TAKE - More I/O path work

[Steve Lord <lord@sgi.com>]

This changes how O_DIRECT read is implemented, we now
use more xfs specific code again rather than the low
levels of the generic direct I/O path. That code cannot
cope with the XFS realtime subvolume, and this way we
should go faster too. The rest of the change is to support
this, but also means we can keep the 2.4 and 2.5 versions
more in sync.

Date:  Tue May 21 12:46:57 PDT 2002
Workarea:  jen.americas.sgi.com:/src/lord/xfs-linux.2.4

The following file(s) were checked into:
  bonnie.engr.sgi.com:/isms/slinx/2.4.x-xfs


Modid:  2.4.x-xfs:slinx:119746a
linux/fs/xfs/linux/xfs_lrw.c - 1.139
	- Pass the pbtarget into the bmap structure rather than the dev_t

linux/fs/xfs/linux/xfs_iops.c - 1.142
	- Make XFS do more of direct I/O itself, we work with the kiobuf mapped
	  by the generic direct I/O path, but do not go via the bottom of
	  the generic direct code anymore - it cannot cope with a realtime
	  subvolume.

linux/fs/xfs/pagebuf/page_buf_io.c - 1.37
	- The bmap structure now contains a pb_target not a dev_t

linux/fs/xfs/pagebuf/page_buf.h - 1.12
	- make the pb_bmap contain a pb_target