Re: a draft about on-link and submit prefixes

[Lorenzo Colitti <[e-mail]>]

On Tue, Mar 14, 2017 at 8:14 AM, 神明達哉 <[e-mail]> wrote:

> I've submitted a new individual draft on the separation between
> on-link and (SLAAC) subnet prefixes, at the risk of stating the
> obvious: https://datatracker.ietf.org/doc/draft-jinmei-6man-prefix-
> clarify/


Thanks for attempting to clarify the topic. I think it is useful to do so,
since the distinctions here are very subtle and even many in the WG (not to
mention in the IETF or in the industry) are not fully aware of their
implications.

If possible, I would suggest making this document even more general. It's
mostly geared to how implementations treat PIOs, and refers to RFC 4861 and
RFC 4862, but really I think the clarifications it makes have much broader
impact. As I see it, the core issues here are:

   - IPv6 addresses don't specify any on-link information
   - For most (but not all) unicast addresses, subnet prefixes are 64 bits
   per RFC 4291.
   - A given subnet prefix can be spread across multiple links, and a given
   link can support multiple subnet prefixes.

This is not immediately obvious to people who are familiar with the IPv4
subnet model, because it is quite different.

For example, it is an entirely valid configuration for an RA to contain a
PIO for 2001:db8::/64 A=1 L=0 and a PIO for 2001:db8::/96 A=0 L=1. In this
case, the host can form an address via SLAAC that *is not on link*. This is
a a perfectly supported configuration that has no equivalent in IPv4. In
IPv4 the subnet prefix and the onlink prefix have to match because things
wouldn't work otherwise (e.g., the broadcast address would not be on link
any more).

A corollary to the above is that subnet prefixes don't specify any on-link
information. The fact that the subnet prefix is "almost always" 64 bits
long doesn't mean that the on-link prefix cannot be /48, /89 or /120. The
fact that the on-link prefix is /120 doesn't mean that nodes on that link
have to have addresses in that /120.