Use fib4_lookup_nh_ext() in ip_output().

[glebius]

Summary

Existense of PCB route cachine doesn't allow us to use new fast route
lookup KPI in ip_output() like it is already used in ip_forward().

However, when there is no PCB provided we can use it. Typical scenario
is a DNS server that would use sendto(2) over a not connected UDP socket.

[melifaro]

sys/netinet/ip_output.c

306 ↗

(On Diff #55755)

Nitpick: if we're revisiting this fancy condition, would it be possible to consider replace its occurrence with something like

if (RTE_IS_INVALID(ro->ro_rt) || dst->sin_family != AF_INET ...)

?

[glebius]

sys/netinet/ip_output.c

306 ↗

(On Diff #55755)

Do we have RTE_IS_INVALID? But we always first need to check that ro->ro_rt isn't NULL anyway.

[glebius]

More comments & fix a bug with broadcast.

[glebius]

• Copy RTF_HOST to NHF_HOST.
• Export in_ifaddr as nh_ia. This allows to fix ifa packets/bytes accounting

[melifaro]

sys/netinet/in_fib.h

46 ↗

(On Diff #55789)

Would it be possible to move it to a spare field?

sys/netinet/ip_output.c

443 ↗

(On Diff #55789)

Seem to always fallback to the in_ifaddr_broadcast() check.

Originally RTF_BROADCAST flag was added 22 years ago in r15652 to avoid calling in_ifaddr_broadcast(). We had route cloning back then, so it was a fair assumption - if the lookup result was a host route (and it almost always was), then we could trust its broadcast flag, as it was set in the in_addroute() after verificaton.

Currently we don't have route cloning and the only way of having RTF_BROADCAST set on the route is to explicitly add host route to the broadcast address:

route add -host 10.0.0.255 -iface vtnet0

assuming 10.0.0.0/24 belongs to vtnet0

Given that, the check here and in line 400 (if (ro->ro_rt->rt_flags & RTF_HOST)) will aways failback to in_ifaddr_broadcast().

Maybe it's worth just eliminating this condition and always check for in_ifaddr_broadcast().

306 ↗

(On Diff #55755)

No, we don't. I was suggesting adding it as we have these checks repeated at least twice in the ip_output()

[glebius]

sys/netinet/ip_output.c

443 ↗

(On Diff #55789)

But as long as we allow adding such routes manually, we should support them, shouldn't we?

[glebius]

• Remove spare, since nh_ia was added.

[melifaro]

LGTM. The last one: are you planning to implement the same functionality in ip6_output()? :-)

[glebius]

Yes, as a separate change.