Re: [dhcwg] DUID+IAID

[Tomek Mrugalski <[e-mail]>]

On 12-03-29 16:41, Ted Lemon wrote:
>> Ted (don't mean to call you out here, but it is prevalent to my
>> question) wrote this on the ISC DHCP list:
>>> DUID+IAID uniquely identifies the interface; DUID uniquely
>>> identifies the host.
>> My question is can this be guaranteed?
> 
> No, I was oversimplifying.   The intent was that there might well be
> more than one IAID per interface.   In practice, I would expect an
> IA_NA and an IA_TA to use different IAIDs, for instance.  Similarly
> for IA_PD.   It was also intended that the client could ask for more
That is not guaranteed. I just checked RFC3315 and it is somewhat
contradictory. Is sending IA_NA and IA_PD legal? Page 11 says that
"Each IA has an IAID, which is chosen to be unique among all IAIDs for
IAs belonging to that client.". This suggest that the same iaid for
IA_NA and IA_PD is not valid. On the other hand, in section 4.2 there
following text "A binding containing information about an IA is indexed
by the tuple <DUID, IA-type, IAID>.". It suggests that IAID number
spaces for different IA types are orthogonal.

Oh well, yet another thing to clarify in 3315bis.

> than one set of IA_NA addresses per interface if it wanted to present
> more than one identity to the network.
Agree. There may be many IA_NA instances (or IA_PD or other IA types) in
a single message.

> Even if what I'd said weren't an oversimplification, it wouldn't work
> to use the MAC address as the IAID, because it's unnecessarily big.
Some implementations use part of MAC. I'm not sure what were the
original reasons for this, but that is mostly valid approach.

> I expect it to be rare that a single DHCP request will require 2^48
> IAs.
I expect it as well. Let's keep iaid as it is now - 32 bit is enough.

And if some brave soul insists on sending 2^48 IAs, he would have to
make sure that his DHCPv6 packets are fragmented properly. Which brings
us to yet another possible 3315bit thing. While I haven't seen any
DHCPv6 messages that are even close to 1280 bytes, it is something that
could possibly be clarified.

Tomek