I suggested something like this a few months back, and was dismissed out
of hand. Does this mean my idea wasn't completely insane after all?
The advantage of this method of course, is that it could work over any
filesystem, not just ext2.
Whether you use the kludge of a separate block device, or the cleaner
(but harder) solution of rewriting the block device layer, in either
case it will work over any filesystem, not just ext2.
In either case, you will need to make changes to the base filesystem to
support writing files in the correct order. So saying it will work for
any filesystem is true, but that statement is sorta like saying "all we
need to do is bell the cat".
The problem the kludged block device is that it has a all sorts of
problems which I listed --- takes up extra memory, slower because of the
enforced extra memory-to-memory copy, etc. It's also painful to
configure, since you'd need to create this extra pseudo device, and then
somehow configure it much like the current loopback devices are
configured with a losetup-like command *before* the filesystems get
mounted. So it complicates the boot scripts, too.
On the other hand, if someone is desparate for metadata logging, it's a
way to start on the filesystem modifications without requiring the block
device layer to be rewritten first.
- Ted
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu