On Fri, Jul 25, 2008 at 04:15:03PM +0300, Jussi Kivilinna wrote:

Quoting "Jarek Poplawski" [off-list ref]:

...

quoted

Is it's really needed? Actually, if it's only for stats, I really
can't see the reason why not to get this info old way - before
enqueing? (And this could be not enough if we want packets count
like in htb.)

qdisc_enqueue recalculates packet size if inner qdisc has size table  
set, so qdisc_pkt_len might return different length after qdisc_enqueue 
than what it was before.

The question is how much does it matter for statistics. Packets' size
(and number) can change many times in many places. Does it mean all
these places should count the same or "real" size? But, if it's really
necessary here, IMHO, it should be done "clean" way, like additional
parameter(s) to qdisc_enqueue(), because in cases like NET_XMIT_CN we
don't even know what happened with an skb.

Jarek P.