IPv6 "bloat"
Enno Rey
erey at ernw.de
Sun Mar 20 11:59:13 UTC 2022
Hi,
On Sat, Mar 19, 2022 at 07:11:47PM -0400, Matt Hoppes wrote:
> I misspoke... it's not UUID... It's DUID.
>
> This isn't a backend management issue. This is a protocol issue. The
> MAC of the interface needs to be sent with a DHCP request so that it can
> be properly authenticated to the physical device.
>
> As long as the client and DHCPv6 server are on the same network
> interface -- it all works fine. However, when you relay that
> information, you now lose the MAC address information.
RFC 6939 solves that, since a long time.
See also: https://insinuator.net/2015/02/is-rfc-6939-support-finally-here-checking-the-implementation-of-the-client-link-layer-address-option-in-dhcpv6/
>
> Further, because the MAC is disconnected in IPv6 it becomes more
> difficult to make the connection between IPs on a dual-stack client.
Not sure if I agree with that either. That connection can be made by various other means, see
https://theinternetprotocolblog.wordpress.com/2020/03/14/does-one-need-dhcpv6/
cheers
Enno
>
> Everyone prints the MAC (a unique ID on devices and devices packaging).
> Almost nobody prints the DUID on a device, so how do you pre-populate
> your DHCP server? I can see that it encourages "one interface per
> network" and so encourages bonding, bridging or whatever, but is being
> able to differentiate the interfaces of a host really so bad? I can't
> help but feel that it would have been nice for DHCPv6 to send DUID and MAC.
>
>
>
> On 3/19/22 7:03 PM, Michael Thomas wrote:
> >
> > On 3/19/22 3:56 PM, Matt Hoppes wrote:
> >>
> >>
> >> On 3/19/22 6:50 PM, Michael Thomas wrote:
> >>>
> >>> On 3/19/22 3:47 PM, Matt Hoppes wrote:
> >>>> It has "features" which are at a minimum problematic and at a
> >>>> maximum show stoppers for network operators.
> >>>>
> >>>> IPv6 seems like it was designed to be a private network
> >>>> communication stack, and how an ISP would use and distribute it was
> >>>> a second though.
> >>>
> >>> What might those be? And it doesn't seem to be a show stopper for a
> >>> lot of very large carriers.
> >>
> >> Primarily the ability to end-to-end authenticate end devices. The
> >> primary and largest glaring issue is that DHCPv6 from the client does
> >> not include the MAC address, it includes the (I believe) UUID.
> >>
> >> We have to sniff the packets to figure out the MAC so that we can
> >> authenticate the client and/or assign an IP address to the client
> >> properly.
> >>
> >> It depends how you're managing the network.?? If you're running PPPoE
> >> you can encapsulate in that.???? But PPPoE is very 1990 and has its own
> >> set of problems.?? For those running encapsulated traffic,
> >> authentication to the modem MAC via DHCP that becomes broken.?? And
> >> thus far, I have not seen a solution offered to it.
> >
> > I was honestly more interested in the bloat angle, but this sounds like
> > a backend problem of your own making most likely. But I'm not motivated
> > to see if it's actually the case or just a misunderstanding.
--
Enno Rey
Cell: +49 173 6745902
Twitter: @Enno_Insinuator
More information about the NANOG
mailing list