Union mounts

[Bill Rugolsky Jr.]

On Sun, Oct 19, 2003 at 05:52:29PM -0400, Mike A. Harris wrote:
> Linux _period_ does not support union mounting, and never has - 
> at least to my knowledge.  I've heard kernel folk discussing that 
> this is something which might happen in 2.7.x however.  Just so 
> you know that there's nothing "Fedora" specific about the lack of 
> union mount capability.
 
Both union mount (i.e., union of directories), and unionfs (overlay
of filesystems) has been on Al Viro's TODO list for *years*.  He
intends to get the semantics right and avoid the problems of the
BSD implementations.  Linux's has multimount and private namespaces
complicate the matter somewhat:

http://groups.google.com/groups?hl=en&lr=&ie=UTF-8&oe=UTF-8&selm=linux.fsdevel.Pine.GSO.4.21.0111161318040.7552-100000%40weyl.math.psu.edu&rnum=17c%40ifi.uio.no&rnum=1&prev=/groups%3Fq%3Dalexander.viro%2Bunion%2Bmount%26hl%3Den%26lr%3D%26ie%3DUTF-8%26oe%3DUTF-8%26selm%3Dfa.lad3g7v.7ngljc%2540ifi.uio.no%26rnum%3D1

At some point Viro had a more-or-less complete union-mount and was
soliticing comments; see this thread:

http://groups.google.com/groups?hl=en&lr=&ie=UTF-8&oe=UTF-8&frame=right&th=3137774f238d4626&seekm=fa.lad3g7v.7ngljc%40ifi.uio.no#link1

Unfortunately, Viro has seemingly been occupied with higher priority
things during the 2.5 series, including cleaning up cruft, fixing
races, and cleaning up the block layer interfaces and dev_t.

Erez Zadok and his students at Columbia U. are working on the
more general problem of filesystem "fan-out" as part of the FiST
stackable template filesystem project.  Their work encompasses
unionfs, versioning, replication (where, e.g., one or more of the
filesystems could be remote mounts, like NFS/CIFS/AFS, etc.),
and other forms of stacking.

Zadok said back in August that they hoped to have something released
by late October.  Whether their semantic design will meet Viro's stringent
requirements remains to be seen.

Regards,

	Bill Rugolsky