Re: [homenet] draft-boutier-homenet-source-specific-routing-00

[Ray Hunter <[e-mail]>]

Juliusz Chroboczek wrote:
>>> You're assuming that non-source-specific routers drop source-specific
>>> routes, right?
>> Doesn't really matter, at least for the routes to the gateways.
>
> Hmm...
>
> Consider the following topology:
>
>
>   -- A -- X -- B --
>        \     /
>         - Y -
>
> A is announcing (::/0, alpha), B is announcing (::/0, beta).  X is
> source-specific.  Y is a legacy router.
>
> Suppose now that rather than dropping the source-specific routes, Y is
> casting them into non-specific default routes, and that it has
> selected A as its next hop for the default route.
>
> Now X crashes:
>
>   -- A         B --
>        \     /
>         - Y -
>
> A's routing table is now just (::/0, alpha) and (::/0, ::/0), with the
> latter pointing at Y.  But Y is still pointing its default route at A,
> so you're in trouble.
>
> If legacy routers just drop specific routes -- you get a blackhole
> instead of a routing loop.
>
> -- Juliusz
>
Hardly surprising. Whenever you have multiple routing technologies
active in one network, you're going to get pathological cases.

What would a current standard OSPF implementation do today?
I presume ignore the LSA's for source-specific routing, and cause a
black hole.
That's not so bad IMHO.

What's the fix?

Detect routing loops automatically so Y disables injecting the
synthesised default route when it sees packets bouncing?
If you're going to do that then you've had to update Y's code anyway, so
you may as well implement source-specific routing.

Mandate all Homenet routers MUST support source-specific routing, and
update the version number of the routing protocol?

Advise users to move any legacy routers to the leaves of their network,
or remove them completely?
That's pretty much what's going to happen anyway with legacy non IPv6
speaking routers.

regards,
RayH