RIP Metric export to kernel table ?

[Florian Lohoff]

On Tue, Jun 20, 2000 at 01:47:47AM +0200, Martin Mares wrote:
> Hello!
> 
> > This is something i neglect - I have seen bird delete the local ethernet
> > route but continuing to try to add routes to gateways on the net unreachable
> > due to deletion of the route and the kernel refusing to add the route 
> > because of destination unreachable.
> 
>    Unless something got terribly misconfigured, BIRD should never do such a thing
> -- device routes have absolutely the highest preference and nothing should make
> BIRD remove them or replace them by anything else.

But if you dont have a device statement in the config (as i had) there
is no "device" routes in the bird process and it deletes those from the kernel.

> Which system configuration did it happen with? linux-20 or linux-22? (Unless
> you need to run an old kernel, I recommend using the latter one since the old
> BSD-ish interface to the routing tables is somewhat fishy.)  With linux-22
> BIRD surely doesn't even touch any device routes in kernel tables and it lets
> the kernel itself take care of them.  The linux-20 configuration could in theory
> have such problems, because it does a lot of magic things when synchronizing
> the routing tables and it probably isn't tested well as machines running 2.0
> kernels are very rare here.

Kernel 2.2.16

>    Is the problem reproducable? I would be very interested in a test case
> where it deterministically happens.

It is - It was a combination of not having a direct statement in the
config and a device not capable of split-horizon and reannouncing
the "directly" connected ethernet. Bird than took the announcement
and deleted the original direct route and tried to replace it 
with the route to the buggy device which that failed because it
wasnt reachable anymore.

Flo
-- 
Florian Lohoff		flo at rfc822.org		      	+49-waiting-4-telekom
     "If you're not having fun right now, you're wasting your time."