How to get the offset in a block device of an inode in a deleted partition

Question

During a fresh installation, I accidentally formatted a disk containing datas. I have tried using some tools: testdisk, foremost, but I did not get good results. (see my unsuccessful post on superuser).

So I have decided to read some docs about ext2 filesystem structure, and I could get some results:

The deleted partition have a directory tree like that:

dev
|-scripts
|-projects
|-services
|-...
Medias
|-downloads
|-Musique
|-...
backup
...

So, based on the ext2 directory entry format:

Directory Entry
Starting_Byte Ending_Byte Size_in_Bytes Field_Description
0   3       4   Inode
4   5       2   Total size of this entry (Including all subfields)
6   6       1   Name Length least-significant 8 bits
7   7       1   Type indicator (only if the feature bit for "directory entries have file type byte" is set, else this is the most-significant 8 bits of the Name Length)
8   8+N-1   N   Name characters

I tried to find some datas matching this structure.
I used this script:

    var bindexOf = require('buffer-indexof');

    var currentOffset=0;
    var deviceReadStream = fs.createReadStream("/dev/sdb");

    deviceReadStream.on('error',function(err){
        console.log(err);
    });

    deviceReadStream.on('data',function(data){

        var dirs = ["dev","scripts","services","projects","Medias","downloads","Musique","backup"];
        dirs.forEach(function(dir){

            dirOctetFormat = new Buffer(2);
            dirOctetFormat.writeUInt8(dir.length,0);
            dirOctetFormat.writeUInt8(2,1);// type is directory
            dirOctetFormat= Buffer.concat( [dirOctetFormat, new Buffer(dir)]);

            var offset = bindexOf( data, dirOctetFormat );
            if( offset >= 0 ){    
                console.log( dir + " entry found at offset " + (currentOffset + offset) );
            } 

        });
        currentOffset += data.length;
    });
}

I found data which seems to be the directory entry of the dev directory:

 ===== Current offset: 233590226944 - 217.5478515625Gio ====== 
scripts entry found at offset 233590227030
services entry found at offset 233590227014
projects entry found at offset 233590228106

If it is the case, I got the inode numbers of its children directories: scripts, projects, services,...

But I do not know what to do with that! I tried to deduce the location of these inodes, based on this guide, but as I was unable to find a superblock of the deleted filesystem, I just have to make guesses about the block size, the number of blocks, ... and that seems a little bit fuzzy to me to hope obtaining a result.

So could you have some intervals for all values needed to obtain the offset of an inode, and a more formal formula to get this offset?

Francinefrancis

Answer 1

If you have only erased the partition table (or modified it) you can still get your data, if data has not been reused for something else.

ext2 filesystems have a MAGIC number in superblock, so to recover your partition you have only to search for it. I did this on one machine and was able to recover not one, but seven partitions in one disk. You have some chances to get invalid numbers, but just search for that magic. Magic number is defined in include/uapi/linux/magic.h and value is #define EXT2_SUPER_MAGIC 0xEF53 (it's found at offset #define EXT2_SB_MAGIC_OFFSET 0x38 ---from file include/linux/ext2_fs.h)

To search for the superblock, just try to find 0xef53 at offset 0x38 in one sector of the disk, it will mark the first block of the partition. Be careful, that superblock is replicated several times in one partition, so you'll find all the copies of it.

Good luck! (I had when it happened to me)

---

Edit (To illustrate with an example)

Just see the magic number in one of my own partitions:

# hd /dev/sda3 | head -20
00000000  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
*
00000400  40 62 08 00 00 87 21 00  26 ad 01 00 f6 30 15 00  |@b....!.&....0..|
00000410  1d 31 08 00 00 00 00 00  02 00 00 00 02 00 00 00  |.1..............|
00000420  00 80 00 00 00 80 00 00  90 1f 00 00 cf 60 af 55  |.............`.U|
00000430  fc 8a af 55 2d 00 ff ff  53 ef 01 00 01 00 00 00  |...U-...S.......|<- HERE!!!
00000440  36 38 9d 55 00 00 00 00  00 00 00 00 01 00 00 00  |68.U............|
00000450  00 00 00 00 0b 00 00 00  00 01 00 00 3c 00 00 00  |............<...|
00000460  46 02 00 00 7b 00 00 00  5a bf 87 15 12 8f 44 3b  |F...{...Z.....D;|
00000470  97 e7 f3 74 4d 75 69 12  72 6f 6f 74 00 00 00 00  |...tMui.root....|
00000480  00 00 00 00 00 00 00 00  2f 00 61 72 67 65 74 00  |......../.arget.|
00000490  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
*
000004c0  00 00 00 00 00 00 00 00  00 00 00 00 00 00 18 02  |................|
000004d0  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
000004e0  08 00 00 00 00 00 00 00  00 00 00 00 93 54 99 ab  |.............T..|
000004f0  aa 64 46 b3 a6 73 94 34  a3 79 46 28 01 01 00 00  |.dF..s.4.yF(....|
00000500  0c 00 00 00 00 00 00 00  e5 61 92 55 0a f3 02 00  |.........a.U....|
00000510  04 00 00 00 00 00 00 00  00 00 00 00 ff 7f 00 00  |................|
00000520  00 80 10 00 ff 7f 00 00  01 00 00 00 ff ff 10 00  |................|

Remember it is on offset 0x38 counted from the block origin, and assume the super block is the second block (block 0 reserved for bootcode, so it will be block 1, with two sectors per block, to make 1k blocksize) in the partition, so you'll have to rewind 0x438 bytes from the beginning of the magic number to get the partition origin.

I have run the command on my whole disk, getting the following result:

# hd /dev/sda | grep " [0-9a-f][0-9a-f]  53 ef" | sed -e 's/^/    /' | head
006f05f0  ee 00 00 11 66 0a 00 00  53 ef 00 00 11 66 2d 00  |....f...S....f-.|
007c21d0  55 2a aa 7d f4 aa 89 55  53 ef a4 91 70 40 c1 00  |U*.}...US...p@..|
20100430  fc 8a af 55 2d 00 ff ff  53 ef 01 00 01 00 00 00  |...U-...S.......|
2289a910  0f 8f 4f 03 00 00 81 fe  53 ef 00 00 0f 84 ce 04  |..O.....S.......|
230d4c70  0a 00 00 00 1c 00 00 00  53 ef 01 00 00 00 00 00  |........S.......|
231b7e50  a0 73 07 00 00 00 00 00  53 ef 0d 00 00 00 00 00  |.s......S.......|
23dbd230  d5 08 ad 2b ee 71 07 8a  53 ef c2 89 d4 bb 09 1f  |...+.q..S.......|
25c0c9e0  06 00 00 00 00 4f 59 c0  53 ef 32 c0 0e 00 00 00  |.....OY.S.2.....|
25d72ca0  b0 b4 7b 3d a4 f7 84 3b  53 ef ba 3c 1f 32 b9 3c  |..{=...;S..<.2.<|
25f0eab0  f1 fd 02 be 28 59 67 3c  53 ef 9c bd 04 30 72 bd  |....(Yg<S....0r.|

Clearly, there are much more uninteresting lines in this listing than the ones we need. To locate the one interesting here, we have to do some computing with the numbers. We have seen that sectors are 512 bytes long (this is 0x200 in hex) and we can have the superblock magic at offset 0x438, so we expect valid offsets to be at 0xXXXXXX[02468ace]38 only. Just select the lines with offsets ending in that expression, and you'll get the first superblock valid (in the third line) at offset 0x20100430. Substract 0x430 to give the byte offset of the partition (0x20100000, and then, divide the result by 0x200, giving 0x100800, or 1050624)

# fdisk -l /dev/sda | sed -e 's/^/    /'

Disk /dev/sda: 931.5 GiB, 1000204886016 bytes, 1953525168 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 4096 bytes
I/O size (minimum/optimal): 4096 bytes / 4096 bytes
Disklabel type: gpt
Disk identifier: DF97DAD4-727D-4BB3-BD7B-3C5A584A2747

Device         Start        End    Sectors   Size Type
/dev/sda1       2048     526335     524288   256M EFI System
/dev/sda2     526336    1050623     524288   256M BIOS boot
/dev/sda3    1050624   18628607   17577984   8.4G Linux filesystem  <-- HERE!!!
/dev/sda4   18628608   77221887   58593280    28G Linux filesystem
/dev/sda5   77221888   85035007    7813120   3.7G Linux filesystem
/dev/sda6   85035008  104566783   19531776   9.3G Linux filesystem
/dev/sda7  104566784  135817215   31250432  14.9G Linux swap
/dev/sda8  135817216  155348991   19531776   9.3G Linux filesystem
/dev/sda9  155348992 1953523711 1798174720 857.4G Linux filesystem

Infarction

Thank you for your answer. I already searched the ext2 signature without results. – Vowell

if the volume is large, you'll get at least the superblock copies.... or perhaps you have formatted the disk with the option to erase all disk. – Infarction

I was thinking that the new partition had overwritten all the superblocks copies. But it was because the sequence is written in little endian, so I had to search 0x53EF. ANd I got a lots of results! – Vowell

superblock copies are spread all along the partition (there's one per block group) so perhaps you have lost more than 80% of the volume. The little endian problem is illustrated in my example. Just have a look at the #define and the grep pattern to search (I have included the commands to illustrate in depth) – Infarction

I have checked in addition that the previous field to the magid number is the maximum number of mounts allowed. By default this field is initialized to -1 (0xff 0xff) So try to search using grep 'ff ff 53 ef' and you'll get out a lot of them. If you have not changed the maximum mount limit this can be of some help. – Infarction

Also, normally, fdisk -l shows you the minimum/optimal block size of the unit. This allows to check for different offset values. A 4Kb blocksize allows you to specify offsets with ^[0-9a-f]*430 as regexp, instead of the one I used (for 1kb block, ^[0-9a-f]*[02468ace]30 ) – Infarction