[Debian] IPv6 trouble with Lenny kernel

Benoit Branciard Benoit.Branciard at univ-paris1.fr
Wed May 27 12:02:40 EDT 2009


Hi all,

I'm running into trouble while trying to get IPv6 running inside VEs 
with Lenny OpenVZ kernel.

Our production HW nodes are running:
- Debian Etch AMD64
- linux-image-2.6.18-14-fza-amd64 028stab056.1dso1 (Debian-Systs) kernel
- vzctl 3.0.23-1dso1~etch0
- IPv6-enabled Etch and Lenny VEs with venet devices

This setup works happily for both IPv4 and IPv6 protocols.

Now I upgraded one test HN to Lenny, and installed 
linux-image-2.6.26-2-openvz-amd64 2.6.26-15lenny2 (Debian-provided) 
kernel, all other things remaining the same (most notably, no VE were 
modified nor upgraded).

This resulted in a sort or ARP-related problem:

- during a short time after HN boot (three minutes or so), VEs have a 
working IPv6: HN, local network and non-local hosts can be successfully 
accessed.

- After this delay, only the HN address can be reached from inside VEs, 
all other IPv6 addresses become unreachable.

This problem does not affect IPv4, and does not affect the IPv6 
capability of the hardware node itself (all IPv6 addresses can be pinged 
from the hardware node).

Rebooting the HN on the linux-image-2.6.18-14-fza-amd64 kernel (while 
maintaining the Lenny binaries) restores total IPv6 functionality, so 
the problem seems kernel-related.

This problem resembles an older bug related to "ip" command parsing in 
/usr/lib/vzctl/scripts/vps-functions (see: 
http://openvz.org/pipermail/users/2008-January/001658.html), but is not 
the same. Running the "vz_get_neighbour_devs" function manually returns 
the correct value here:

. /etc/vz/vz.conf ; . /usr/lib/vzctl/scripts/vps-functions ; vzgetnetdev 
; vz_get_neighbour_devs MY:IPV6:ADD:RESS

displays the expected HN public interface.

Any idea on how to resolve this ?


-- 
Ce message a ete verifie par MailScanner
pour des virus ou des polluriels et rien de
suspect n'a ete trouve.



More information about the Debian mailing list