:: Re: [DNG] OT: disk recovery softwar…
Góra strony
Delete this message
Reply to this message
Autor: Simon Hobson
Data:  
Dla: dng@lists.dyne.org
Stare tematy: Re: [DNG] OT: true read-only disk
Temat: Re: [DNG] OT: disk recovery software (was: true read-only disk)
While watching the disk struggle reading bits of metadata that's been read before, it gave me an idea for a tools - I just wish I even had a fraction of the skills needed to build it.
While the "image the whole thing and work form the image" does work - it doesn't work for disks like these f***ing Seagates with their crappy firmware, and it can need a disproportionately long time to get results. Even without the Seagate firmware problem, it can take weeks to finish imaging a large drive and you can't do anything at all with the copy until it's finished.

So I was thinking ...
To mount the partition/filesystem needs a certain amount of metadata to be readable, and in particular every time I have to restart the process, it has to re-read that information. So what if I had a cache that would cache stuff that had been read, so that you would only ever need to read any block off the faulty disk one time ?

The way I thought it might work is this ...
It would present a fake drive device that you'd use as a proxy for the real disk. Every time you read from this fake drive, it would do a back to back read against the real drive IF it hasn't already read that block - in which case it would just return the cached block. It would of course need a chunk of disk space large enough for an image of the whole drive (or partition) plus the overhead of a bitmap for which blocks have been cached, and which have had persistent read errors.
It would need options, perhaps interactive*, to determine whether it re-tries failed blocks, or just replies with a media error result to the calling program. I'm fairly certain that many of the blocks I get errors on have been read at some point, so over time it could build up a fairly good cache - and thus allow mounting a filesystem much more reliably.

* That might also need a helper program that would determine what the block contained**. Taking this case, if it's a media file I'd not bother retrying until I'd done all the easy stuff. But if it's filesystem metadata (sometimes when it fails, mount give the error "mount: /dev/sdf5: can't read superblock") then it needs retrying.

** Supplmentary question ...
Is there any tool which given a block number will work out what it belongs to ? Using ddrescue to image a drive, at the end you'll be left with a file of what couldn't be copied - which is pretty useless in terms of working out if any particular file is affected or not.



As for my progress ...
I've recovered about 25% of the TV recordings files, which is better than the "nothing" I was half expecting. Sometimes the disk will crap out before the filesystem is mounted, sometimes it won't - hence the idea of the caching setup above. I am certain that if I tried to image the drive then I'd be at it for weeks, and probably get nothing back - worse, I'd have an image but random files would have holes in them and I'd have no idea which. Doing it file by file I know whether a file is complete or not - and as I say, I have about 25% recovered with no errors.