Re: Brocken Raid & LUKS

[Phil Turmel <philip@turmel.org>]

On 02/22/2013 09:58 AM, Stone wrote:
> Am 22.02.2013 14:53, schrieb Phil Turmel:
>> On 02/22/2013 05:31 AM, stone@heisl.org wrote:
>>> to work on the live cd is very slow.
>>> i will kick out my two system drives and take one new and install a old
>>> system (ubuntu 11.04, i think on this system i have created the first
>>> time the raid) to it.
>>>
>>> do you have new infos from the hexdump or other news to try out some
>>> things the get the raid and the luks running?
>> Unfortunately, no.  The hexdump had no real superblock candidates that I
>> could see.  That strongly suggests that there remain some ordering
>> issues.  I would try chunk sizes down to 8k.  If that still doesn't
>> work, consider re-creating with a different drive order--it's a slim
>> possibility that "sdc1 sdd1 missing sdf1" isn't correct.
>>
>> Meanwhile, you haven't supplied the complete hexdump of your luks
>> signature sector.  It may not help, but it would show the payload offset.

What about this part?

> i have installed the system now with one system drive.
> the raid devices are now: sdb1 sdc1 sdd1(brocken not sync) sde1

Ok.

> i have now tested all chunk's from 512k to 8k
> 512 Open Luks but no superblock
> 256 Open Luks but no superblock
> 128 No key available with this passphrase
> 64 No key available with this passphrase
> 32 No key available with this passphrase
> 16 No key available with this passphrase
> 8 No key available with this passphrase

Ok, but on the smaller chunk sizes, the device order could impact
interpretation of the key material.  You should repeat the small chunk
tests with the drive order variations below.

Make a grid with chunk size on one axis, and drive order on the other
axis.  Mark each combination with yes or no if it can open luks.  If it
can, save the output of "fsck -n" in a file.  This would be a good thing
to script.

After the script is done, look at all the saved files to see if any look
like possible solutions.

> 512k and 256k working better...
> next tests:
> mdadm --create /dev/md2 --assume-clean --chunk=512 --verbose --level=5
> --raid-devices=4 /dev/sde1 /dev/sdb1 missing /dev/sdc1
>     No Luks
> mdadm --create /dev/md2 --assume-clean --chunk=512 --verbose --level=5
> --raid-devices=4 /dev/sdc1 /dev/sdb1 missing /dev/sde1
>     No Luks
> mdadm --create /dev/md2 --assume-clean --chunk=512 --verbose --level=5
> --raid-devices=4 /dev/sdc1 missing /dev/sdb1 /dev/sde1
>     No Luks
> mdadm --create /dev/md2 --assume-clean --chunk=512 --verbose --level=5
> --raid-devices=4 /dev/sdb1 /dev/sde1 /dev/sdc1 missing
>     fsck.ext4: Invalid argument while trying to open /dev/mapper/md2_nas
>     fsck.ext4: Bad magic number in super-block while trying to open
> /dev/mapper/md2_nas
> mdadm --create /dev/md2 --assume-clean --chunk=512 --verbose --level=5
> --raid-devices=4 /dev/sde1 /dev/sdc1 /dev/sdb1 missing
>     No Luks
> mdadm --create /dev/md2 --assume-clean --chunk=512 --verbose --level=5
> --raid-devices=4 /dev/sdc1 /dev/sde1 /dev/sdb1 missing
>     No Luks
> mdadm --create /dev/md2 --assume-clean --chunk=512 --verbose --level=5
> --raid-devices=4 /dev/sdb1 /dev/sdc1 /dev/sde1 missing
>     fsck.ext4: Invalid argument while trying to open /dev/mapper/md2_nas
>     fsck.ext4: Bad magic number in super-block while trying to open
> /dev/mapper/md2_nas
> 
> do you think that i should try do mount the partion as RO? but i think
> this is not working because the damaged filesystem. right?

Do *not* mount at all.  Even a read-only mount isn't really
read-only--it will try to play back the journal, and will try to write
to the superblocks.

Phil