Re: [PATCH net-next] tcp: Fix tcp_syn_flood_action() if CONFIG_IPV6=n

From: Geert Uytterhoeven
Date: Fri Nov 18 2022 - 03:30:45 EST


Hi Jamie,

On Fri, Nov 18, 2022 at 2:50 AM Jamie Bainbridge
<jamie.bainbridge@xxxxxxxxx> wrote:
> On Thu, 17 Nov 2022 at 08:15, Jakub Kicinski <kuba@xxxxxxxxxx> wrote:
> > On Thu, 17 Nov 2022 08:39:43 +1100 Jamie Bainbridge wrote:
> > > > if (v6) {
> > > > #ifdef v6
> > > > expensive_call6();
> > > > #endif
> > > > } else {
> > > > expensive_call6();
> > > > }
> > >
> > > These should work, but I expect they cause a comparison which can't be
> > > optimised out at compile time. This is probably why the first style
> > > exists.
> > >
> > > In this SYN flood codepath optimisation doesn't matter because we're
> > > doing ratelimited logging anyway. But if we're breaking with existing
> > > style, then wouldn't the others also have to change to this style? I
> > > haven't reviewed all the other usage to tell if they're in an oft-used
> > > fastpath where such a thing might matter.
> >
> > I think the word style already implies subjectivity.
>
> You are right. Looking further, there are many other ways
> IF_ENABLED(CONFIG_IPV6) is used, including similar to the ways you
> have suggested.
>
> I don't mind Geert's original patch, but if you want a different
> style, I like your suggestion with v4 first:
>
> if (v4) {
> expensive_call4();
> #ifdef v6
> } else {
> expensive_call6();
> #endif
> }

IMHO this is worse, as the #ifdef/#endif is spread across the two branches
of an if-conditional.

Hence this is usually written as:

if (cond1) {
expensive_call1();
}
#ifdef cond2_enabled
else {
expensive_call1();
}
#endif

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds