CentOS – Configuring Firewalld to Allow Quagga OSPF Messages

Guacamole pointed me in the right direction. I did a bit of debugging on the ProCurve and in quagga and it emerged that one of them was seeing the area as a stub area, but the other saw it as a normal area, and that's where the communication broke down. Once I had both configured to see the area as the same, it all worked like a charm.

Why would the routing table change at all? R2 already has routes to everything via its links to R1 and R3. Even if R1 loses its route to 172.16.3.0/24, R3 is still advertising it back to R2 (as it did before the link went down).

You should see some change in the OSPF database itself, but the host's routing shouldn't need to change. Try comparing "show ip ospf database" on each of the routers before and after shutting down the link.

Worth noting - Without multipath enabled in your ospf configuration only one path to 172.16.3.0/24 will be injected into your routing table. If it sees routes from both routers then it will choose one of the paths based on a tie breaker - the router-id's. If R2 had already selected the path where the link -didn't- go down then nothing changes...

A couple of points-

1.) Why are you using "redistribute connected" - particularly on an unrestricted basis. The "network x.y.z.q" command already takes care of the links themselves. Redistribute connected is, at best, going to yield local routes showing up as external. Remove this statement and add another network statement for the 192.168.x interfaces (mark these interfaces as passive if they're supposed to be for end-host connectivity). If you must redistribute connected routes then apply filtering to limit the routes that are actually redistributed. Unrestricted redistribution is only pain waiting to happen.

2.) It's been a while since I've played with Quagga, but I think you also need "link-detect" on the ospf interface definitions.