Re: ShadowFS and me :)

[Moritz Schulte]

José María <nav3gant3@terra.es> writes:

[one note: there are ideas about renaming shadowfs to unionfs, to
emphasize the similarity between that Hurdish {shadow,union}fs and
BSD's unionfs.]

> Let's go to rock. I have a problems: there isn't a ShadowFS 
> specification!!

True, there is no single paper describing the exact semantics of
shadowfs; but you can find many information in the mailinglist
archives.  Basically it should be something like BSD unionfs, just
more powerful (IIRC unionfs allows only two "filesystem layers"):

shadowfs lets you "stack" filesystem trees; it then provides a new,
virtual filesystem, which is the result of stacking the specified
filesystems.  The order does matter; if there is a node with the same
name on multiple filesystems, then the upper filesystem "shadows" the
lower filesystem.  shadowfs should have many nice features, like for
instance copy-on-write, which lets you manipulate the virtual shadowfs
without really manipulation the underlying filesystems.

Many things will be possible with shadowfs.  For example, the /usr
part of a GNU system can be on a different partition than the root
partition and later be "added" to the root filesystem with shadowfs.
shadowfs could be used to "collect" libraries from several places in
the system and bundle them in /lib.  Furthermore, users could have a
lot of fun with it: they could create a copy-on-write shadowfs in
their home directory with / as the underlying filesystem; after a
chroot in that new shadowfs, they have a "writable" root filesystem.

> I don't know if someone is working on it!!

I did, but that is quite some time ago.  There is one version,
shadowfs-0.1.9, which seems to work ok - although it has it's
limitations.  But that version uses the wrong approach internally and
should be rewritten; once I started to work on shadowfs-0.2 - Wolfgang
should have the latest code - but it doesn't really work yet.

Regarding the differences between 0.1.9 and 0.2: 0.1.9 doesn't keep
track of used nodes, it doesn't have a "node cache".  All looked up
nodes in 0.1.9 (means: directory nodes, since they are the only nodes
managed by shadowfs) were created on the fly and destroyed after use.
0.2 should contain a node cache, to improve performance and
functionality.  If you want to work on that, that would be great.

                moritz
-- 
moritz@duesseldorf.ccc.de - http://duesseldorf.ccc.de/~moritz/
GPG fingerprint = 3A14 3923 15BE FD57 FC06  B501 0841 2D7B 6F98 4199