On Thu, 22 Sep 2016, Paul Jakma wrote:
Just the session setup with the test tools the final issue. I've been
adapting Martin's bgptool test code to test BGP collision handling.
Not found the problem so far. Unfortunately, I have other stuff
occupying me until early next week now.
So, the issue here seems to be that:
- the other side of Martin's tests waits for a connection from bgpd, to
know that bgpd is Active. The other side does not send SYN|ACK though,
the bgpd side host is resending its SYNs.
- however somehow bgpd thinks its connection succeeds, so it raises
TCP_connection_open, and the peer goes into a state where
collision-detection will consider it (OpenSent).
- the other side connects to bgpd, this succeeds.
- bgpd closes the inbound connection, cause it thinks its outbound tcp
is open, BGP FSM is in OpenSent, and its own RID is higher - so its
outbound has priority.
The problem presumably lies in bgp_connect_check() getting things wrong
somehow, but how....
If so, this must be a very old issue.
regards,
--
Paul Jakma | p...@jakma.org | @pjakma | Key ID: 0xD86BF79464A2FF6A
Fortune:
A homeowner's reach should exceed his grasp, or what's a weekend for?
_______________________________________________
Quagga-dev mailing list
Quagga-dev@lists.quagga.net
https://lists.quagga.net/mailman/listinfo/quagga-dev