I didn't know that LFS supported all of the things requested.
LFS is very fast, and (I gather) recovers very quickly after
a crash.

Someone offered on the lists, a few years ago, a program for
changing the size of a BSD ffs filesystem.  I don't think that
it worked while the filesystem was mounted, but it would allow
you to grow a filesystem, I think, if you had simply concatenated
drives (not interleaved) and were willing to have the filesystem
offline while it is adjusted to the new space.  (I assume that
you could also shrink a filesystem, if you wanted to reclaime an
unused hard drive in a set.)


Re. softdeps: Is it true that softdeps render the filesystem
unreliable?  As I always understood it, it was like so:

 * Old ffs was okay, and not as slow as a fully synch. mount.
   Short of hardware failure, you didn't lose files.  (^&  Even
   if you pull the power plug while writing.  But it was not as
   fast as one might like.

   Unless you did something like mounting fully asynch.  (Or unless
   the filesystem code has a nasty bug.  (^&)

 * GNU/LINUX's ext2fs was fast, but was essentially the equivalent
   in terms of robustness (and speed) of a fully asynch. ffs mount.
   You can lose files if the system goes down or crashes at a bad
   time.

   If you like ext2fs as GNU/LINUX systems use it, you can get the
   same effect with an asynch. ffs mount.

 * softdep does *not* introduct robustness issues, but *does* get
   most of the performance of a fully asynch. mount.

I could certainly have been wrong about softdep, but that's what
I understood.

-- 
  "I probably don't know what I'm talking about."  http://www.olib.org/~rkr/