[plug] undelete files in ext3 file system

[Craig Ringer]

> Let us know if you find out more about this! Presumably such a system
> requires a bit of planning, to allocate fixed space on the same disk
> (maybe like a swap parition) where the versions can be stored

That was my understanding of how most of these 'layered' filesystems 
would work. That mechanism, of course, goes very well with the 2.6 
device-mapper based LVM2, where you can easily dynamically allocate and 
deallocate blocks of storage. I'm getting rather used to it with LVM1 on 
the Xeon at work, and would strongly suggest that /everybody/ who 
doesn't need to dual boot use it. It's wonderful.

> I
> suppose it would be nifty if past versions were accessible as regular
> file paths, like with AFS backups, so that this versioning system could
> be used without requiring extra user software when users are dealing
> with network / workgroup mounts.

I think it would generally be best to have history accessed via a 
separately mounted read-only FS, to avoid the horror of recursion on 
file history. Think /mnt/bigdisk and /mnt/bigdisk.history  as two 
separate filesystems. It's a concept a bit like snapshotting really, but 
a file-level snapshot updated at every change (preferably with a maximum 
frequency timer, too).

> Actually, this rings a bell for some
> reason. I seem to recall some sort of filesystem where the past
> revisions are not visible during directory traversal but can be
> retrieved if you use a special filename. This might not have been a real
> general-purpose filesystem "layer", though -- it might have been an
> application feature that was just supposed to look like a filesystem
> feature.

AFAIK, lots of bigger UNIXes have this. Hell, I think OpenServer might 
support filesystem versioning that way, but it doesn't do it by default.

Craig ringer