C10k/PPPoE/IPv6 -> mess

Jørgen Hovland jorgen at hovland.cx
Fri Aug 4 10:43:10 CEST 2006


Wild guess here..
What l2tp version are you/deutsche telecom using?


-----Original Message-----
From: ipv6-ops-bounces+jorgen=hovland.cx at lists.cluenet.de [mailto:ipv6-ops-bounces+jorgen=hovland.cx at lists.cluenet.de] On Behalf Of Sebastian Abt
Sent: 4. august 2006 10:38
To: ipv6-ops at lists.cluenet.de
Subject: C10k/PPPoE/IPv6 -> mess


we're actually having a somewhat weird scenario with "IPv6 on
DSL"-services over here in Germany.  We started offering v6-only access
services about 1.5 year ago, which went pretty smooth right from the
start.  Now we encounter the problem that some customers can't use this
service any longer with neither having something changed on their side,
nor on our end.  Our DSL customers are brought to us via L2TP, so a
typically setup looks like:

customer <-- PPPoE --> DSLAM <----> LAC <-- L2TP --> ISP

The problem is that Deutsche Telekom has replaced their ERXes with C10k
LACs in some cities in Germany which don't handle encapsulated IPv6
traffic correct.  While the ERXes were deployed, everyhting was fine
(and still those aggregated by ERXes can use IPv6). 

When having a deeper look at the packets we saw that the C10ks put wrong
values in the length-field of the PPPoE header (constantly 0x02,
regardless of the actual payload size) while keeping all other things in
good order.  The packets we are receiving on our LNS are all good.  We
even did some testing with explicitly setting the L-Bit and the
length-field in the L2TP header (which by default is not set on Cisco
routers and even can't be switched on by software it seems), hoping the
C10k would use these information and stop computing wrong values, but
unfortunately that did not help in any way.

This problem has been reproduced with two customers and three ISPs now.

Unfortunately I don't know any details about the C10k's configuration or
software being used and neither got anyone from Deutsche Telekom, nor
from Cisco to pay attention to this bug, so my question is: has anyone
on this list seen similar/exactly the same problems with PPPoE, C10ks
and IPv6? 



More information about the ipv6-ops mailing list