This would be the 'Heidemann Framework', presented in John Heidemann's
PhD thesis. It is implemented in 4.4BSD, and there are even some
applications of it:
- null fs(?): delegates all access to another file system:
mount -t null / /mnt
- null fs(?): return EINVAL for all operation. Used to unmount a file
system which has open files. All inodes are replace with null inodes
under each file.
- union fs: mount one fs on top of the other
- uid map filesystem: translates user ids
Used for NFS servers which have different uids than their clients.
There is some discussion of how one could split the flat file system
(inodes) and the hierarchical file system (directories), thus allowing
to put one hierarchy on top of another flat system. I don't think there
is a proof-of-concept implementation of that idea.
Anyway, I'm not sure whether such an architecture is worth the effort.
The only advantage of implementing the Heidemann framework for Linux
is that porting BSD fs drivers to Linux might be simpler.
Regards,
Martin