[[meta copyright="Copyright © 2009 Free Software Foundation, Inc."]]

[[meta license="""[[toggle id="license" text="GFDL 1.2+"]][[toggleable
id="license" text="Permission is granted to copy, distribute and/or modify this
document under the terms of the GNU Free Documentation License, Version 1.2 or
any later version published by the Free Software Foundation; with no Invariant
Sections, no Front-Cover Texts, and no Back-Cover Texts.  A copy of the license
is included in the section entitled
[[GNU_Free_Documentation_License|/fdl]]."]]"""]]

[[meta title="Union Mounts"]]

When setting a translator on Hurd -- similar to mounting a file system on UNIX
-- the new node(s) exported by the translator are obscuring the original node
where the translator is set, and any nodes below it in the directory tree. The
translator itself can access the underlying node (which is a very nice feature,
as it allows translators presenting the contents of the node in a different
format); but it's no longer accessible from the "outside".

Plan9 has a feature where a file system can be mounted in union mode: the new
file system doesn't obscure the mount point in this case, but instead the
contents are combined. (This feature has also been under discussion in Linux
for a couple of years now, under the label "VFS-based union mounts".)

This kind of union mounts is generally useful, as it's sometimes more
convenient than unioning existing filesystem locations with unionfs -- it's not
necessary to mount a file system that is to be unioned at some external
location first: just union-mount it directly at the target location.

But union mounts also allow creating passive translator hierarchies: If there
is a passive translator on a parent node, and further passive translators on
child nodes, the union mount allows the child nodes with the further translator
settings still to be visible after the parent translator has started.

This could be useful for device nodes for example: let's say we have an
ethernet multiplexer at /dev/veth. Now the virtual subnodes could all be
directly under /dev, i.e. /dev/veth0, /dev/veth1 etc., and explicitely refer to
the main /dev/veth node in the translator command line. It would be more
elegant however to store the virtual nodes direcly below the main multiplexer
node -- /dev/veth/0, /dev/veth/1 etc.

There are two possible approaches how union mounts could be implemented in the
Hurd. The first one is to let the various translators handle union mounts
internally, i.e. let them present the underlying nodes to the clients in
addition to the actual nodes they export themselfs. This probably can be
implemented as some kind of extension to the existing netfs and diskfs
libraries.

The other possible apporach is less efficient and probably more tricky, but
probably also more generic: create a special unionmount translator, which
serves as a kind of proxy: setting the union-mounted translator on some
internal node; and at the actual mount location, presenting a union of the
nodes exported by this translator, and the nodes from the underlying file
system.

The goal of this project is implementing union mounts using either of the
approaches described above. (Though it might be useful initially to prototype
both for comparision.) The ethernet multiplexer shall serve as an example use
case -- any changes necessary to allow using it with the union mount
functionality are also to be considered part of the task.