Re: [mpls-linux-general] bug: mpls/tc and mpls/netfilter, the same problem ?

[Vincent J. <ja...@en...>]

> I look to the tc code examples, and i'm a doubt. TC expect after the
> flowid a CLASSID
> value. But, CLASSID is of form X:Y where X and Y are unsigned short.
> Moreover, CLASSID represent a cbq where the packet is enqueue, and
> flowid tell tc to put the packet into the CLASSID class if it's
> accepted. Finally, i'm serious doubt about the mpls README.ingress
> example. Look again to the source, let me know that tc_index is setup
> when you specify the keyword "set_tc_index" in the tc command line. So,
> i'm totally confused with the tc behaviour. Have some doc about tc in
> addition to the "2.4 Advance Routing HOWTO" ?

see http://snafu.freedom.org/linux2.2/docs/howto-qos/

2 months ago, I had started to write one from a user point of view. I will 
post it this afternoon or next week. But I do not think that it will answer 
to your question.

> I just finish a new patch for mpls+netfilter (i send it to the mailing
> list this afternoon).

Great, I will not test it for a long time, I am going back home in Bretagne.

> Now we use nfmark+fwmark (aka
> CONFIG_IP_ROUTE_FWMARK) to setup the label or mpls_index+fwmpls (same
> behaviour than fwmark but only for mpls). So, you can put a mark (nfmark
> or mpls_index) to a packet with iptables and then use this mark as a
> filter for tc. My test (both on atm and eth) work for iptables but not
> with tc. Again, i'm a doubt about tc and i don't know (in the packet
> processing) when the ip stack call the sched stuff. Do you know that ?

It is called by the enqueue function of the main interface. And then 
according to a QoS hierachy on this interface (the classifiers), the 
specifique enqueue function is called. Then it is this function that chooses
to enqueue or to drop the packet. see 
http://snafu.freedom.org/linux2.2/docs/howto-qos/node3.html

The dequeuing process is controlled by the main interface too. When the 
interface is ready to send a new packet, it calls its dequeue function that 
looks for the right packet to dequeue in order to provide bandwidth 
limitation and/or delay support.

> I'm afraid that mpls stuff strap the sched stuff.

Everything is always possible ;-)

Vincent