> > At different points in the code, we use a different chain of pointers
> > to get an ill_t...
> >
> > Are there any occasions when ire->ire_stq->q_ptr (when ire_stq is
> > non-null) is going to be different to ire->ire_ipif->ipif_ill?
>
> Yes. ire_ipif is the cached source address to be used when using this
> ire, which may not be assigned to the same interface as the output
> interface (which is (ill_t *)ire_stq->q_ptr). This is often true when
> using IPMP, or when using vni (or even lo) to host source addresses, or
> with IPv6 when the best choice of source address when doing source
> address selection belongs to an interface other than the output interface
> (if candidate addresses on the output interface are deprecated, for example).
Indeed. The IPMP Rearchitecture work makes this especially common --
e.g., the IRE_CACHE entries have an ire_ipif that refers to the IPMP ill
and an ire_stq that refers to an underlying interface in the group.
BTW, there are a number of bits of code that get this wrong. For
instance, a number of calls to ipif_select_source() pass dst_ill when they
should instead pass ipif->ipif_ill. This is why ipif_select_source() has
to do silly things like:
if (ill->ill_usesrc_ifindex != 0) {
usill = ill_lookup_on_ifindex(ill->ill_usesrc_ifindex,
B_FALSE, NULL, NULL, NULL, NULL, ipst);
if (usill != NULL)
ill = usill; /* Select source from usesrc ILL */
else
return (NULL);
}
The above code would be needless if ipif->ipif_ill was properly passed in.
--
meem
_______________________________________________
networking-discuss mailing list
[email protected]
