In message <8846.989400784@itojun.org>itojun@iijlab.net writes
>	i guess i need a flag bit onto a routing table, for PMTUD blackhole
>	detection.  the problem i have is expressed in the following comment
>	(sys/netinet/ip_icmp.c):
>	/*
>	 * XXX:   RTV_MTU is overloaded, since the admin can set it
>	 *        to turn off PMTU for a route, and the kernel can
>	 *        set it to indicate a serious problem with PMTU
>	 *        on a route.  We should be using a separate flag
>	 *        for the kernel to indicate this.
>	 */
>
>	not sure if PMTUD blackhole detection is a generic sys/net problem,
>	or a problem just for sys/netinet.  anyway we have RTV_MTU behavior
>	already in sys/net so I guess it is okay to add another bit for
>	PMTUD blackhole detection.

I'm not sure what other protocols could benefit from PMTUD BHD right
now, but it certainly seems like a general, rather than IP-specific
problem.  Putting new flags into the route metrics makes alot of sense 
to me (but then, I wrote the comment above).  

At some point, I considered having several flags -- a sort of,
"We think we've detected a black hole, and we're trying various
sized packets to see what gets through," and a "We've successfully
got packets of this size through".  That seems like overkill, since
we can just turn off PMTUD on the route altogether, I guess.

Kevin
kml@dotrocket.com