> This required some hacking in mrouted and the kernel in order to
> decouple the notion of "IP address" from "interface identifier".

I'll see if this change might be ok, though..

> I did this in two stages:
> 
> 	- Rip out support for mrouted-created v4-in-v4 tunnels.
> 	  They are wire-compatible with gif(4), which is more
> 	  generic, so we should just use the generic mechanism.

I for one would say ok.

> 	- Changed the MRT_ADD_VIF socket option to pass an
> 	  interface name rather than an IP address.  I left
> 	  the old version in there as OMRT_ADD_VIF so that
> 	  old mrouted(8) versions would continue to work, so
> 	  long as they did not use VIFF_TUNNEL (which could
> 	  not work, since I ripped out all the kernel support
> 	  for it :-).

I agree that we don't use VIFF_TUNNEL as an interface (vif) flag
anymore. However, VIFF_TUNNEL and VIFF_SRCRT aren't needed to be
changed to OVIFF_TUNNEL and OVIFF_SRCRT.
Because daemon codes, mrouted, pimd etc. have these flags and adding
"O" doesn't have any effectiveness, these can be still kept in
ip_mroute.h without change, I think.
(As for pimd, there are entries of VIFF_TUNNEL and VIFF_SRCRT, but
they have no meanings anymore. Even so, retaining a portability
between other OSes must be cared...)
(pimd newly defines VIFF_REGISTER for a pseudo interface to handle PIM 
Register messages. just fyi.:)

Regards.
--
Hitoshi Asaeda