[ILUG] RAID-5 recovery

[Ronan Cunniffe]

On 28/02/12 07:23, Niall O Broin wrote:
> On 27 Feb 2012, at 23:22, Ronan Cunniffe wrote:
>
>>     (yeah, first post to the list in ages, and it's 'cos I need help.... )
>>
>>     I have an mdadm RAID-5 array with - you guessed it - 2 failed disks.
>>
>>     Well, somebody else had it and didn't monitor it; so now it's dead, and has landed on my desk (the data is irreplaceable, of course).
>>
>>     Some dd_rescue'ing later, I have images of all the disks, 2 of them with ~5-10% unreadable each.  The good news is that the unreadable bits mostly don't overlap, so I actually have N-1 good data for ~99% (maybe all - still checking) of the array, and so it's at least theoretically possible to reconstruct the whole thing.
>>
>> Clearly, I need a tool that lets me nominate which-disk-is-dead for different block ranges.  I don't think mdadm will help, so does anybody know of anything like this?
> Never came across anything like that, but two thoughts occur:
>
> 1) Get another disk, and assemble it from the readable parts 2 slightly dead disks.
I think this will be complicated, because for each stripe, one of the 
disks is parity, not "real" data: if I reconstruct a disk out of the 
working bits, some of the reconstruction will be bit-inverted due to 
parity calculations.  Whatever-it-is needs to do a *real* RAID-5 
reconstruction using a map of which disk is dead.

2) Write a kernel module which returns blocks from disk A when asked for blocks from disk B for the relevant block ranges.

A kernel module?  This is how you'd solve it? Niall, I am in awe! ;-)
> Niall
>