> 
> But why not?  The difficulty I've had lies in trying to be precise
> about exactly why it's not reasonable to expect the kernel to survive
> dd if=/dev/zero of=/dev/mem but it is reasonable to expect it to
> survive, say, experiments with routing sockets.

I think (hard today, I've a stinking cold...) that even root should only
be able to mess up the parts of the system the command is expected to play
with.

So if you mess with routing sockets your routing might get stuffed.
If you mess with memory all bets are off!
However kernel code should do sufficient validation of parameters to
ensure the intergrity of its own data.
The user process can, of course, have its data trashed if it passes an
inappropriate user address.
Anything that assumes that the kernel namelist (read from /netbsd for
example) matches the running kernel needs its head examined.

    David