mdadm & raid6: does "re-create(with different chunk size) + resync" destroy orig. data?

Question

I would like to rescue data on my software raid-6 array. I did some stupid actions (described bellow) with this original array.

Main Question:

I need to know if original data stored on raid-6 array are definitely lost (or not) after the following actions have been prepared upon this array (executed in the order listed bellow):

1. zeroing super-blocks of all the active disks/partitions registered in the array

2. executing the "mdadm --create ..." command using different options (see bellow for list) than have been used when array have been created originally: -> different chunk size -> different layout -> different disks order

3. resync-ing the array

Note: Specific values of mdadm parameters should not be relevant here, since this is about the principle how mdadm works ...

I think points 1) & 2) should not even touch the original data since they are supposed to manipulate just superblocks

I see the 3) point as most critical from data lost point of view: I'm not sure what exactly is happening with array during resync, but based on heavy activity of all involved hard drives (for ~7 hours) I assume the data storage area is completely re-processed ...

Sub-questions:

1. Does ordering of hard drives/partitions (as they are ordered on mdadm command line) play role for raid6 creation & initial resync?

2. What everything is necessary to backup after array creation to be able to safely re-create the array in similar situation as my is (e.g. backup superblocks informations & partition table info for each disk involved in the array ...)?

Comment:

The mdadm wiki article (http://en.wikipedia.org/wiki/Mdadm) should be revised and author should be kicked in to ass a little bit, or more then just a little bit ...
The article mentions the zeroing superblocks & subsequent array re-creation as solution for getting rid of the "mdadm: Cannot open ...: Device or resource busy" issue.
Author somehow forgot to mention important step - to backup the parameters the original array (superblocks) as the first step ... and also my investigation seems to point out that the ordering of the involved disks/partitions plays role as well ...

Thanks for answers,
Peter

Answer 1

I looked at the wikipedia article, and the bit about deleting the superblock etc. is right, but it's supposed to be used for solving a different problem than what you were apparently trying to solve. The intention in the article is to "clean out" a drive that has old RAID settings on it so that you can use it in a new array.

You should, of course, never clear the superblock on a drive that currently contains data that you want to preserve. The superblock contains critical information regarding how the data on that drive is organized, so erasing or changing the superblock it is unhelpful unless you intend to throw away the data with it.

Answer 2

You're probably right about only affecting the superblocks until the drives began to re-sync, but at this point that's hardly a significant detail. By now your data is in an unknown state if it's even there at all.

I don't think you'll be helped any further by advice you read on the Internet, even from this site. Simply trying out things you read online is only a valid course of action if failure is not a problem.

If the data is very important to you, I'd recommend turning the drives over to a data recovery lab and letting them work on it. If your data is still intact in some form, they'll have the experience to know how to fix your problem without causing any further damage.

Edit
Luckily, since rebuilding your parity data should ONLY affect the parity data, you theoretically may still have just enough information left untouched to reconstruct your original content, assuming you started out with a clean and healthy array before this mess happened.

However, reconstructing it would be non-trivial and require a lot of work building a profile of the array by hand, especially if you changed some of the dimensions of your array when you overwrote the superblocks.

It would almost definitely not survive doing this sort of thing twice, though.

Answer 3

The confusing part is:

create and then resynced for 7 hours.

When an array is created, the filesystem is generally created on top and there is no resyncing. Zeroing the superblocks makes it difficult for the raid set to figure out which order the drives were in when the raidset tries to assemble itself automatically. What did you do with the create? create it with 2 drives, missing 2, then grow it and add the others? or did you create it as a raid-6 set and add all drives at the same time.

If the latter, it should not have resynced. IF you didn't grow the array and the drives resynced, it sounds almost like it detected a portion of the array. After the resync, did you have to make a filesystem or did a filesystem already exist?

If the chunk size changed, that's not good. If the alignment of the partitions changed, again, not good. Order of the drives changed, again, not good. You have a number of things that probably have made it extremely difficult to recover the data if at all.

The one thing that I find very curious is the resync. Raid-6 can take two drive failures and still have the ability to recover. But, some of the things you did almost make me think you're in a situation that is going to be almost unrecoverable except by experts.