- make io_getevents use lifo wakeup order (spotted by John Myers)
- various improvements based on feedback from Andrew Morton:
	- don't memset / init_waitqueue_head in brw_kvec_async
	- experiment: get rid of the waitqueue_active removals by
	  changing add_wait_queue_cond
	- fix for brw_kvec_async error case
	- copy_from_user -> EFAULT fix in io_submit
- fix io_getevents: was not blocking (inverted if)
- fixed aio_fput.  really!
- fixed a bug in page cache writes that were fully cached
- added cancel operation for aio_poll
- userland apps include fs.h: move #include <linux/aio.h> into protected area
- don't leak requests on -EFAULT
- fail requests for an ioctx with 0 slots
- properly destroy io context when io_setup -EFAULTs
- fixed bug in generic page cache aio read that resulted in corrupt state
- poison aio contexts on completion
- workaround potential request starvation
- BUG() on negative ioctx usage counts
- clear ioctx_list on exec
- check for cancel != NULL
- enable debugging printk for raw io
- return 0 past eof on files instead of EINVAL
- typo fix in wtd_wait_on_buffer
- merge with 2.5.4-pre-something (from bk tree ~8am ca.EST Feb 7th 2002)