In message <200110171757.f9HHvAmZ029543@destroy.selresearch.net>
Kevin Lahey writes:

>It sure would be nice to have hardware that supported multiple MAC
>addresses.

[... multiple MAC addresses ... ]

As Jason noted, "add" semantics is different from "set" semantics.
Ethernet hardware since DIX days has supported "set mac address"
semantics; it goes back to the disagreement over whether (Ethernet)
addresses are NIC addresses or host addresses.  Protocols like DECnet
phase IV (and, um, Chaosnet?) require changing the active MAC address.
(802.11 is harder to change, I forget the details.)

Prior art for changing ether MAC addresses goes as far back as Ultrix.
ISTR that API was like Jason's suggestion.



>complained bitterly about that, my NetBSD pals laughed and suggested 
>that most of the NICs were like that.

Contemporary hardware with CAMs for multicast filtering (or for
802.3ad?)  does support receiving unicast to one of several
addresses. For that, an add/subtract API (as for multicast addresess)
is what I'd want.

And is anyone looking/working on 802.3ad for NetBSD?


>I agree, it'd be fun to have in there, although, IIRC, there were
>some complexities in trying to do things like assign addresses to 
>particular MAC addresses.  I wound up creating multiple virtual
>interfaces to handle this, and that rapidly got ugly.

I suspect ARP entries will get ugly however you attack the problem.