Re: [PATCH -next 2/3] net: sched: remove AT INGRESS/EGRESS

[Florian Westphal <fw@strlen.de>]

Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:
> On Fri, May 15, 2015 at 07:21:15PM +0200, Florian Westphal wrote:
> > So, from ifb point of view it makes no difference, G_TC_FROM+AT_STACK
> > causes skb to be dropped and IFB doesn't care about G_TC_AT() at all.
> 
> yes. your change is technically correct. It's not causing ifb regression,

Thanks.

> but it removes information in a way that will be very hard to add it later.

Are you sure?  Would you mind elaborating a bit?

> > AT_STACK cannot even happen for the G_TC_AT case from looking at the
> > code since dev_queue_xmit forces AT_EGRESS & rx sets AT_INGRESS.
> 
> yes, if we only consider ingress and egress hooks.
> I want to use this stack/ingress/egress indication with socket filters.

Hmm... I'm sorry, I fail to understand where problem is.

> If we make stack==egress, I would need to refactor this code all over again.

If you mean "skb was not forwarded", you could just check for
skb->skb_iif = 0?

If not, what info do you need, and why could we not extend proposed enum
if absolutely required?

> It's not broken today. You're doing this aliasing only two squeeze a bit.

Yep, but its was also to minimize the state machinery down to whats
required.

Sorry Alexei, I'm just trying to find out what exactly is needed,
perhaps if you can clarify/explain I might be able to re-spin this in a
way that will meet your requirements.