Re: pkt_sched: gen_estimator: more fuel for Jarek and Changli

[Eric Dumazet <eric.dumazet@gmail.com>]

Le mercredi 09 juin 2010 à 06:51 +0000, Jarek Poplawski a écrit :
> On Wed, Jun 09, 2010 at 08:13:17AM +0200, Eric Dumazet wrote:
> > 
> > With un-modified kernel, I ran following scripts on my machine
> 
> Why not modified with your other patch quite obviously needed by
> rateest?:
> 

First patch is obvious, but I cooked same script to trigger both bugs,
because you needed some evidences.

> > [PATCH net-2.6] pkt_sched: gen_estimator: add a new lock
> > 
> > gen_kill_estimator() / gen_new_estimator() is not always called with
> > RTNL held.
> 
> Btw, I agree with Changli that adding RTNL to rateest (if possible),
> and doing the RTNL replacement later, seems more proper.
> 

I wont be the guy adding RTNL to netfilter, thats for sure. That would
be a step backward. 
Sometimes, the 'obvious' fix is also a dumb one.
Do you really think I didnt had this idea too ?

xt_RATEEST is an unfortunate domain intersection (netfilter / sched).

We can solve this using a fine grained lock, instead of interesting
lockdep issues, yet to be discovered.

You can submit your patch, but I wont Ack it, I'll Nack it for all these
reasons.

Why dont we remove all locks we have 'because we can use RTNL and be
with it' ?

qdisc_mod_lock could be removed quite instantly, qdisc_base could be
protected by RTNL... And so on...