> I'm not sure I understand.  The VIFF_TUNNEL checks were to handle the
> special-case of ip_mroute.c doing the encap.  "Logical" interfaces such
> as gif, gre, etc. can simply be treated as normal network interfaces,
> as can e.g. vlan interfaces.

Yes, you are right.
The reason why I asked you is that if you handle all of available
interfaces without checking these attributes (i.e. vif flags), you
have to change members of vifctl structure, like your change. However,
if you can distinguish each interface by each vif flag which is gotten 
from if_type, IFT_..., then you don't need to change vifctl
structure. (I mean your kernel patch eliminates vifc_lcl_addr and
vifc_rmt_addr from vifctl.)

Following is what I want to say.
Since you are complaining about using local/remote address, the patch
tries to use ifp instead of the address. But when you can see the
interface is a phisical or logical (virtual) after mrouting
initiation, you can skip checking these addresses for logical
interfaces. I feel this is reasonable rather than that.
And newly defined vif flag (e.g. VIFF_LOGICAL) by above policy stands
completely different meaning from VIFF_TUNNEL. 

(Because I've not checked your codes deeply yet, I may miss some
point around this. If so, please correct me.)
--
Hitoshi Asaeda