Re: virtual wireless interfaces / WDS

In message: <[email protected]>
            David Young <[email protected]> writes:
: On Mon, Apr 14, 2003 at 05:00:10PM -0600, M. Warner Losh wrote:
: > In message: <[email protected]>
: >             David Young <[email protected]> writes:
: > : Like a wds(4), you will ordinarily put all the sta(4) into a bridge(4),
: > : which will take the IP number.
: > : 
: > : Is the usefulness of sta(4) clear?
: > 
: > Not entirely.
: 
:   The argument for sta(4) is that link conditions in a BSS are not
:   always alike for every STA, so it is not always desirable for STAs to
:   have configuration and statistics in common. Contrast this with a wire
:   ethernet where, no matter where you attach a station, link conditions
:   are virtually the same.  A single broadcast interface suits the latter
:   in the way that it does not suit the former.
: 
:   Put another way, a wireless interface is in many ways an aggregation
:   of point-to-point links, not a broadcast link.
: 
:   (Perhaps meaningful comparisons can be drawn between wireless and VCs
:   in an ATM network or VLANs in an Ethernet; I am not too familiar with
:   those technologies.)
: 
:   But besides the link conditions consideration, sta(4) is a handy way
:   to expose STAs' status and configuration (WEP keys, media, data rate,
:   powersave state, etc.) to the operator and to daemons.
: 
:   I choose to model wireless links with virtual network interfaces because
:   a network interface is a familiar abstraction to BSD operators, because
:   there are opportunities for code re-use, and because it lets me more
:   effectively use already-existing, unmodified utilities and daemons
:   (ifconfig, routed, ospfd, etc.) without modification in 802.11 networks.

I don't doubt that.  I should have added that most of the other
reasons to create these things which I understand.  However, one
doesn't create a virtual interface for every host on your wired
network, so I'm torn.  In a wired environment you have routes to each
mac address.  802.11 tries to mimic the wired environment, even though
it has some different characteristics than the wired env.  I can see
some obvious advantages.  Enough to want to see more of a fleshed out
proposal :-).

: > How does one put all the sta(4) things into a bridge(4)?  These things
: > come and go on a regular basis.
: 
:   Off the top of my head: a daemon listens to the wi(4) parent for the AP
:   to send the STA an association response or, in an IBSS, for the STA to
:   send a beacon or any other packet. Then it creates a sta(4), adds it
:   to the bridge, and attaches it to the wi(4), in that order. Sometimes
:   the daemon will not create the sta(4) and add it to the bridge before
:   the STA sends a packet. I think that is ok if the wi(4) belongs to
:   the bridge, also.

So you are moving the association into a daemon.  I'm not saying that
this is necessarily bad.  I'm just trying to understand what you are
proposing.  I'd prefer that the staN appear and disappear on their own
and that the daemon would just react, just like a routing daemon
reacts to interfaces appearing and disappearing.  I'd rather not have
the daemon be the agent for this creation, even if the daemon is
responsible for associating and disassociating the stations.  However,
these sorts of details can be worked out.  I just didn't see how the
connections between the sta devices and the bridge devices happened.

:   It occurs to me that a sta(4) can and should be exempted from spanning
:   tree computations.

That's likely true.

How does one get a list of associated stations for a given interface?
I have a system right now that has two wi interfaces, wi0 and wi1,
each with a directional antenna.  Let's say I have 2 clients on each
one.  In your proposal, that would mean I'd have sta0, sta1, sta2 and
sta3.  ifconfig would, if I understand you, list four sta's,
sta0..sta3.  How would I know which one belongs to wi0 and which ones
belong to wi1.

I'm assuming that all devices that support AP would use staN
attachments?  There's talk of a hostap mode for the an driver, there's
the amd driver in the pipeline, there's the Atheros driver that's in
the pipe, and there's the awi driver.  All of these support hostap in
some form or another.

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

Warner
--
*bsd wireless list, a bawug thing <http://www.bawug.org/>
[un]subscribe: http://lists.bawug.org/mailman/listinfo/bsd-wireless/