where things are heading

[Zach Brown]

Hi all,

I thought I'd send out a little mail to describe some experiments that
will be breaking the crfs code base in the coming weeks.

First, we're getting crfsd out of the business of implementing the btrfs
disk format.  Instead we're trying having it call into btrfs.ko to read
and write items in the file system.  We get the huge advantage of using
the relatively mature btrfs kernel code at the non-trivial cost of
requiring a linux kernel on the server.

Second, I want to change the crfs kernel module to not maintain an
explicit cache of items in pages.  Instead it'll manipulate vfs caches
directly.  This will remove a ton of code from the kernel client but it
requires manipulating the various caches in ways that the vfs
historically doesn't quite anticipate.  I think the end result will be
worth it, but getting there will probably be reasonably frustrating.

I'll be working towards these goals in a series of commits.  To keep the
commits small and understandable, though, lots of the file system will
be left broken at each commit.  The general pattern will be adding new
infrastructure, migrating paths to use it, then removing the old unused
infrastructure.

So, that's where I'm headed.  We'll see how it goes :)

- z