Re: [RFC PATCH 1/1 v2] net: sched: Introduce conndscp action

[Cong Wang <xiyou.wangcong@gmail.com>]

On Fri, Mar 22, 2019 at 1:50 PM Kevin 'ldir' Darbyshire-Bryant
<ldir@darbyshire-bryant.me.uk> wrote:
>
>
>
> > On 22 Mar 2019, at 20:05, Cong Wang <xiyou.wangcong@gmail.com> wrote:
> >
> > On Fri, Mar 22, 2019 at 11:26 AM Kevin 'ldir' Darbyshire-Bryant
> > <ldir@darbyshire-bryant.me.uk> wrote:
> >>
> >> Hi Cong,
> >>
> >> Thanks for your questions.
> >>
> >>> On 22 Mar 2019, at 17:39, Cong Wang <xiyou.wangcong@gmail.com> wrote:
> >>>
> >>> Hello,
> >>>
> >>> On Fri, Mar 22, 2019 at 7:09 AM Kevin 'ldir' Darbyshire-Bryant
> >>> <ldir@darbyshire-bryant.me.uk> wrote:
> >>>>
> >>>> Conndscp is a new tc filter action module.  It is designed to copy DSCPs
> >>>> to conntrack marks and the reverse operation of conntrack mark contained
> >>>> DSCPs to the diffserv field of suitable skbs.
> >>>>
> >>>
> >>> Is it possible and feasible to integrate this into connmark?
> >>
> >> I started off coding it that way but quickly ran into my limitations with netlink messaging and became frustrated.  Aside from my own limitations, conndscp ab/uses tcf_qstats requeues & overlimits to indicate DSCP->MARK->DSCP operations and has been useful in proving DSCP/marking operations are occurring in the right times/places.  Integrating with connmark which itself uses overlimits to indicate conntrack mark to skb->mark restoration would lose that differentiation/confirmation/debug ability.  A possibility is to ab/use the drop count instead but I fear that would cause confusion.
> >
> > This sounds problematic, why a flag/parameter doesn't work?
> >
> I don’t understand the question?

You said conndscp uses some stat to save some configuration
information, that is, DSCP->MARK->DSCP operations. But
configuration information is usually saved in a parameter struct
or some priviate flag. So, I have to ask why a flag/parameter doesn't
work for this case?

And, you also implied this is a barrier for you to reuse connmark
action.

Am I misunderstanding anything here?

>
> >
> >>
> >>> Both are intended to retrieve information from conntrack and store
> >>> it into skb. I know the name "connmark" already says it is a mark,
> >>> while yours isn't, I still want to see if we can avoid code duplications.
> >>
> >> I understand your quest :-)  I think conndscp does a bit more than connmark.  Conndscp is two way diffserv<-->conntrack mark operation.  connmark is a single way conntrack mark->skb.mark operation.
> >
> > I am not sure if it is a good idea to modify conntrack in TC,
> > as conntrack doesn't even belong to TC. Retrieving information
> > from conntrack and saving it to skb is fine, as we modify skb
> > in many different ways.
>
> OK, this is why I wanted to ask as RFC before I went too far implementing stuff.  AFAIUI you’re saying it’s tc is okay to restore stuff from a connmark but not to set/change the conntrack mark.  So I need to find a legal place to store a DSCP into a conntrack mark.

Yes.

I guess you should look into netfilter to modify any conntrack attribute,
it is at least where conntrack belongs to. :)

Thanks.