I am backing out this change, it breaks doio, and probably anything else
which fills holes in files. I think we had a similar test condition in
this code path before, and it was removed to fix corruption. There has
got to be another variant of this which we can use.
> With this change streaming writes perform almost
> at the speed of streaming read, as measured
> with lmdd and bonnie even when 1K & 2K chunks
> are used for write.
> Date: Mon Aug 28 16:12:45 PDT 2000
> Workarea: waco.engr.sgi.com:/build1/ananth/slinx-xfs-new
> The following file(s) were checked into:
> Modid: 2.4.0-test1-xfs:slinx:73209a
> linux/fs/pagebuf/page_buf_io.c - 1.24
> - Fix a long-standing problem where write performance
> was sub-standard when < 4K chunk-sizes were used.