Re: ICMP responses to broadcasts

Steve Deering (deering@su-pescadero.ARPA)
25 Apr 1986 17:03-PST

John Romkey writes:

    My opinion is that internally there should be some additional state
    associated with incoming packets marking them as a broadcast (the
    hardware driver can usually figure this unambiguously), and that any
    code that cares whether or not a packet is broadcast should inspect a
    flag rather than having to compare the addresses.

But the decision to withhold ICMP error reports should be based on
whether or not the packet was broadcast or multicast AT THE IP LEVEL,
NOT the local network level. If a packet destined to a unicast IP
address happens to be delivered as a local net broadcast or multicast,
it should still be treated as an IP unicast for the purposes of ICMP.
Conversely, an IP broadcast or multicast may be transmitted as a sequence
of local net unicasts; it should still be treated as a broadcast by ICMP.
I really think you should inspect the IP address to make such decisions.
I understand that it can be messy checking for all the different flavours
of IP broadcast address (at least IP multicast addresses can all be
recognized with a single test), but these tests tend to be out of the
performance-critical main path, so I don't think it's such a big deal.

                                                        -- Steve Deering

This archive was generated by hypermail 2.0b3 on Thu Mar 09 2000 - 14:36:06 GMT