Improve the UDP tunneling callback

[bryanv]

Summary

For vxlan, the UDP tunneling callback needs to provide an opaque context and the source address.

Reduce some duplicated code in IPv6 UDP by moving it into udp6_append(). This makes the IPv6 code look more like the existing IPv4 code.

[gnn]

Have you run this code with WITNESS to make sure that holding the lock when running through the tunnel does not lead to LOR?

[bryanv]

In D383#3, @gnn wrote:

Have you run this code with WITNESS to make sure that holding the lock when running through the tunnel does not lead to LOR?

Yes. It is a bit annoying that we hold both the INP_INFO_RUNLOCK() and INP_RUNLOCK() when calling the callback. The code could probably be restructured to drop the INP_RUNLOCK() lock before (vxlan doesn't need the lock, and I don't think SCTP does either). I use netisr_queue_src() so we don't recurse into the stack to process the inner Ethernet frame.

[gnn]

For now leave it as is, but leave a comment in. Better to be safe than fast on the first go round.