Re: [PATCH iproute2-next v4] ip-link: add support for nolocalbypass in vxlan

[Stephen Hemminger <stephen@networkplumber.org>]

On Fri, 19 May 2023 11:11:12 +0200
Andrea Claudi <aclaudi@redhat.com> wrote:

> On Fri, May 19, 2023 at 11:50:03AM +0800, Vladimir Nikishkin wrote:
> > 
> > Stephen Hemminger <stephen@networkplumber.org> writes:
> >   
> > > On Thu, 18 May 2023 21:46:01 +0800
> > > Vladimir Nikishkin <vladimir@nikishkin.pw> wrote:
> > >  
> > >> +	if (tb[IFLA_VXLAN_LOCALBYPASS]) {
> > >> +		__u8 localbypass = rta_getattr_u8(tb[IFLA_VXLAN_LOCALBYPASS]);
> > >> +
> > >> +		print_bool(PRINT_JSON, "localbypass", NULL, localbypass);
> > >> +		if (localbypass) {
> > >> +			print_string(PRINT_FP, NULL, "localbypass ", NULL);
> > >> +		} else {
> > >> +			print_string(PRINT_FP, NULL, "nolocalbypass ", NULL);
> > >> +		}
> > >> +	}  
> > >
> > > You don't have to print anything if nolocalbypass.  Use presence as
> > > a boolean in JSON.
> > >
> > > I.e.
> > > 	if (tb[IFLA_VXLAN_LOCALBYPASS] &&
> > > 	   rta_getattr_u8(tb[IFLA_VXLAN_LOCALBYPASS])) {
> > > 		print_bool(PRINT_ANY, "localbypass", "localbypass", true);
> > > 	}
> > >
> > > That is what other options do.
> > > Follows the best practices for changes to existing programs: your
> > > new feature should look like all the others.  
> > 
> > Sorry, I do not understand. I intended to do exactly that, and I copied
> > and adjusted for the option name the code currently used for the
> > "udpcsum" option. Which is exactly
> > 
> > 		if (is_json_context()) {
> > 			print_bool(PRINT_ANY, "udp_csum", NULL, udp_csum);
> > 		} else {
> > 			if (!udp_csum)
> > 				fputs("no", f);
> > 			fputs("udpcsum ", f);
> > 		}
> > I just replaced that option name with [no]localbypass. Fairly
> > straightforward, prints noudpcsum or udpcsum. Later Andrea C
> > 
> > Then Andrea Claudi suggested that print_bool knows about the json
> > context itself, so the outer check is not needed, so I removed that.
> > But the "model option" I used (really the simplest one), does have
> > output both when set to true, and when set to false. I have neither an
> > opinion on this nor an understanding what is better for scripting. But I
> > do not understand the suggestion "do like the other options do", when
> > seemingly, other options do what I suggest in the first place.
> >  
> 
> If I get Stephen corretly, he is simply suggesting that printing
> "nolocalbypass" is unnecessary. If you find don't find "localbypass" in
> the output, you know it's not enabled.
> 
> Unfortunately iplink_vxlan does not work according to this logic, as you
> are pointing out, but there are several places where this happens, like
> link_gre6.c:543. So you can do that for "localbypass" here.
> 
> Fixing old options, on the other hand, is not easy, as we may end up
> breaking user scripts relying on "no<whatever>" option. I can work on a
> patch for that, but we probably need some kind of deprecation warning to
> users.
> 
> Stephen, what do you think?
> 

Scripts parsing non-json output are fragile. There was never a hard
guarantee that non-json output was stable.


I was looking at the existing vxlan_print_opts() and it already does
this for several options.

static void vxlan_print_opt(struct link_util *lu, FILE *f, struct rtattr *tb[])
{
...

	if (tb[IFLA_VXLAN_COLLECT_METADATA] &&
	    rta_getattr_u8(tb[IFLA_VXLAN_COLLECT_METADATA])) {
		print_bool(PRINT_ANY, "external", "external ", true);
	}

	if (tb[IFLA_VXLAN_VNIFILTER] &&
	    rta_getattr_u8(tb[IFLA_VXLAN_VNIFILTER])) {
		print_bool(PRINT_ANY, "vnifilter", "vnifilter", true);
	}