[Olsr-dev] Removing default route after program termination

[John Hay]

On Wed, Oct 10, 2007 at 04:12:41PM +0200, Bernd Petrovitsch wrote:
> On Mit, 2007-10-10 at 15:55 +0200, Hannes Gredler wrote:
> > On Wed, Oct 10, 2007 at 03:31:40PM +0200, John Hay wrote:
> > | On Wed, Oct 10, 2007 at 02:48:27PM +0200, Mitar wrote:
> > | > Hi!
> > | > 
> > | > On 10/10/07, Bernd Petrovitsch <(spam-protected)> wrote:
> > | > > Short: I had (on a stock Linux kernel) no problems with > 1 default
> > | > > route.
> > | > 
> > | > I can second that. I also have good (working) cases of Linux based
> > | > nodes with VPN and Internet uplink working nicely together with OLSR.
> > | > But on Linux you can declare multiple default gateways (with different
> > | > metrics). On Mac OS X this is something nonexistent (if I understood
> > | > things correctly).
> > | 
> > | It is nonexistent on FreeBSD also. I think the thinking is that an app
> > | (like a routing daemon) is much better at deciding when to use what,
> > | than the kernel can.
> > 
> > hmm not really .. that thinking implies that there is only one source
> > (routin daemon) supplying routing information.
> 
> Yes.
> And that is clearly not always the case (and it's even worse IMHO;-) -
> see below).
> 
> > like bernd has described many of us olsrd
> > developers use a openvpn tunnel to some olsr cloud to get a link-state
> > feed in order to test olsrd. here i already have two sources
> > for the default route ... one (static) to my ISP and another one
> > pointing over the openvpn (tap) interface.
> 
> The root of the problem is IMHO:
> First, we have two different "layers" of routing:
> - "pure" IP routing in the tech-only domain. One uses e.g. RIP, OSPF for
> that.
> - administratively limited routing (which - as a consequence - must be
> configured
>   by a human) usualy at the border of an AS: BGPv4 is state of the art
> there.
> [ Technically, BGPv4 as such is not necessary and one could - pure
> technically
> seen - use OSPF there. But usually you have at least to pay there so you
> want/need
> it controlled. ]
> 
> These annoying home users (like /me) are sitting on such an
> administrative border -
> without need for BGPv4 because I have only one IP-address and my cable
> provider
> probably doesn't want to route through my node anyways;-) (and I have
> lots of
> firewalls rules there too). But I have also one of the tech-only domain
> routers
> on exactly the same host.
> Thus there is no single daemon on my node which controls all of the
> routing
> (because "cable client part is quite statically configured - especially
> the
> firewall).
> I probably could install and use Quagga and have such a single route
> controlling instance on my node but up to now it was not necessary (and
> it
> saved time and brain capacity to not learn the next tool).

I guess it all depends where you come from. :-) Maybe your case is
clear cut, but in a more generic "routing" problem with more than
one routing daemon running, you will still need a way to set the
metric/weight a daemon use for its routes... or even different
metrics depending on the route. For instance I might want to say
the default route from daemon xyz needs to have the best metric
while for the routes x.y/16 routes from daemon yyy should be the
most prefered. Or even, I never want to learn the default route
from xxx.

And yes I do come from a background of haveing run gated/zebra/quagga
with multiple protocols simultaneously and using the weight
adjustments filtering capabilities. :-) I would love it if I could
tell olsr not to learn certain routes. :-)

John
-- 
John Hay -- (spam-protected) / (spam-protected)