[quagga-dev 5420] Re: [PATCH] RFC 2328, chap 8.1:

Joakim Tjernlund joakim.tjernlund at transmode.se
Mon Jun 2 15:23:26 BST 2008


On Mon, 2008-06-02 at 14:29 +0100, paul at clubi.ie wrote:
> On Mon, 2 Jun 2008, Joakim Tjernlund wrote:
> 
> > Yes, find all links that have the same lowest cost and add these links.
> 
> Ok. There shouldn't be a need for that - it should be done at a 
> higher level (by the SPF algorithm searching through LSAs).
> 
> > ... here is were I got into trouble, how do I find the right oi? Perhaps
> > any oi with matching routerID will do? Or NULL oi should be used
> > instead?
> 
> The oi should correspond to the ifindex in l->link_data, no? Add a 
> "ospf_if_lookup_by_ifindex" function to ospf_interface and use that 
> ;).

hmm, that could work once lsa_link_ptop_set() uses ifIndex I guess.
But this might be an incompatible change and require some sort
of manual config I think.

Any chance one can get away with a NULL oi?

> 
> > As you can see in my patch I tried to use nh->router.s_addr = 0 but 
> > that won't work if I don't have the correct oi, I guess.
> 
> What to use for the nh->router is a good question.. Not certain of 
> the answer to that one. I'd guess searching the W LSAs links for, in 
> order of preference:
> 
> a) The address from a stub network link
> b) Any IP address of the router
> 
> Failing either of those being found (which would be surprising, as an 
> OSPF router must have at least one IP address, even if all interfaces 
> are unnumbered), use the router-ID as a last-resort fallback (it's 
> not supposed to be a routeable IP address, but in practice it often 
> is).

Seems easier to just use a 0.0.0.0 IP for the router, iff I have the
correct oi. Should not that work too?

I was hoping to to get "unnumbered" PtP i/f's to work automatically
iff one had choose the local PtP IP addresses wisely, namely the
routerID. That is, all PtP i/f's would use the routerID as its local
IP address. link_data in the LSAs would then hold the routerID as well.

 Jocke



More information about the Quagga-dev mailing list