Help with DHCPv6 client-identifiers

Rudy Zijlstra rudy at grumpydevil.homelinux.org
Fri Nov 18 13:57:03 UTC 2011


On 11/18/2011 02:43 PM, Alex Bligh wrote:
>
>
> --On 18 November 2011 13:48:33 +0100 Bjørn Mork <bjorn at mork.no> wrote:
>
>> head-in-sand... You may want to consider:
>
> In some deployments yes. In others, nearly everything you have written is
> irrelevant. For instance, We use DHCP to give IP addresses to virtual
> machines. We know the MAC address we give them, as we build the NIC (and
> indeed sometimes filter other MAC addresses out). We know the networking
> topology. We don't know the host OS which determines how the DHCP 
> packet is
> built. This works fine on IPv4. It's completely broken on IPv6 as 
> there is
> no reasonable way to identify the client, so we just dropped DHCP support
> for IPv6, and use auto config instead. We could instead have looked at 
> the
> MAC address the packet actually comes from (i.e. in the ethernet 
> layer) as
> we don't happen to use relays or similar, but really life is too short 
> - I
> would support a patch to dhcpd to do that.
>
> In my view, it was an error not to put the hardware address in the DHCPv6
> packet as an option. The server after all does not have to use it. It use
> any of the multiplicity of other ways of allocating addresses.
>
Which is why cablelabs added IPv6 vendor options to include the MAC 
address... It is pretty much the only way to identify a CM.



More information about the dhcp-users mailing list