> This is something Julian, who has configured ea4rct, has pointed out, and
> he suggested that maybe this is a bug in the linux kernel routing algorithm.
> This sounds worrying, and a glaring mistake like this should have been found
> and corrected a long time ago. That or something else is happening.
It's unlikely to have anything to do with the kernel routing implementation.
All of the table manipulation occurs from user space.
I've noticed other related sorts of issues with rspfd use on my system.
I think Craig just needs to take a look at the policy he has adopted for
determining when to replace an existing route.
> I didn't think it was necessary to specifically name a route to the hosts on
> a network interface, as this is implicit in the device's configuration
> parameters, but maybe this is something which must be done if dynamic routing
> is used.
In the case of rspf, yes.
> I guess I'll have to wait and see what Craig says, and check whether the
> problem is "his", ours (bad configuration), or a wider kernel problem (which
> I would think is unlikely).
I'd guess it's a combination of the first two.
1> The other rspfd shouldn't be advertising the network route in the first
place. If he is a linux machine check whether he has a nodegroup entry
for that interface, he probably shouldn't.
2> your rspfd probably shouldn't be replacing the existing one with the
new one.
regards
Terry