[ moved from source-changes to tech-net.  --thorpej ]

On Thu, Jul 06, 2000 at 07:01:47AM +0900, itojun@iijlab.net wrote:

 > 	... or ip_mroute.c should be converted to use encap_attach{,_func}()?

That would be another way, yes.  However...

The vifs are only used by mrouted(8) to create v4-in-v4 tunnels per the
`tunnel' keyword in mrouted.conf.  We could conceivably rip the `tunnel'
support out of mrouted(8) completely, use `gif' interfaces exclusively to
configure the tunnels, which should work using mrouted.conf's `phyint'
keyword.

...this would certainly clean up some code that is in desperate need
of cleaning up.

 > 	the current code (encap4_input) has last-resort fallback case at the
 > 	end, so there should be no problem.

Right, the issue is when something else (e.g. `gif') adds a tunnel that
conflicts with an MRT tunnel or vice-versa ... if two conflicting `gif's
are configured, and error is returned, but the check is not done if a
conflict is caused by MRT, so the MRT path silently fails.

(Although it may `work' by sheer coincidence, now that I think about it :-)

-- 
        -- Jason R. Thorpe <thorpej@zembu.com>