Why don't we use the DCE UUID mechanism that already exists?
I thought about it; however, DCE UUID's are much more complicated
(although they are more extensible). At the cost of the extensibility,
though, they are a real pain to generate. This is a problem, because I
had originally intended for the kernel to gen a UUID and fill it into
the superblock if the UUID field as blank. However, it would be much
more complicated to do this with a DCE-style UUID.
Ultimately, it might be useful to put a DCE-style UUID generator in the
kernel, since some applications require a *large* number of UUID's, and
the DCE UUID really assumes that a central UUID authority per network
node is handing out UUID's.
However, for the ext2 superblock mechanism, these concerns don't arise,
since we won't be making new ext2 filesystems all the time --- but by
the same token, there's no reason to use the complicated DCE-style UUID
in the superblock, either, since there's no question of on-the-wire
conflict. So, we're probably much better off sticking with something
simple.
- Ted