>	* When you mmap() a file, you can wind up sharing pages with a
>	  previous map which is out of date with the actual file.  It
>	  should not be necessary for every program to msync() after
>	  every mmap() call just to get a current view of the file.

It turns out that the part of my patch intended to fix this problem
(adding a vm_map_clean() to vm_mmap() for normal files) causes some
programs to hang when run twice at once (e.g. "echo foo | awk '{print
$1}' | awk '{print $1}'").  I'm not sure why.  I'll check in the
msync() fix, but that won't fix the vi problem unless we add msync()
calls after the mmap() calls in vi.