Re: [lwip-users] Routing in lwIP.

[Jonathan Larmour]

Alain M. wrote:
> Jonathan Larmour escreveu:
>
> > Alain M. wrote:
> >
> > > Ranjan Sinha escreveu:
> > >
> > > > Hi,
> > > >
> > > > I was wondering about how lwIP decides routes. I have gone through the
> > > > available documentation and source code as well and to me it seems
> > > > that routing is implemented in a very simple way; One interface is
> > > > marked as default and we match the destination addresses with
> > > > interface addresses keeping netmask in mind.
> > > >
> > > > Has anybody implemented more complex routing scenarios? For example,
> > > > assigning metrics and seting up a routing table.
> > > One simple thing that comes to my mind and that can solve most problems
> > > is to select the interface for each connection.
> > >
> > > It would certainly be faster then searching a routing table and as an
> > > optional feature the interface selected could be a pointer to the actual
> > > interface.
> > >
> > > This should work in reduced environments as things don't usualy change
> > > so much and it would still be light wheight as in LWip :)
> >
> >
> > The current ip_route is just a simple linear search anyway, so with only
> > one netif, it would take a trivial amount of time in any case. If you had
> > some sort of route cache, you'd have to be able to know when to 
> > invalidate
> > it, which would add complexity, thus reducing any difference.
>
>
> I don't understand why you would want a routing table with a single 
> netif.

I assumed that that was what you meant in your last paragraph.... I didn't 
understand why you'd want to cache the interface for each connection if 
there's only one netif :-).

If there are several interfaces, I can believe that a concept of a cache 
could work. So the PCBs could have a struct netif *cached_route, used for 
PCBs which have an associated connection (which also applies to UDP PCBs 
in lwIP). The cache is used all the time, but as soon as a netif changes 
(e.g. with netif_set_up, netif_set_down, netif_set_default) then the cache 
in all PCBs gets wiped out, which in most systems won't happen often.

But I think the original pointer is interested in more complicated 
routing, possibly even dynamic routing. I think that would require a fair 
bit of new code in lwIP.

Jifl
--
eCosCentric Limited      http://www.eCosCentric.com/     The eCos experts
Barnwell House, Barnwell Drive, Cambridge, UK.       Tel: +44 1223 245571
Registered in England and Wales: Reg No 4422071.
------["The best things in life aren't things."]------      Opinions==mine