Bug#534470: Using mdadm 2.6.7.2 to assemble a raid array created withmdadm 1.9.0 will corrupt it making mdadm 1.9.0 to crash whentrying to reassemble

[NeilBrown]

On Fri, June 26, 2009 7:25 pm, RUSSOTTO François-Xavier 200103 wrote:
> Thank you for your answer Neil.
>
> I assumed that an update to the superblock checksumming routine was the
> reason though I didn't find very much information about this update and
> potential consequences on older systems. But, above all, I am very
> surprised that the new mdadm does not even warn the user about potential
> raid corruption when used on an old one.

Yes, I guess it might be possible to put some useful message somewhere...

>
> As you suggested, I tried recreating the array using option
> --assume-clean. So, I booted a Sarge cdrom, checked that chunksize and
> layout was default ones then I typed:
>
> $ mdadm --create /dev/md0 --level=raid5 --raid-devices=3 --auto=yes
> --assume-clean /dev/sda2 /dev/sdb2 /dev/sdc2
>
> And that worked! recreating the array without loss of data: I ran fsck on
> /dev/md0, no error were reported, then mounted it to check that all data
> was there; no pb.
>
> Then, I rebooted the server and... GOT A KERNEL PANIC AGAIN... this time
> for: invalid uuid ! I patiently rebooted the Sarge cdrom again and tried
> to reassemble the raid updating the uuid (which I fortunately backed up),
> using:
>
> $ mdadm --assemble /dev/md0 --auto=yes --uuid=#old#:#uuid#:#back#:#up#
> --update=uuid /dev/sda2 /dev/sdb2 /dev/sdc2
>
> Then, I got an error from mdadm saying that option "uuid" for --update was
> not valid! Only "sparc2.2", "summaries" and "resync" seams supported on
> that version of mdadm (1.9.0).
>
> Now, I don't know what to do and would be grateful for any suggestion to
> make my system boot again.

Use --create --assume-clean again but add the --uuid= option there.

If that doesn't work (and I'm not 100% sure it will), you will need to
find a way to rebuild your initrd.

NeilBrown