Re: [v6ops] IPv6 link-local and URLs @ IETF119

[Brian E Carpenter <[e-mail]>]

On 19-Mar-24 13:25, Karsten Walther wrote:
> Dear v6ops,
> 
> I like that there is a discussion running on the zone name, since this is a constant pain for us in the last years with our IoT devices (virtual ones aka containers as well as physical IoT devices.
> 
> So my take on this:
> 
> For me a zone name shall never be part of an URL, because it is an information, which is valid on a certain machine only, while an URL should be usable by any machine in a network. In my understanding it defines a location and not how a certain program can reach it.

That is in fact the model in RFC6874, which no browser supports, by saying that the zone ID should be removed before the URL goes out on the wire.

> However, I can understand that a certain user may wants to select a local interface to use e.g. for troubleshooting, thus I would leave it to the application how to specify (command line option, UI setting) the interface to use.

When the application is a browser, the browser community has explicitly refused to do this, either as defined in RFC6874 or in draft-ietf-6man-rfc6874bis.

If the application is ping, that works today. For apps in general, that is exactly what draft-carpenter-6man-zone-ui describes (and what I implemented as a proof of concept at https://github.com/becarpenter/misc/tree/main/zelect ).

> 
> According the use in API programmin like restconf, I discourage the use of zone name, since such configuration is meant to be used somewhere in the network. 

Yes, but my understanding is that sometimes interfaces need to be specified for remote management, so they are defined in relevant YANG modules (using a numeric index, not a name).

> Again in APIs no URLs which are valid only on a certain machine should be used. I think it is possible today start applications in virtual environments, which have the specific network interface as default interface. 

The default zone, as described in RFC4007, is optional and not supported on Linux.

> This would only require, that virtualization host don’t act as routers.
> 
> To have a scoping namespace, which is unambigious across machines would require some sort of registry, which of course contradicts to the .local approach, which means “all next to me”.

More precisely, it means "all next to me on a particular interface". mDNS is quite explicit about that.

Regards
     Brian

> 
> Regards
> 
> Karsten
> 
> *Von: *Toerless Eckert <[e-mail]>
> *Datum: *Dienstag, 19. März 2024 um 09:27
> *An: *[e-mail] <[e-mail]>
> *Cc: *[e-mail] <[e-mail]>, [e-mail] <[e-mail]>, [e-mail] <[e-mail]>
> *Betreff: *IPv6 link-local and URLs @ IETF119
> 
> Dear v6ops
> 
> You may want to think going to http(bis) WG this week for the slot on
> draft-schinazi-httpbis-link-local-uri-bcp. In it, he argues that rfc6874 should be
> retired/made-historic, because it was never implemented in browsers.
> 
> For those who've been absent to the discussion:
>    rfc6874 says URLs can represent IPv6 link-local addresses as [<ipv6addr>:<zone-name>]
>      and David Drafts lays out why this is difficult for browsers
>    rfc6874bis was held up (indefinitely) by Murray (ART AD) on the prmise that the
>      browser vendors decided to not implement it rfc6874 nor bis.
>    draft-carpenter-6man-zone-ui from Brian Carpenter was receiving various criticisms in 6MAN,
>      leading to David to write his draft. Where he primarily promotes to use .local mDNS
>      instead of IPv6 link local addresses
> 
> My take on this:
> 
> 1. The only poor souls who should ever have to use IPv6 link-local addresses in a browser
>     field are IPv6 Network Opertors (aka: here, this group), when interacting in a browser
>     with a router (e.g.:web interface off a browser) and entering URLs. Everybody else
>     should use names (including .local), so it is certainly a minority use-case, but
>     i would hope an important minority use-case. Without network admins being able to
>     troubleshoot even if/where DNS is not working, we can not provide running IPv6 networks.
> 
> 2. I find Murray's DISCUSS on rfc6874bis not convincing, because scoped IPv6 link-local
>     addresses in URLs are not only needed for browsers, but for any type of API in programming
>     languages that use URI, such as restconf or the like. Besides, i do not see why we as
>     the IETF should constrict what we deem to be necessary by the implementation problems
>     of effectively very few browser cores in the industry, neglecting the broader use
>     of URLs. The argument alone that the IETF should not be able to demand what's needed
>     for an IPv6 network archtiecture because some application land work is hard is just
>     what has continued to slow down adoption of anything IPv6 for now 2 decades.
> 
> 3. That being said, i would love to see Davids draft progress to help eliminate the
>     non-working of .local addresses in Browsers today (aka: create standadrd/demand for
>     mDNS in browsers to work), because they actually do have a good
>     amount of actual really cool IoT use-cases (not v6ops). I just don't want the work to call
>     for retiring rfc6874. I just want it for rfc6874 to become only necessary where no other
>     option helps.
> 
> Cheers
>      Toerless
>