Linux QOS ipv4 affects kernel using zebra for ipv6

[Xavier Sudre]

Hi everybody,

I sent 3 days to find the trouble I had with dynamic routing tool zebra
and I have isolate a problem into the netfilter code of the kernel or
the iptables tool.

In faact I have QOS on my linux box running Debian with 2.4.18 usagi
kernel. The QOS is applied to ipv4 stack using iproute tool routing with
the help of packet marquing.
To mark packets I use iptables and the mangle table like this:

iptables -t mangle -A OUTPUT -m length --length 0:500 -j MARK --set-mark
3

iptables -t mangle -A OUTPUT -m length --length 500:1500 -j MARK
--set-mark 4

I use tc tool to create cues and filter packet marked by iptables.

I have isolated this two lines of iptables marking process and they are
doing a very starnge stuff on my dynamic routing process in ipv6.
I use zebra bgp facilities for ipv6 and using it it does not
redistribute routes to the kernel when marking process is used.
In this case I have the following error message from the kernel:

kernel: route_me_harder: ip_route_output_key(dst=213.91.4.3, src=80.65.229.50,
oif=7, tos=0x0, fwmark=0x0) error -19

I searched for a solution and switched form kernel 2.4.18 to 2.4.18 usagi, and 
I have also tried the 2.4.19-pre8 kernel but I get always the same eror
message. 

I am using the latest iptables 1.2.6a, and there is no bug known on
their web-site for that problem.

Do you have any solution or any experience with this problem ?

I really hope that you would be able to help me because QOS is a big part
of my ipv4 routing and I dont want to slow down in ipv6 because of
ipv4...

Thanks,
-- 
Xavier Sudre
URL: <http://www.sudre.fr>
EMAIL: <xavier@sudre.fr>