2003-03-25 17:51:15

by Trever L. Adams

[permalink] [raw]
Subject: 2.4.20 and 6to4 tunneling

I have found a problem, I am not sure if it is in my configuration or
not. I have no dedicated ipv6 addresses, however, my internal network
and the router are using 6to4 tunneling and uses the appropriate 2002:
addresses. I am doing this to learn about ipv6 and possibly to develop
some software using ipv6.

The problem is that a website I set up on a friend's box in Finland
(which has a 2001 ipv6 address and doesn't use ipv6) loads the text just
fine. However, I never do get the images, and if I do a netstat on the
server (in Finland), the connection shows open, with a sendq of about
the image file sizes.

This sounds like the old mtu problems where path mtu discovery was
broken. However, I am unable to find much information. Is this a bug
in the kernel, in the setup, or just me being silly?

vichu-ip6.innernet.org is the website, remove the -ip6 to get the ipv4
site. Yes, it is a silly little website of me playing around. The
image that definitely shows the problem is right at the top of the page.

I can provide ifconfig output or any other details necessary. I have
tried to do ethereal analysis of the connection. I don't seem to see
anything but the opening syn/syn of the connection carrying the image.
It just seems the other end is broken. The server is running RedHat 8
all errata, kernel 2.4.18-24.8.0. Mine is a largely RedHat Phoebe with
a few packages from the newer rawhide.

Sorry if this is off topic, but I don't seem to be finding any
references to such problems w/ Google, so I thought it might be a kernel
bug, but am not quite positive of that yet.

Trever
--
"In protocol design, perfection has been reached not when there is
nothing left to add, but when there is nothing left to take away." --
RFC1925: The Twelve Networking Truths


2003-03-28 05:56:43

by Trever L. Adams

[permalink] [raw]
Subject: Re: 2.4.20 and 6to4 tunneling

With the help of someone else, it appears that the client acks and the
server fails to respond. I am not seeing this on any other connections
other than this box. I am guessing it is a configuration problem or a
bug in the software/libraries/kernel from RH 8.0 with errata.

Trever
--
First Law of System Requirements: "Anything is possible if you don't
know what you're talking about..." -- Unknown