> Can anyone see a problem with the above proposal?
> Attached is a proposed change to implement this.
> Comments?

It's a nit but it would probably be better to avoid sending back, say,
ICMP port unreachables when the interface is in this state.

You could do this by setting M_BCAST on packets matching the magic
case as an expedient, if exceedingly disgusting, way of achieving the
desired result.

Dennis Ferguson