Re: A 3rd try at a proposal for draft-ietf-6man-rfc4291bis-07

[Lorenzo Colitti <[e-mail]>]

On Tue, Mar 7, 2017 at 1:56 PM, David Farmer <[e-mail]> wrote:

> It is only making explicit that, manual config and DHCPv6 address can be
> associated with subnets of any length, and it's clear that at least some
> implementors think it says that already, and I think your in the minority
> of people that think's it doesn't.
>

If you're talking about the current state of the standards, It doesn't
matter whether I am in the minority or not because the standard is pretty
clear about how things work. The current state is that (for unicast
addresses not starting with 0b000):

   - IIDs are 64 bits in length, because of RFC 4291 §2.5.1
   - Subnet Prefixes are 64 bits in length, because of RFC 4291 §2.5.0
   and 128-64 = 64
   - IIDs are required to be unique in a subnet prefix, because of RFC
   4291 §2.5.1
   - On-link determination is separate from the Subnet Prefix, because of
   RFC 5942 (partly motivated by RFC 4903 §4.2, which the current discussion
   seems to completely ignore)
   - On-link determination is determined by PIOs with L=1, because of RFC
   5942.
   - When L=1 is not specified the on-link prefix length of any given
   address is undefined, and the IAB recommends that implementations treat the
   prefix length as undefined, or treat it as /128.

So, if by "subnet" you mean Subnet Prefix as defined in RFC 4291, then it's
clear that the subnet is 64 bits long. If by "subnet" you mean "on-link
prefix", then a DHCPv6-assigned address is not associated with *any*
on-link prefix length, because on-link prefix lengths are configured
completely separately.

Manual configuration is not specified in the standards so it's not clearly
defined anywhere what the /64 in "2001:db8::1/64" means. We may have
opinions, and observe the behaviour of implementations, but those remain
opinions and observations.