Christoph Hellwig, Rick Wheeler, Trond Mykleburst, Dave Howells,
Badari Pulavarathy, Suparna Bhattacharya


Error handling/recovery
	- Media going away errors not handled well
	- Error recovery difficulties since VFS uses uninterruptibe i_sem
	- interruptible mount implementation is difficult (sigmask in 
	  filesystem ?)
Throttling page cache
	- needs to be implemented at the vm level in page_cache_alloc()
	- hooks for fs to notify congestion ?
		- current approach is very block specific

Use of PG_private for bufferheads
	- XFS uses it as a bitmap if PG_private is not set

Batched fsync (Ric Wheeler)
	- filesystem specific sync instead of fs-level (sync_fs + barrier)
		should flush disk cache
		ext3 doesn't perform a cache flush on fsync
	- Ric's results with batched fsync

Debugging/validation issues
	- Difficult to verify filesystem changes .. esp issues related to
	  disk flushing.

Generic support for preallocation
	- can benefit cachefs
	- cachefs would also like to be returned an error if reading a hole,
	  (instead of a return of zeros), e.g using O_DIRECT with an added flag
	  or having some way to easily identify holes. XFS has an ioctl
	  to return extents ... which is needed by NFS as well (David Shenoy)
	
	a comment about parallel NFS that I do not remember now

dcache
	- change global hash to a per-filesystem hash, allow parallel updates
	- SGI noticed this issue, but it wasn't a priority

repair time/recovery time scaling
	e2fsck time with extents