On Thu, Jun 03, 2004 at 14:38:55 +0900, Jun-ichiro itojun Hagino wrote:

> 	i guess udp pcb is keep pointing to MTU=1500 route (like default)
> 	somehow.  normally udp pcb will get notified from icmp layer using
> 	*ctlinput, and should point to MTU=1280 route.

I think you are right. In my setup sending of large udp packets on bones
is done by bind9. When there is a host route with mtu 1280 and I restart
bind9, packets are sent with max 1280 size.

So, it looks like those sockets that were present before the host route
are not informed by a mtu change.

This is reproducable now:
- delete host route
- start named
- send dns query 
- dns reply is sent with > 1280, no action on icmp6 packet too big
(host route is present)
- send dns query 
- dns reply is sent with > 1280, no action on icmp6 packet too big
- restart named
- send dns query
- dns reply with <= 1280, OK

	rvdp