Julian Assange <proff@suburbia.net> just posted a test suite for
mmap on the freebsd-hackers list that is rather interesting.

The results from -current are:

HAVE_MMAP_FILE_PRIVATE_READ
HAVE_MMAP_FILE_PRIVATE_CHILD_INHERIT
HAVE_MMAP_FILE_PRIVATE_WRITE
HAVE_MMAP_FILE_SHARED_READ
HAVE_MMAP_FILE_SHARED_CHILD_INHERIT
HAVE_MMAP_FILE_SHARED_CHILD_READ_PARENT_WRITE
HAVE_MMAP_FILE_SHARED_PARENT_READ_CHILD_WRITE
HAVE_MMAP_DEV_ZERO_PRIVATE
HAVE_MMAP_DEV_ZERO_PRIVATE_CHILD_INHERIT
HAVE_MMAP_DEV_ZERO_SHARED
HAVE_MMAP_DEV_ZERO_SHARED_CHILD_INHERIT
HAVE_MMAP_DEV_ZERO_SHARED_CHILD_READ_PARENT_WRITE
HAVE_MMAP_DEV_ZERO_SHARED_PARENT_READ_CHILD_WRITE
HAVE_MMAP_ANON_PRIVATE
HAVE_MMAP_ANON_PRIVATE_CHILD_INHERIT
HAVE_MMAP_ANON_SHARED
HAVE_MMAP_ANON_SHARED_CHILD_INHERIT
HAVE_MMAP_ANON_SHARED_CHILD_READ_PARENT_WRITE
HAVE_MMAP_ANON_SHARED_PARENT_READ_CHILD_WRITE

and the difference between those results and the ones from 
FreeBSD 2.2-current is that NetBSD screws up HAVE_MMAP_FILE_SHARED_WRITE,
which is writing to mmaped memory, calling munmap, and reading from
the file that had been mmaped.

I figure it might be handy for people who have had the most
experience with the NetBSD VM flaws to sanity-check Julian Assange's
code, given that it's likely to be part of the GNU autoconf package.

	Sean.