Re: virtual wireless interfaces / WDS

>>>>> "mwl" == M Warner Losh <[email protected]> writes:
>>>>> "dy" == David Young <[email protected]> writes:

   mwl> one doesn't create a virtual interface for every host
   mwl> on your wired network, so I'm torn.

I think it is the correct level of complexity.  The status-quo
alternative is that a simple crippled version of bridge(4) is
re-implemented inside the hostap code.  This design decision, whether
it is a decision of 802.11 or a decision of wi(4), causes some
problems, for example that attaching a bridge(4) between a
station-mode radio and an Ethernet doesn't work.  I think it is
cleaner to have a real bridge/switch/thing.

If you get a bunch of quad-port Ethernet cards and make your own
switch, you do it the same way.  Just because Cisco has succumbed to
their own marketing snowjob and added a bunch of weird IOS commands
doesn't mean we also have to confuse switches and hubs.  I think
switches are bad enough by giving us two types of things and BROADCAST
flags that lie, and we should stick with things are either a bridge(4)
or else things where tcpdump work for everyone.  802.11 is this
weirdass third kind of thing.

   mwl> I'd prefer that the staN appear and disappear on their own and
   mwl> that the daemon would just react,

one of the annoying things with dynamically appearing and disappearing
BSD devices is, if sta147 appears, disappears, and reappears, you
don't know if it's the same sta147, it's a different sta147, or you
don't know if it's the same or different.  Sometimes I think this
3-case information is ascertainable by the driver but there is no way
to pass it up.  maybe that is more with mounted disks than network
devices, but if you run into that problem unexpectedly while working
on this it'd be nice if you solved it cleanly for everyone else, too.

   mwl> Does the sta get an IP address, or does the wi?

bridge(4).  so it should work both ways.  I mean, you shouldn't be
_required_ to put all the sta into a bridge.  You should be allowed to
give each one a different IP address and make no bridge among them.

but there also needs to be some way to give an IP address to something
that doesn't vanish.  In Ciscoland i heard sometimes they have routing
daemons listen/originate on an alias of the loopback interface,
because othe rinterfaces are appearing and disappearing so often.  If
you are running a one-IP-address AP you don't want to be bothered with
its address vanishing when there are no stations.

    dy> if the wi(4) belongs to a bridge, it should add all of its sta
    dy> to the bridge?

Does the abstraction support hostap drivers that can create more than
one nwid, setting different parameters, and IP addresses, for each?  

That is a nice thing to do in many situations.  and it's better to do
it with one card not just to save money, because then the vhostaps
might be smart enough not to interfere among themselves by two of them
transmitting or granting voice to two stations in the same interval.

-- 
Le fascisme est la dictature ouverte de la bourgeoisie.
		-- Georg Dimitrov
--
*bsd wireless list, a bawug thing <http://www.bawug.org/>
[un]subscribe: http://lists.bawug.org/mailman/listinfo/bsd-wireless/