0

I have a failing hard disk (let's call it sda) which contains, among other things, a 1.5 TB partition (let's call it sda3). There is another disk (sdb), which has a 1.5 TB partition (sdb1) as well. Both used to be part of an mdadm level 1 RAID using metadata version 1.2. Inside this RAID partition (let's call id md5), there is a LUKS encryption container (let's call it md5_uncrypted). This container should contain an ext4 partition.

At some place in time around August 10, 2012, I somehow restarted the RAID array with sdb1 missing and didn't even notice that. When I wanted to replace the RAID yesterday (three months later), I started copying data from sdb1 until I realized that it was out of date. So I took a look at the old sda3. By mistake, I ran mdadm --create instead of mdadm --assemble to restart md5 with only sda3 available. Accordingly, I ignored all warnings and let mdadm --create continue. cryptsetup didn't like the content of the new RAID. I didn't actually think mdadm --create would corrupt data if the same metadata version is used? Well, apparently it did.

I compared the first 20 MB of sda3 and sdb1 and noticed that they are equal starting at about 8 MB. So I copied the first 8 MB of sdb1 to sda3 (I have a backup of the old first 20 MB of sda3) and tried to assemble md5 (with only a single drive, sda3). Unfortunately, this gave me an error:

failed to add /dev/sdb1: Invalid argument

I also tried using the LUKS header from sdb1 on a freshly mdadm --created sda3, which cryptsetup happily did (of course), but it contained garbage.

My question is: Is there any chance to restore at least some of the data from sda3? Since I have the state from three months ago, everything helps, even just a list of files, or a list of files with modification dates.

Edit:

# mdadm --examine /dev/sdb1
         Magic : a92b4efc
       Version : 1.2
   Feature Map : 0x0
    Array UUID : 76a25475:70aa881c:dd30cc93:bbae03b7
          Name : ubuntu:0
 Creation Time : Fri Mar 16 20:52:16 2012
    Raid Level : raid1
  Raid Devices : 2

Avail Dev Size : 2930272256 (1397.26 GiB 1500.30 GB)
    Array Size : 1465129848 (1397.26 GiB 1500.29 GB)
 Used Dev Size : 2930259696 (1397.26 GiB 1500.29 GB)
   Data Offset : 2048 sectors
  Super Offset : 8 sectors
         State : clean
   Device UUID : b9012482:afa502cf:7794f4fb:2a0da196

   Update Time : Wed Nov 21 20:51:51 2012
      Checksum : 4e54a07 - correct
        Events : 15003


  Device Role : Active device 1
  Array State : .A ('A' == active, '.' == missing)

# mdadm --examine /dev/sda3
          Magic : a92b4efc
        Version : 1.2
    Feature Map : 0x0
     Array UUID : 54ea031e:42367512:b6a8675b:91f2cb6f
           Name : willow:5  (local to host willow)
  Creation Time : Wed Nov 21 18:03:35 2012
     Raid Level : raid1
   Raid Devices : 2

 Avail Dev Size : 2929999872 (1397.13 GiB 1500.16 GB)
     Array Size : 1464999744 (1397.13 GiB 1500.16 GB)
  Used Dev Size : 2929999488 (1397.13 GiB 1500.16 GB)
    Data Offset : 262144 sectors
   Super Offset : 8 sectors
          State : active
    Device UUID : 93c70c36:3cc540a5:13817695:bd4f327c

    Update Time : Wed Nov 21 18:03:35 2012
       Checksum : 321ddb3e - correct
         Events : 0


   Device Role : Active device 1
   Array State : .A ('A' == active, '.' == missing)

1 Answer 1

2

First if you have a spare HDD I would strongly advise to mirror sda3 and just work with the mirror.

mdadm --create with the same options shouldn't corrupt data unless defaults of unspecified options are changed between the version which initially created the array and the current version.

Did you compare the superblocks on sdb1 and sda3 with mdadm --examine?

Unless you've added/changed/removed keys the luks header should be identical. Have you tried to restore luksHeaderBackup from the sdb1 array to the created array on sda3?

Different offsets of the luks header {'L','U','K','S',0xba,0xbe} on sdb1 and sda3 would explain garbage in luks volume.

8
  • 1
    Thanks for your suggestions. I think, version 3.2.3-2 created the array, and right now I'm on 3.2.5-3, so there's probably no change in default values involved. As for comparing the superblocks – the original sda3 superblock is lost, so which should I compare against? I will have a look at the luks header positions; I didn't change the keys in the last three months on that system. Nov 22, 2012 at 15:39
  • Unfortunately, both LUKS headers start at exactly 1MB. Nov 22, 2012 at 16:04
  • And have exactly the same content. I wonder if I can't just take everything from the LUKS header to the end and try to luksOpen it? I would probably need to correct LUKS' payload offset (which currently has 1936220465 if I'm not mistaken). Nov 22, 2012 at 16:26
  • compare the old superblock from sdb1 against a newly created on sda3 and the one you made during the initial assemble/create mishap. Nov 22, 2012 at 16:43
  • 1
    1MB from the start of the partition I presume? And directly at the start of the raid array? 1936220465 sounds wrong for the payload offset, that would be beyond the end of HDD Nov 22, 2012 at 16:49

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .