Power off the server, connect a disk to another computer which won’t recognize it as RAID (in particular, don’t connect to LSI RAID cards and HBAs). Make a dump (
dd if=/dev/sdX of=/path/to/an/image.img).
Now, you have virtually unlimited recovery attempts. You can:
- explore a dump contents with a hex editor and see if it still contains data
- experiment with the disk, including the rollback you mentioned; if you ever wash data from it, you can always rewrite disk from the dump (the dd command with if and of reversed), or
- experiment with the dump; for example, you can make it read-only, make read-write overlay image over it (which will contain changes) and do any forensics you are able to: scan for partitions and file systems, etc.
I want to point out again. If you have any valuable data you still hope to recover, always do dump first, before any other action takes place. Don’t try running any recovery operations if you aren’t absolutely sure you know what you doing until you have a clone lying in safe place. If you recover some kind of RAID, it is often required to have a clone of each component device to be able to retry after bad recovery attempt.
UPD: That’s true, regular automatic backup, monitored and tested, often saves many hours and some hair.