Summer of Code: Magic Links, FiST/vnode stacking?, ReiserFS

[Erez Zadok]

In message <20050617020803.GC95979 at afields.ca>, Allan Fields writes:

> * FiST / vnode stacking (Coordinating w/ Erez Zadok?):
> 
> This is one I wanted to see move forward on *BSD.  What FiST offers
> is a unified approach for template file systems which can aid in
> cross-platform development of filesystem code (potentially) saving
> a significant amount of developer effort and duplication.  It's
> been around for a number of years now and FreeBSD templates do
> exist.

Good timing: we just released fistgen-0.1.2, which contains updated ports to
FreeBSD-5.x.  You may get it from

	ftp://ftp.filesystems.org/pub/fist/fistgen-0.1.2.tar.gz

> The Size-Changing Algorithm (SCA) code is Linux specific,
> so I'm not certain of the status.

The SCA code is indeed only in the Linux templates and hasn't been ported to
the FreeBSD ones.  There are also several SCA-related bugs in our bugzilla
server, yet to be fixed.

> Areas left to address: FreeBSD Templates (remaining build issues,
> keep templates up to date, etc.), SCA: Size-Changing-Algorithm (some
> work on this internally w/ the Stony Brook team?), 

Our group would love to work with anyone in the freebsd community to get
more and stable stackable file systems into freebsd.

> Cache Coherency issues, etc.

Cache coherency has always been an issue with stackable file systems.  What
OSs need is something like Heindeman did in his SOSP 1995 paper.

> [http://www.fsl.cs.sunysb.edu , http://www.filesystems.org ]
> 
> Desirable/Related File System Ports: ncryptfs and/or ecryptfs,
> gzipfs, unionfs

[ne]cryptfs would indeed be very nice to have, but I thought that freebsd
already had some pretty powerful crypto device-level support, no?

gzipfs will "just work" once the SCA code is ported to freebsd.

Unionfs is another story.  We've developed and released a unioning f/s for
Linux, and in the 6+ months since it's been released, we've had many
downloads and users -- and consequenetly many bugs reported and mostly
fixed.  We've found out that doing namespace unification properly (i.e.,
unix semeantics compliant) is much harder than it initially seems.  We've
written a detailed tech report on it.  If anyone is going to tackle fixing
the freebsd unionfs, you should read our tech-report first.

Cheers,
Erez.