to use its space.

In some configurations you may not be able to grow the array until you have removed the internal bitmap. You can add the bitmap back again after the array has been grown. mdadm --grow /dev/md1 -b none mdadm --grow /dev/md1 -b internal

Growing an array by upgrading devices
An array may be upgraded by replacing the devices one by one, either as a planned upgrade or ad hoc as a result of replacing failed devices.

mdadm /dev/md1 --fail /dev/sda1 (replace the first drive with the new, larger one then partition it) mdadm --add /dev/md1 /dev/sda1

Allow the new drive to resync. If replacing all the devices repeat the above for each device, allowing the array to resync between repetitions. Finally, grow the array to use the maximum space available and then grow the filesystem(s) on the RAID array to use the new space.

mdadm --grow /dev/md1 --size=max

Deleting an array
mdadm --stop /dev/md0 # to halt the array mdadm --remove /dev/md0 # to remove the array mdadm --zero-superblock /dev/sd[abc]1 # delete the superblock from all drives in the array (edit /etc/mdadm/mdamd.conf to delete any rows related to deleted array)

Convert an existing partition to RAID 5
Assume that the existing data is on /dev/sda1:

mdadm --create /dev/md1 --level=5 --raid-devices=3 missing /dev/sdb2 /dev/sdc2 mdadm -Es >>/etc/mdadm/mdadm.conf update-initramfs -u dd if=/dev/sda1 of=/dev/md1 (add /dev/md1 to your boot loader's menu) (reboot into /dev/md1) mdadm --add /dev/md1 /dev/sda1 (update your boot loader)

Notes:
 * 1) A partition may be given as missing to act as a placeholder so that it can be added later.
 * 2) The /boot directory should be elsewhere, possibly on /dev/md0 or its own partition.
 * 3) If the reboot fails, do NOT add /dev/sda1 into the array until the problem is corrected!

Known problems
A common error when creating RAID devices is that the dmraid-driver has taken control of all the devices that are to be used in the new RAID device. Error-messages like this will occur:

mdadm: Cannot open /dev/sdb1: Device or resource busy

Typically, the solution to this problem involves adding the "nodmraid" kernel parameter to the boot loader config. Another way this error can present itself is if the device mapper has its way with the drives. Issue 'dmsetup table' see if the drive in question is listed. 'dmsetup remove ' will remove the drive from device mapper and the "Device or resource busy" error will go away as well.

RAID already running
First check if the device isn't in use in another array: cat /proc/mdstat

Probably you will have to stop the array with: mdadm --stop /dev/

Check the /etc/mdadm/mdadm.conf file (and restart system if possible): vi /etc/mdadm/mdadm.conf

Then you should be able to delete the superblock of this device: mdadm --misc --zero-superblock /dev/sdxN

Now the device shouldn't be busy any more.

Sometimes dmraid "owns" the devices and will not let them go. There is a solution.

Tweaking the kernel
To solve this problem, you need to build a new initrd without the dmraid-driver. The following command does this on a system with the "2.6.18-8.1.6.el5"-kernel: mkinitrd --omit-dmraid /boot/NO_DMRAID_initrd-2.6.18-8.1.6.el5.img 2.6.18-8.1.6.el5

After this, the system has to be rebooted with the new initrd. Edit your /boot/grub/grub.conf to achieve this.

Alternatively if you have a self customized and compiled kernel from a distro like Gentoo (the default option in gentoo) which doesn't use initrd then check kernel .conf file in /usr/src/linux for the line
 * 1) CONFIG_BLK_DEV_DM is not configured

If the above line is set as follows: CONFIG_BLK_DEV_DM=yes

then You might have to disable that option, recompile the kernel, put it in /boot and finally edit grub conf file in /boot/grub. PLEASE be careful NOT to disable CONFIG_BLK_DEV_MD=yes

(Note the MD instead of DM) which is essential for raid to work at all!

If both methods have not helped you then booting from live CD probably will (the below example is for starting a degraded raid-1 mirror array and adding a spare hdd to it and syncing. Creating a new one shouldn't be more difficult because the underlying problem was 'Device or resource busy' error): modprobe raid1 mknod /dev/md1 b 9 1 mknod /dev/md3 b 9 3 mdadm --assemble /dev/md1 /dev/hda1 mdadm --assemble /dev/md3 /dev/hda3 mdadm --add /dev/md1 /dev/hdb1 mdadm --add /dev/md3 /dev/hdb3

It might be easier to try and automatically assemble the devices mdadm --assemble --scan

Remember to change the corresponding md* and hd* values with the corresponding ones from your system. You can monitor the sync progress using: cat /proc/mdstat

When the sync is done you can reboot in your Linux normally.

Zeroing the superblock
Another way to prevent the kernel autostarting the raid is to remove all the previous raid-related information from the disks before proceeding with the creation, for example: mdadm --stop /dev/md0 mdadm --zero-superblock /dev/sd[abcd]1

And now the usual create, for example: mdadm --create /dev/md0 --level=5 --raid-devices=4 --spare-devices=0 /dev/sd[abcd]1

Recovering from a loss of raid superblock
There are superblocks on the drives themselves and on the raid (apparently). If you have a power failure, hardware failure, that does not include the drives themselves, and you cannot get the raid to recover in any other way, and wish to recover the data, proceed as follows:

Get a list of the devices in the raid in question: mdadm --detail /dev/md[x]

Result something like this:

/dev/md127: Version : 1.2 Creation Time : Sun Aug 21 23:35:28 2011 Raid Level : raid6 Array Size : 7814047744 (7452.06 GiB 8001.58 GB) Used Dev Size : 1953511936 (1863.01 GiB 2000.40 GB) Raid Devices : 6 Total Devices : 4 Persistence : Superblock is persistent Update Time : Sun Jan 1 11:43:17 2012 State : clean, degraded Active Devices : 4 Working Devices : 4 Failed Devices : 0 Spare Devices : 0 Layout : left-symmetric Chunk Size : 512K Name : clop:1 (local to host clop) UUID : 7ee1e93a:1b011f80:04503b8d:c5dd1e23 Events : 62 Number  Major   Minor   RaidDevice State 0      8       33        0      active sync   /dev/sdc1 1      8       49        1      active sync   /dev/sdd1 2      8       81        2      active sync   /dev/sdf1 3      8       65        3      active sync   /dev/sde1 4      0        0        4      removed 5      0        0        5      removed

RaidDevice order (sdc1,sdd1,sdf1,sde1) and Chunk Size are critical

Record all your raid member parameters: mdadm --examine /dev/sd[abcde...]1 | egrep 'dev|Update|Role|State|Chunk Size'

Look carefully at the Update time. If you have raid members attached to the motherboard and others attached to a raid card, and the card fails, but leaves enough members to keep the raid alive, you want to make a note of that. Look at Array State and Update Time. For example:

/dev/sdc1: Update Time : Wed Jun 15 00:32:35 2011 Array State : AAAA.. ('A' == active, '.' == missing) /dev/sdd1: Update Time : Thu Jun 16 21:49:27 2011 Array State : .AAA.. ('A' == active, '.' == missing) /dev/sde1: Update Time : Thu Jun 16 21:49:27 2011 Array State : .AAA.. ('A' == active, '.' == missing) /dev/sdf1: Update Time : Thu Jun 16 21:49:27 2011 Array State : .AAA.. ('A' == active, '.' == missing) /dev/sdk1: Update Time : Tue Jun 14 07:09:34 2011 Array State : ....AA ('A' == active, '.' == missing) /dev/sdl1: Update Time : Tue Jun 14 07:09:34 2011 Array State : ....AA ('A' == active, '.' == missing)

Devices sdc1, sdd1, sde1 and sdf1 are the last members in the array and will rebuild correctly. sdk1 and sdl1 left the array (in my case due to a raid card failure).

Also note the raid member, starting with 0, the raid needs to be rebuilt in the same order. Chunk size is also important.

Zero the drive superblocks

mdadm --stop /dev/md0 # to halt the array mdadm --remove /dev/md0 # to remove the array mdadm --zero-superblock /dev/sd[cdefkl]1

Reassemble the raid

mdadm --create /dev/md1 --chunk=4096 --level=6 --raid-devices=6 /dev/sdc1 /dev/sdd1 /dev/sdf1 /dev/sde1 missing missing

'missing' tell the create command to rebuild the raid in a degraded state. sdk1 and sdl1 can be added later

Edit /etc/mdadm.conf and add an ARRAY line with a UUID. First get the UUID for your raid:

mdadm -D /dev/md

then:

nano /etc/mdadm.conf

and add something similar to the file (notice there is no # in front of the active line you are adding

ARRAY /dev/md1 UUID=7ee1e93a:1b011f80:04503b8d:c5dd1e23
 * 1) ARRAY /dev/md0 UUID=3aaa0122:29827cfa:5331ad66:ca767371
 * 2) ARRAY /dev/md1 super-minor=1
 * 3) ARRAY /dev/md2 devices=/dev/hda1,/dev/hdb1

Save with 

Last, mark the array possilbly dirty with: mdadm --assemble /dev/md1 --update=resync

Monitor the rebuild with

watch -n 1 cat /proc/mdstat

All your data should be recovered!

You can also use Netdata to monitor your mdadm disk arrays.

Increasing RAID ReSync Performance
In order to increase the resync speed, we can use a bitmap, which mdadm will use to mark which areas may be out-of-sync. Add the bitmap with the grow option like below : mdadm -G /dev/md2 --bitmap=internal


 * Note: mdadm - v2.6.9 - 10th March 2009 on Centos 5.5 requires this to be run on a stable "clean" array. If the array is rebuilding the following error will be displayed:


 * mdadm: failed to set internal bitmap.


 * And the following line is added to the log file:
 * md: couldn't update array info. -16

then verify that the bitmap was added to the md2 device using cat /proc/mdstat

you can also adjust Linux kernel limits by editing these files /proc/sys/dev/raid/speed_limit_min and /proc/sys/dev/raid/speed_limit_max

You can also edit this with the sysctl utility sysctl -w dev.raid.speed_limit_min=50000

Increasing RAID5 Performance
To help RAID5 read/write performance, setting the read-ahead & stripe cache size for the array provides noticeable speed improvements.

Note: This tip assumes sufficient RAM availability to the system. Insufficient RAM can lead to data loss/corruption

echo 16384 > /sys/block/md0/md/stripe_cache_size blockdev --setra 16384 /dev/md0 Write Performance: dd if=/dev/zero of=/mnt/family/10gb.16384k.stripe.out bs=1M count=10240 10240+0 records in 10240+0 records out 10737418240 bytes (11 GB) copied, 94.5423 s, 114 MB/s

Read Performance: dd if=/mnt/family/10gb.16384k.stripe.out of=/dev/null bs=1M 10240+0 records in 10240+0 records out 10737418240 bytes (11 GB) copied, 28.5435 s, 376 MB/s

These changes must be done on any reboot (add to an init script to set on start-up)

Monitoring Status

 * , included in  package

cat /proc/mdstat
Review different  outputs :

Output with no devices configured: cat /proc/mdstat Personalities : [linear] [multipath] [raid0] [raid1] [raid6] [raid5] [raid4] [raid10] unused devices: cat /proc/mdstat Personalities : [raid1] [raid6] [raid5] [raid4] md_d0 : active raid5 sde1[0] sdf1[4] sdb1[5] sdd1[2] sdc1[1] 1250241792 blocks super 1.2 level 5, 64k chunk, algorithm 2 [5/5] [UUUUU] bitmap: 0/10 pages [0KB], 16384KB chunk

unused devices: