use netlink events to emit lighthouse updates instead of a timer

[JackDoan]

This needs to be

• confined to Linux (afaik Android will kill you if you try to even touch netlink)
• made configurable
• less log spammy
• do we still need to poke the lh every so often to make punchy work?
• debounced

[nbrownus]

The main reason to stay in communication with a lighthouse is to learn about roaming real underlay addresses which might be opaque to the client.

Real underlay address changes are learned by simply receiving a properly crafted/encrypted packet though, so as long as the client sends encrypted packets on some regular cadence, we can squash the host update traffic.

Generally I would rely on test packets for this behavior but we made a change recently to reduce the amount of paperwork a nebula mesh has to deal with. If we are receiving traffic from the lighthouse then the tunnel is declared alive and we will never initiate a test packet to it. The only time we would send a test packet to a lighthouse is if we haven't sent or received a message from them within a 5 second period (by default) so we likely would want to adjust this to send test packets regardless of rx traffic.

[JackDoan]

The main reason to stay in communication with a lighthouse is to learn about roaming real underlay addresses which might be opaque to the client.

oof I forgot about NAT 😢

Real underlay address changes are learned by simply receiving a properly crafted/encrypted packet though, so as long as the client sends encrypted packets on some regular cadence, we can squash the host update traffic.

Generally I would rely on test packets for this behavior but we made a change recently to reduce the amount of paperwork a nebula mesh has to deal with. If we are receiving traffic from the lighthouse then the tunnel is declared alive and we will never initiate a test packet to it. The only time we would send a test packet to a lighthouse is if we haven't sent or received a message from them within a 5 second period (by default) so we likely would want to adjust this to send test packets regardless of rx traffic.

yeah I like this approach