Re: Better link-local code

On Thu, 16 Dec 2004 09:30:47 -0500, Dan Williams <dcbw redhat com> wrote:
> On Thu, 2004-12-16 at 15:18 +0100, Sebastien Estienne wrote:
> > > Hmm, I have to think about this.  Previously, failure of DHCP would
> > > trigger a switch to a new access point.  In this new code with autoip,
> > > failure assigns a link-local address to the card rather than searching
> > > for a new AP.
> > It could do both: assign a link local address on the wired
> > connection(if there is some link) and also look for a new AP on the
> > wireless card
> > It doesn't hurt to have a link local ip definie on the wired
> > connection, because they can't interferate with other ip that a dhcp
> > could offer on the wireless card.
> >
> > the problem is when no dhcp are found on both wireless and wired
> > connection, we can't assign a link local adress on both i think.
> Yeah, though NetworkManager currently doesn't allow more than one device
> to be "up" at the same time.  Physically with IFF_UP, yes it does (so we
> can link-check and scan) but not in the sense of having an IP and
> routing table entries.  One of the design considerations we made was to
> limit current operation to only one "active" device at a time.
> The next major revision (whenever that will be, hopefully branch and
> start working on it soon) should have this support though.
> In any case, I'm trying to figure out when link-local should be done for
> wireless infrastructure networks.  NM uses failure of DHCP to basically
> tell you that the WEP key is wrong in some cases, since that's really
> the only way to figure that out.  We can't kill that particular behavior
> by making link-local succeed where DHCP would otherwise fail, so I've
> got to think this through a bit.  Perhaps have link-local only occur
> when you've run out of access points to connect to or something?  I'm
> not sure.
> Ad-Hoc is a different story, ATM I don't want to run a DHCP server on
> the main machine, so Ad-Hoc is a separate path that just does link-local
> by default.  That was actually the main reason for me to apply Tom's
> patch :)  Maybe I'll just keep the current behavior for Infrastructure
> connections, and just use link-local for Ad-Hoc networks until we sort
> this thing out.
> Dan

you could use link local, when:
-the user only have one network card (no wireless) and dhcp failed
-the user has wired and wireless ,but only one has link, and dhcp
failed on this one
-the user specificaly choose either wireless or wired and that the
dhcp fail on the choosen interface.

i think that having some kind of diagram to explain what the software will do.

Because maybe somday we'll have to be able to support bluetooth device for:
- ethernet : pand
- ppp : dund

Is it the goal of networkmanager to manage all  kind of network configuration?
Because supporting other wireless technologie like umts/gprs will be a
growing demand in the future.

like in this scenario:
wired at home
gprs or utms on the road to work
wifi (802.11) at work

i talked about bluetooth because some personnal gateway (set top box)
using bluetooth to access broadband are appearing

but maybe all these things are beyond the scope of networkmanager.
Sebastien Estienne

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]