TAKE - properly account clustered IO

[Ananth Ananthanarayanan <ananth@xxxxxxxxxxxxxxxxx>]

This change improves dbench performance for smaller number of clients by 5-10%.

Date:  Mon Mar  5 12:12:26 PST 2001
Workarea:  waco.engr.sgi.com:/build1/ananth/xfs-tot

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


Modid:  2.4.x-xfs:slinx:88916a
linux/fs/buffer.c - 1.58
        - Account for the fact that write_buffer for clustered IO can push out
          more than 1 block at a time. This makes back ground scanning only as
          aggressive as necessary in bdflush.

linux/fs/pagebuf/page_buf_io.c - 1.63
        - Return number of pages clustered for use by write_buffer so
          watermark accounting is done properly. Other misc. cleanups.