Re: [v6ops] DHCPv6/SLAAC Make Hosts Confusing-//RE: new draft: draft-liu-bonica-v6ops-dhcpv6-slaac-problem

[Mikael Abrahamsson <[e-mail]>]

On Tue, 22 Oct 2013, Mark ZZZ Smith wrote:

> Is this saying that the prefix length of the address on host is /128?

A single address is always a /128. This /128 can be part of an on-link 
other network, or it isn't.

> If that is the case, as the prefix length in doesn't indicate on-link or 
> off-link status (as RA PIOs do), is there any specific reason for the 
> prefix length the DHCPv6 server hands out to be /128 instead of the 
> subnet's prefix length of /64? 

The DHCPv6 server always hands out /128. This /128 can be within a subnet 
advertised in RA, or it can be outside of it. The host doesn't care. When 
a subnet is being advertised in RA you get a network route pointing to the 
interface without an IPv6 address as next-hop.

So it's perfectly achievable today (and it works) to have the following:

Host gets 2001:db8:fff::1/128 from dhcp
host gets on-link 2001:db8:1::/64 from RA and creates a route towards the 
interface for this, but doesn't do SLAAC because A=0.
Host gets default route from router and installs it.

Now, the *router* needs to understand that 2001:db8:fff::1/128 is on-link, 
but no other hosts on the network does not.

So what networks are being announced in RA is completely decoupled from 
addresses handed out by DHCPv6 IA_NA. It's perfectly valid to hand out a 
/128 and then have no on-link prefixes at all, or have some with A=0.

-- 
Mikael Abrahamsson    email: [e-mail]