>Michael Graff writes:
>> Why should we _allow_ our IP output code to send bad packets?  IMHO,
>> trapping errors is much more important than the (somewhat specific)
>> use of sending bad packets to test stacks.

>Agreed.

I *dis*agree.  I wonder if you've read the 4bsd manual Go read the 

>As you note, there is always BPF for real games.

I think describing formerly-legitimate use of raw sockets (modulo
kernel bugs) as "games: begs the question of whether it's a worthwhile
thing to support. 

Opinion seems to be divided here.  There are those of us who are
already use raw sockets and don't want them taken away.  Then there
are those who don't seem to think it's legitimate to be doing such
things at all.

The original intent of the raw-socket interface was to provide as
close an interface as possible to the in-kernel environment. That
includes an interface close to the bottom of the protocol/interface
(i.e., protocol/ifnet) division.  That's different, in subtle but
quite important ways, from the BPF interface.   Those who want to
develop (and debug!) networking stacks using packetfilters already
have the option of using the x-kernel.

I would much, much, rather see the mbuf-chain bugs fixed, and continue
to make BSD raw  sockets available to those who know how to use them.