Migrating RAID Volumes Built Under the 0.4x (Old) Raid Tools to the 0.90 (New) Raid Tools



To test and illustrate this procedure I booted a brand new machine into the 2.2.16nfs3 kernel which uses the old RAID software, built a level 0 array and populated it with data. I then booted the same machine into 2.2.16-cfht which uses the new RAID software and converted the array using the following procedure. I believe that there are other ways to do this operation but I've tested this one and it does work; I "diff"ed the files in the array against an identical file structure on another volume and there was no data corruption and e2fsck reported no errors.

The old RAID software configures itself each time at startup via the /etc/mdtab file which is created manually or by using the mdcreate utility. The new software uses the /etc/raidtab file and self initializes at bootime using persistent superblocks and the "linux raid autodetect" partition type, which is mostly valuable for root filesystem arrays, and means the raidtab file is no longer used after the array is first created.

The general technique I used was to create a raidtab file that described the same array as the old mdtab file, then "create" the array with the new raidtools. Since the interleaving algorithm is the same for both, a new array is created which is identical to the old one and contains the original data. Removing the mdtab file prevents the old raid software from being invoked and the array is henceforth treated as a new type volume.

The mdtab and raidtab files are different syntactically but one can be manually translated to the other. An mdtab type description specifies a level 0 array like this:

/dev/md0        raid0,64k,0,cbb29e52    /dev/hda2 /dev/hdc2 /dev/hdd2 /dev/hde2 /dev/hdf2 /dev/hdg2 /dev/hdh2

Whereas a raidtab description of the same array would look like this:

raiddev /dev/md0
       raid-level      0
       nr-raid-disks  7
       nr-spare-disks  0
       chunk-size      64
#        persistent-superblock 1
       device          /dev/hda2
       raid-disk       0
       device          /dev/hdc2
       raid-disk       1
       device          /dev/hdd2
       raid-disk       2
       device          /dev/hde2
       raid-disk       3
       device          /dev/hdf2
       raid-disk       4
       device          /dev/hdg2
       raid-disk       5
       device          /dev/hdh2
       raid-disk       6

Note that the interleaving factor, or chunk-size is the same for both (64k), and that a persistent-superblock is not used. I tried it and it works but e2fsck complains about the superblock being incorrect though it shows no other errors.

Partition types for the old arrays did not need to be "linux RAID autodetect" because the software initialized the array via the mdtab file at startup. For the new RAID tools to initialize however the partition types must be set to autodetect, otherwise you will need to manually start the array after each reboot or via an rc script. The partition type can be changed without loosing any data.

To migrate and array such as the one above, first make a raidtab file from the mdtab file. Move the mdtab file out of the way:

mv /etc/mdtab /etc/mdtab.old

Change the system to boot from the new kernel and reboot. When it is up simply initialize the array with the new raidtools:

mkraid /dev/md0

Check it:

cat /proc/mdstat

Change the partition types to autodetect at this point if you feel comfortable with it and reboot to make sure it starts itself O.K. Check dmesg to make sure you have no errors.

-Kanoa