Allow testing multiple GoBGP instances on localhost.

[wenovus]

Currently GoBGP does not accept UPDATE messages with nexthops pointing to a loopback address. This disallows multiple GoBGP instances from running at the same time on 127.0.0.0/8.

This PR proposes removing this constraint when the RouterID of the current GoBGP instance itself resides within the testing subnet of 127.0.0.0/8.

[fujita]

nexthops pointing to a loopback address

routerId?

[wenovus]

nexthops pointing to a loopback address

routerId?

The code just checks loopback for now:

gobgp/pkg/packet/bgp/validate.go

Line 172 in 8bfcc66

if p.Value.IsLoopback() || isZero(p.Value) || isClassDorE(p.Value) {

Are you suggesting simply to change the check for p.Value.IsLoopback() to checking whether it equals the RouterID? So essentially a less-restrictive check?

Or do you mean I have a bug here that I need to check whether next-hop == RouterID as well, even in this test environment?

[fujita]

What assigned mean?

[wenovus]

"assigned" means that the GoBGP server is expected set the next-hop attribute to one of these addresses, which generally should not happen.

The use case I have is I'm starting multiple GoBGP servers on the same host on different loopback addresses for lightweight testing. In order for this to work, I need the UPDATE messages to not be dropped when the next-hop is set to 127.0.0.0/8. I set this up by configuring Neighbor.Transport.Config.LocalAddress to the specific IP address within 127.0.0.0/8 for each test GoBGP instance.

In order to distinguish this test scenario from production scenarios, I thought the right way would be to check whether the RouterID falls within 127/8.

I wasn't able to figure out how GoBGP exactly sets the nexthop attribute, which should never be 127/8 in production use cases. Do you think there is a better way to check for this to make GoBGP more testable?

[fujita]

You can set RouterID to anything so I don't think that it's the right way to check if nexthops points to a loopback address or not.
Why not using fsm.peerInfo.Address or fsm.peerInfo.LocalAddress?

The following code updates the next hop attribute:

gobgp/internal/pkg/table/path.go

Line 197 in 146b2b8

func UpdatePathAttrs(logger log.Logger, global *oc.Global, peer *oc.Neighbor, info *PeerInfo, original *Path) *Path {

[wenovus]

Done. Check both Address and LocalAddress.

[fujita]

I think that Checking Address or LocalAddress is enough but I pushed anyway. Thanks!