arp/nd: Cope with late calls to iflladdr_event

[kp]

Summary

When tearing down vnet jails we can move an if_bridge out (as
part of the normal vnet_if_return()). This can, when it's clearing out
its list of member interfaces, change its link layer address.
That sends an iflladdr_event, but at that point we've already freed the
AF_INET/AF_INET6 if_afdata pointers.

In other words: when the iflladdr_event callbacks fire we can't assume
that ifp->if_afdata[AF_INET] will be set.

MFC after: 1 week
Sponsored by: Orange Business Services

[donner]

Would it not better to not setting the callback on shutdown?

[kp]

In D28860#646170, @donner wrote:

Would it not better to not setting the callback on shutdown?

It'd require a reordering of stack setup/teardown, and that'd be a far more impactful and risky change.
We currently NULL out if_afdata[] in if_detach_internal(), which is called by if_vmove() before the if_reassign callback which triggers the bridge cleanup that results in the iflladdr_event that triggers the event this fixes.

To remove the callback first we'd have to shut down the arp/nd6 code before we move interfaces out. That feels like the wrong order, and would almost certainly result in other bugs.

[melifaro]

sys/netinet/if_ether.c

1482

Would it be possible to add a comment here describing why do we check this?

[kp]

Add comment