
  $Id: TODO,v 1.155 2004/05/04 11:37:26 lha Exp $


	Release 0.36
		Redhat kernel + syscall

	Release 0.37:
		Darwin 10.2
			ubc
			find fd/vnode leak (fhopen)
		OSF/1 5.1a
			ubc
		FreeBSD 4.5
			ubc ?
		OpenBSD 3.1 and NetBSD 1.5
			dubble-buffer (X ?)
		Linux 2.4
			ubc
			(sparc64)syscall table 64 vs 32 bits syscalls
		Solaris 9
			ubc

TODO-list, in no particular order.

Please mail arla-drinkers@stacken.kth.se if you want to work on
any of these things so we don't do double work.

Store fcache entries on a list on the appropriate volcache entry to
make volume callbacks faster.

split cred for !bsd to rd_cred and wr_cred 

fix weirdness with fake mount points
  installattr (token w/o data) (ls -l ; ls foo ; ls bar)

write2: bsd

fix usage of token between nnpfs and arla

fix usage of open-token between nnpfs and arla

fs flushall:
-volume ?
is this really good ?

nnpfs/linux: 
check that we DTRT wrt Documentation/filesystem/Locking.txt

nnpfs/*: teach nnpfs to read AFS directories for readdir

   merge in tol's code from winnt

nnpfs/linux: check if we can hook into vm_operations/address_space to
figure out when cache pages are dirty and need to flushed.

   dubble buffer instead, look how its done on netbsd nnpfs

*/Makefile.in: allow creation of shared libraries

*/Makefile.in: homogenize

arlad/nnpfs: implement advisory locking

arlad: more improvements on writing in disconnected mode

arlad: fallback to disconnected mode if networking is not working (possibly per cell)

arlad: more consistency checks before recovering state

arlad: remove unused stuff from FCacheEntry

nnpfs/aix: figure out where the PAG should be stored

nnpfs/irix: fix syscall and implement setgroups wrapper

nnpfs/irix: port to Irix 6.5

nnpfs/solaris: verify multithreadness, merge xnode and vnode mutex locks

nnpfs/sunos: implement setgroups wrapper

OPTIMIZATIONS

This is a list of posible optimizations that could be done to different parts
of arla.

 - Delay reads and writes to cache and AFS-servers. Make the more operations
   in one batch. (possible with rx?) Especially directories don't have to be
   synced for every operation.

 - send more messages to/from arlad in one sweep. (lookup on whole path in one
   message, getattr+getdata, etc..)

MILKO

Filbunke, the nnp-class-server generation:
- callbacks
- logging
- fsck/salvage, stability, more tests
- SPEED, real i*op
- volser (?)
- pts,vldb
- security
