This ticket is tracking the QE verification effort for the solution to the problem described below.

 Problem Description: If there are routers using only IPv6 LLA (common with BGP unnumbered use case) and a router policy is used on a router with nexthop set to the LLA of a different router then the outport is determined by matching subnets of the nexthop address and the LRPs addresses on the router with the policy. In this case, if LLAs are used then all LRPs match the criteria as all LLAs are fe80::/64 on all ports.

https://github.com/ovn-org/ovn/blob/045b14c258c7292d1d51248f243eb201199ee45d/lib/ovn-util.c#L466-L478

 Impact Assessment: This makes routing policy not working as expected and with the reroute action traffic is sent to random ports. The workaround is to use a subnet per each router-router connection.

 Software Versions: All OVN versions

  Issue Type: This is a new issue

 Reproducibility: It can be reproduced consistently.

 Reproduction Steps:

create 3 routers, 1 connecting to the other two and not configuring any IPs on the LRPs. Then use a router policy with "reroute" action and nexthop an LLA of one of the two connected LRPs.

R1 | LRPr1 ---- LRPr2 | R2 | LRPr3 ----- LRPr4 | R3

                                             ^
          match: inport LRPr2 reroute - nexthop: LRPr4 LLA

 Expected Behavior: The outport should be determined based on exact address given in nexthop. This can be very complex with complicated topologies

 Observed Behavior: Routing happens randomly

 Troubleshooting Actions: Looked at the generated logical flows for the router for lr_in_policy stage

 Logs: If you collected logs please provide them (e.g. sos report, /var/log/openvswitch/* , testpmd console)