netlink: provide original ifname in the interface dump.

[melifaro]

Summary

This change provide an alternative to the ifmib "IFDATA_DRIVERNAME" API.

[kp]

sys/netlink/route/iface.c

178

I think that's wrong. Or rather, it only works for most interfaces, not all.

With if_epair you'll end up with 'epair0' for both epair0a and epair0b.
And I think that on if_ovpn interfaces you'll end up with 'openvpn0', even if the interface was originally 'ovpnc0'.

[melifaro]

sys/netlink/route/iface.c

178

Ack, that's a good one. Do you think something based on the cloner-specific callback w/ fallback to this composition would do the trick?

[kp]

sys/netlink/route/iface.c

178

I'm not sure I see how that'd work.

My thinking was that we're just going to have to add a char orig_ifname[IFNAMSIZ] to struct ifnet. It can probably be populated from if_attach() (perhaps with an 'if not already set' check, to cope with if_vmove()).

There's all sorts of fun around interface names and locking and error handling, but if all we do is store the original name and return it to userspace those issues don't matter.

[melifaro]

sys/netlink/route/iface.c

178

I agree that it's the simplest solution, but I have somewhat mixed feelings around that.
If that's a rarely used feature, I'd rather not make struct ifnet even bigger, but try to calculate the original name instead.

Anyway, I guess the composition function needs to be in the if.c and the arguments should permit different implementations. I'll come up with a separate review then, so we can have more targetted discussion :-)

[melifaro]

.