Re: cls_u32 compile failure in current 2.6.12-rc1+BK tree

[Thomas Graf <tgraf@suug.ch>]

* Thomas Graf <20050328195320.GE3086@postel.suug.ch> 2005-03-28 21:53
> * jamal <1112038128.1119.5.camel@jzny.localdomain> 2005-03-28 14:28
> > On Mon, 2005-03-28 at 10:55, Thomas Graf wrote:
> > > * jamal <1112015725.1089.1045.camel@jzny.localdomain> 2005-03-28 08:15
> > > > On Mon, 2005-03-28 at 07:58, Thomas Graf wrote:
> > 
> > > > But you cant completely do this for u32 tcf_exts_dump_stats.
> > > > in the case of ACTION compiled in but someone using old API.
> > > > The code that Meelis just mention that compiler complained about. 
> > > > There was no problem with u32 and tcf_exts_dump - with that you can hide
> > > > all #ifedfs. 
> > > 
> > > I think we can avoid your fixes. 
> > 
> > You are still missing the question;->
> > 
> > You can do this for every classifier but u32. Or is it me who is missing
> > something?
> 
> I think what you are refering to is the TC_U32_KEY() check, right?
> If you look at the place where tcf_exts_dump is called you can
> see that the same pre condition is given since the call is in a
> else branch of a TC_U32_KEY() == 0 condition.
> 
> I haven't had the time to come up with a patch, too many other
> things to take care of but I think it should work out ok.

Before I go ahead, let's make sure I do it right this time. We do
have the following TLVs that need to filled out:

old policer:
  TCA_STATS2 (gnet_stats_*)
  TCA_STATS (tc_stats)

action:
  TCA_ACT_STATS (gnet_stats_*)

action in compatiblity mode:
  TCA_ACT_STATS (gnet_stats_*)
  TCA_STATS (tc_stats)
  TCA_XSTATS

I see that you moved TCA_ACT_STATS into the action TLV array which
doesn't work with the gnet_stats architecture anymore since the
backward compatiblity TCA_STATS and TCA_XSTATS must be outside of
the action TLV. So we need to split the statistics dumping if we don't
want to change iproute2 and have TCA_ACT_STATS being on the same level
as TCA_OPTIONS of the classifier. I didn't notice this breakage when
you posted the patch, sorry.

Idea:
We dump TCA_ACT_STATS in tcf_exts_dump and handle all the backward
compibility including dumping policer stats in tcf_exts_dump_stats.
This means we have to adapt gnet_stats a bit to allow for this.

Thoughts?