Re: [v6ops] New Version Notification for draft-baker-v6ops-cpe-autoconfigure-00.txt

[Lorenzo Colitti <[e-mail]>]

On Sun, Jun 18, 2017 at 6:21 AM, Fred Baker <[e-mail]>
wrote:

> In discussion with John Brzozowski, he expressed a concern that CPEs don't
> come "out of the box preconfigured for IPv6" as they often do for IPv4. I
> had a recent experience with a new router at my home that made the point in
> my mind; IPv6 was an "advanced configuration", and the difference between
> "auto-configure" and "auto-detect" was a little lost on me.
>
> So I have a question. Do we need to spell this out for CPE vendors? I have
> spliced together a draft on the topic. I'd welcome comments that might
> improve it, especially if I have something egregiously wrong. Second, I'm
> curious whether the working group thinks we need this.


We have already spelled it out in RFC 7084, which contains more detail than
this draft.

I don't think this document is a good idea because it oversimplifies what
an IPv6 router needs to do. Critical things that will result in outages and
broken connectivity are missing, and if an implementer just read this
document and built a product, it would be a broken product. Here are some
examples of things that need to be done in order to provide reliable
connectivity.

   1. The downstream RA MUST report an MTU that's <= the MTU of the WAN
   interface
   2. When the WAN default route is lost, the router MUST send out
   zero-lifetime RAs on the LAN
   3. A multi-subnet router MUST send out RIOs if cross-subnet traffic is
   desired.
   4. When the WAN is renumbered, the changes MUST be reflected in the LAN
   RAs.

Based on very recent personal experience, these things are not only
important, completely non-obvious to developers who have build an IPv4-only
product (even an upmarket one), and also very hard to implement in the
typical "just bolt together dnsmasq and dhcpcd" style of implementation
that many of these CE routers use. The open-source ecosystem just isn't
there yet. Example: dnsmasq can't do #1, #2 and #3 at all.

The best advice that we can give to CE router manufacturers is:

   1. Please implement IPv6.
   2. Follow RFC 7084.
   3. Send your box to UNH IOL (or any IPv6-ready logo test lab) repeatedly
   until it passes.

I'm not even sure that #1 is good advice until more people have done #3 and
the open source ecosystem has caught up - unless the plan is to have the
routers occasionally provide broken connectivity and wallpaper over the
problem with happy eyeballs (aka, kick the can down the road until such
time as we need to take away IPv4).