>> The real problem is on the other end: hosts which try to do PMTU-D,
>> but are behind boxes (usually misconfigured firewalls) which drop
>> the ICMPs necessary for PMTU-D to function.
> Not only, some PPPoE setup just won't send the ICMP unreachable
> message, because the equipement at which the MTU is lowered doesn't
> appear at the IP level.

You mean things like ATM?  But that won't listen to the DF bit, either,
will it?  So there's no issue (except the loss rate increase provoked
by fragmentation).

Or you mean there are setups which silently drop too-large packets
without any notification to anyone?  I have trouble calling those
anything better than grossly broken, and wouldn't worry about arranging
for NetBSD to support them any more than I would about doing likewise
for setups which, for example, drop packets containing 0xff octets:
it's the responsibility of the IP-over-whatever layer to actually
implement IP's semantics.

/~\ The ASCII				der Mouse
\ / Ribbon Campaign
 X  Against HTML	       mouse@rodents.montreal.qc.ca
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B