OS6 now works on x86 Legacy WARNING: NO NTGR SUPPORT!

mangrove wrote:

This is partly correct, in that it is NOT a completely data-destructive operation, but its a metadata-destructive one. It is also complicated by the fact that there is a new, blank filesystem on top of the old one, but the simple XORing of a resync isn't a data-destructive operation in itself (it doesn't write zeros, it XORs the existing data, MD doesn't know or care about which file system is on the resulting datablob). Data is still there, but it might be hard getting it out.

We might be talking cross-purposes here.

You're talking about the MD driver and how it distributes across an array; which is - as you say - filesystem agnostic.

But, in the process of upgrading to 6.x he's gone from ext4 as the filesystem on that MD device to another filesystem that has an utterly different on-disk data structure.

Imagine the on disk data "blocks" like this:
For ext4: XXYYYY
For btrfs: XYXYXY
Where X is a bit of metadata, and Y is actual data.
(This is only an example, the structure is utterly more complex and not like this in any way - it serves only as example)

The process of initialising the btrfs filesystem *may* have created those metadata blocks all over the disk, where actual data may have been in the ext4 filesystem. This depends on the init process for btrfs, which I have no idea about.

So, even if you could trick the system into recognising the filesystem as ext4 (unlikely at best), the on-disk data is going to be completely in the wrong format for it to try and recover anything.