On Fri, May 23, 2008 7:59 am, Anders Johansson wrote:

Hi,

I have a fully functioning raid5 array with 4 disks that will not start.
The following has happened:

1. I created the array with 3 disks and ran it for about two years.
2. I recently grew the array to 4 disks, no problem.

I assume the 'grow' completed ?

3. I upgraded to kernel 2.6.18.6 (Debian testing) mdadm v2.6.4.
4. After the upgrade, the array would not start /proc/mdstat marked
   the three original disks as removed and the recently added one as
   active.

I don't suppose you have kernel logs (/var/log/kern*) from the first reboot
into the new kernel that show the array being assembled?

5. I was stupid and assumed that re-add meant add without reconstruct
   and added the three missing disks.

It does, but only if the metadata confirms that this is safe.

6. The array is now still intact but will not start as the disks I
   re-added are marked as spares.

There seems to be two problems here:
The kernel upgrade did something?
How can I mark the spare disks as active without touching the data?


I did test create, but did not complete it. The output is:
mdadm --create /dev/md0 --level=5 --raid-devices=4 --layout=left-symmetric
--chunk=64 --assume-clean /dev/sdc1 /dev/sda1 /dev/sdb1 /dev/sdd1
mdadm: /dev/sdc1 appears to be part of a raid array:
    level=raid5 devices=4 ctime=Fri Mar  9 12:54:19 2007
mdadm: /dev/sda1 appears to be part of a raid array:
    level=raid5 devices=4 ctime=Fri Mar  9 12:54:19 2007
mdadm: /dev/sdb1 appears to be part of a raid array:
    level=raid5 devices=4 ctime=Fri Mar  9 12:54:19 2007
mdadm: /dev/sdd1 appears to be part of a raid array:
    level=raid5 devices=4 ctime=Fri Mar  9 12:54:19 2007
Continue creating array? no
mdadm: create aborted.

You should complete the create.  As you are using "assume-clean"
doing this will not affect any data on the device - only the metadata.
And that is corrupt already.
I would run "fsck -n" after the create to check that the data looks right,
then mount the filesystem readonly and have a look around, but
I suspect everything will be fine.

NeilBrown