Re: refactoring ip_output() and the L2 _output()

[Darren Reed <darrenr%netbsd.org@localhost>]

David Young wrote:
> ...
>
> It's funny that we get into ether_output() with an rtentry that's
> not even usable.  I'm not sure how that happens.  I will have to
> look more carefully at what ip_output() is doing.
>
>                         if ((rt0 = rt = rtalloc1(dst, 1)) != NULL) {
>                                 rt->rt_refcnt--;
>                                 if (rt->rt_ifp != ifp)
>                                         return (*rt->rt_ifp->if_output)
>                                                         (ifp, m0, dst, rt);
>                         } else 
>                                 senderr(EHOSTUNREACH);
>                 }
>
> Yikes!  To call rt->rt_ifp's output routine with a different ifp is not
> advisable, is it?  ISTR finding a bug in carp(4) involving a similar
> pattern, ifpX->if_output(ifqY, ...).
...

If both bge0 and bge1 are on the same subnet then a route out
of either can be used to transmit out either.

So if bge0's IP is 1.1.1.1 and bge1 is 1.1.1.2 and I want to
reach 1.1.1.3, a cached route might point to bge0 however the
kernel could just as easily decide to use bge1 (maybe bge0 is
down.)

So in places where you have, for example:
+            return ip_hresolv_output(rt->rt_ifp, m, dst, rt);

you should replace the "rt->rt_ifp" with just "ifp".

For the purposes of research, you may want to add a statistic
that is bumped whenever "rt->rt_ifp != ifp" when ip_output()
is called.

Darren