Addendum to kern/2657 ([dM] vnd can disk_unbusy without disk_busy):  I
just now experimented, by making a copy of /netbsd that was as full of
holes as possible, then trying first one and then the other with vnd.
dd read the entire vnd when it was configured to the non-holey file and
crashed as described when configured to the holey file.  So I think a
file with holes is a good test case for making VOP_BMAP fail.

Is there any reason why vnd doesn't fall back on VOP_READ/VOP_WRITE
when VOP_BMAP fails?  Sure, as the comment at the top of the file
indicates, it may interact badly with the buffer cache...but I'd rather
double-buffer some data and have it work than have it error out, or
(worse) crash.  The vnd is one case where I'd much prefer to take a
performance hit if necessary to make it work, rather than erroring out.
Especially since it seems to me that doing that would make it work with
NFS-remote files, which the man page indicates it doesn't at present.

					der Mouse

			    mouse@collatz.mcrcim.mcgill.edu
		    01 EE 31 F6 BB 0C 34 36  00 F3 7C 5A C1 A0 67 1D