[IPv6]Re: [v6ops] Re: New draft: "The IPv6 Loopback Address Prefix"

[Maciej Żenczykowski <[e-mail]>]

On Wed, Nov 26, 2025 at 5:19 AM Ole Trøan <[e-mail]> wrote:
>
> Mark,
>
> Since the loopback address is of link-local scope you can already do this with multiple loopback interfaces.
> ::1%loop0
> ::1%loop1

I think in practice implementations make it node local and not scope
local, because they do generate packets (for example icmp errors)
sourced from ::1, and distinguishing the actual interface (besides lo)
would be problematic.

linux-6.16:~$ ping6 ::1%lo
ping6: ::1%lo: Name or service not known

linux-6.16:~$ telnet ::1%lo 22
Server lookup failure:  ::1%lo:22, Name or service not known

linux-6.16:~$ telnet fe80::1%lo 22
Trying fe80::1%lo...
telnet: Unable to connect to remote host: Network is unreachable

So I think this would need both kernel changes (at least Linux afaict
seems to not require scope id for ::1 and not even allow it) and
library changes...

> Cheers,
> Ole
>
> > On 26 Nov 2025, at 11:34, Mark Smith <[e-mail]> wrote:
> >
> > Hi,
> >
> >> On Wed, 26 Nov 2025 at 20:06, Gert Doering <[e-mail]> wrote:
> >>
> >> Hi,
> >>
> >>> On Wed, Nov 26, 2025 at 07:20:17PM +1100, Mark Smith wrote:
> >>> On Wed, 26 Nov 2025 at 18:26, Gert Doering <[e-mail]> wrote:
> >>>> On Wed, Nov 26, 2025 at 11:03:20AM +1100, Mark Smith wrote:
> >>>>> On Wed, 26 Nov 2025 at 10:19, Geoff Huston <[e-mail]> wrote:
> >>>>>> thanks Mark - except that you should  substitute the /32 to a /96 as I got confused between my left and my right!
> >>>>>>
> >>>>>
> >>>>> One of my use cases, which I think I hinted to but didn't specifically
> >>>>> state, was multiple loopback interfaces on a node.
> >>>>>
> >>>>> On Cisco routers you can create multiple loopback interfaces. I've
> >>>>> done that so that I could have say a management loopback interface and
> >>>>> address, a BGP session loopback interface and address and then other
> >>>>> loopback interfaces and addresses for other functions such as an
> >>>>> L2TPv2 end point, RADIUS service client address etc.. Those loopback
> >>>>> interface addresses were all announced into the routing protocol (or
> >>>>> not depending on function).
> >>>>
> >>>> This is a very different thing from what the draft proposes to establish.
> >>>>
> >>>> You are basically deliberately choosing and assigning routable addresses
> >>>> to "non physical" device interfaces, for externally-reachable functions
> >>>> (and this is a very reasonable thing to do, so "all the router people"
> >>>> do it, and "all the anycast service people" do it as well...).
> >>>>
> >>>
> >>> Yes. I mentioned it to demonstrate that I think a larger loopback
> >>> prefix should be big enough to support multiple loopback interfaces
> >>> within a host.
> >>
> >> I find this very hard to understand.
> >>
> >> Why would "there is a larger loopback prefix, used only for machine-internal
> >> communication" have any relevance to "I use multiple loopback interfaces with
> >> manually configured routable IP addresses"?
> >>
> >> Why would "having multiple loopback interfaces out of the well-known
> >> machine-internal loopback address space" have any relevance to anything?
> >>
> >
> > It's in the context of the draft I wrote proposing a larger IPv6
> > prefix. The prefix should be large enough to support subnetting it so
> > that those sub-prefixes can be assigned to multiple loopback
> > interfaces on a host.
> >
> > A ::/96 prefix means having to subnet in an unfamiliar location within
> > the IPv6 address.
> >
> > A /48 for the loopback prefix would allow subnetting at the familiar
> > /48 boundary, resulting in familiar /64s sized subnets that are
> > assigned to different loopback interfaces on the host.
> >
> > Assignment of loopback prefixes to loopback interfaces should be
> > automated. I think that is one of the key values of 127.0.0.1/8 - it's
> > automatically configured. Using at least /48 and then a /64 per
> > loopback interface allows doing something like using the interface's
> > ifIndex as the /64 subnet number (ifindex are typical 16 bits in size,
> > same as the /48 through /64 bits).
> >
> > Loopback interfaces are required to have a link-local prefix per
> > rfc4291, and their prefix length is a /64. The drawback of using a
> > link-local address for testing over a loopback interface is that they
> > require a zone identifier.
> >
> > While a /48 would be fine, since it's better to try to avoid having to
> > grow the loopback space in the future (e.g. which is what has happened
> > with the documentation prefix), a /32 or 1 x 4 billionth of the IPv6
> > address space, for a larger loopback prefix seems worth while.
> >
> > Regards,
> > Mark.
> >
> >> Gert Doering
> >>        -- NetMaster
> >> --
> >> have you enabled IPv6 on something today...?
> >>
> >> SpaceNet AG                      Vorstand: Sebastian v. Bomhard,
> >>                                           Karin Schuler, Sebastian Cler
> >> Joseph-Dollinger-Bogen 14        Aufsichtsratsvors.: A. Grundner-Culemann
> >> D-80807 Muenchen                 HRB: 136055 (AG Muenchen)
> >> Tel: [phone]         USt-IdNr.: DE813185279
> >
> > _______________________________________________
> > v6ops mailing list -- [e-mail]
> > To unsubscribe send an email to [e-mail]
>
> --------------------------------------------------------------------
> IETF IPv6 working group mailing list
> [e-mail]
> List Info: https://mailman3.ietf.org/mailman3/lists/[e-mail]/
> --------------------------------------------------------------------

--
Maciej Żenczykowski, Kernel Networking Developer @ Google