Subject: Re: PMTUD blackhole detection
To: Jun-ichiro itojun Hagino <itojun@iijlab.net>
From: Kevin Lahey <kml@dotrocket.com>
List: tech-net
Date: 05/01/2001 11:11:29
In message <20010501084913.3B8617D4@starfruit.itojun.org>Jun-ichiro itojun Hagi
no writes
>	on PR 12790, I posted a simple patch for TCP PMTUD blackhole detection.
>	if there's no ack for a while, the patch will do the following:
>	- turn off path MTU discovery for IPv4
>	- uses minimum MTU (1280) for IPv6
>	the behavior will remedy path MTU discovery blackhole, if our side
>	is the sender of bigger packet.  it does not solve problems when
>	the other side is sending bigger packet and choking with the lack
>	of icmp too big messages.
>	see RFC2923 section 2.1 for the exact description of the issue.

Cool.  Is there some way that we could set a flag in the route, 
rather than just having a per-connection flag?  That way we only have
to pay the cost of the two timeouts once per black hole, rather
than once per connection...

>	i have no environment to test, so i need your help.

I can try to test it, but I'm not sure how soon I'll have a chance.
I usually set up a test environment with a PPP link or something where
I can easily set the MTU, and then hack the PPP box to not return the 
ICMP too big message.

Kevin
kml@dotrocket.com