Re: [v6ops] out-of-focus: why DHCPv6 breaks Android computers?

Le 30/10/2019 à 05:26, Jen Linkova a écrit :
> On Wed, Oct 30, 2019 at 2:35 PM Rajiv Asati (rajiva)
> <rajiva@cisco.com> wrote:
>> Yes, same prefix can be specified in DHCPv6 server and router RA
>> PIO. If ever a conflict between the IPv6 addresses at a host, then
>> the host could use DAD to resolve the conflict (and recycle the
>> SLAAC derived address). Standard machinery.
> 
> Well, I guess the more tricky case is 'a host got an address from 
> DHCP, DAD is failing because another host already assigned the same
> IP (manually or via SLAAC)'. What's next (besides sending DECLINE
> back to the server)? Split-brain condition in address assignments is
> hard. The more I read this thread the less I see the point of having
> DHCPv6 (for address assignment) when we have SLAAC.
> 
> Oh...I think I'm breaking the promise I made to myself ("never get 
> involved into SLAAC vs DHCP discussion ever') again...;))

I dont want to discuss a vs b either.

Alex

> 
> 
>> -----Original Message----- From: Alexandre Petrescu
>> <alexandre.petrescu@gmail.com> Date: Tuesday, October 29, 2019 at
>> 2:32 PM To: Rajiv Asati <rajiva@cisco.com>, Ted Lemon
>> <mellon@fugue.com> Cc: "v6ops@ietf.org" <v6ops@ietf.org> Subject:
>> Re: [v6ops] out-of-focus: why DHCPv6 breaks Android computers?
>> 
>> 
>> 
>> Le 29/10/2019 à 13:38, Rajiv Asati (rajiva) a écrit :
>>> 
>>> Indeed. CPE router should allow keeping both M and A flags, to
>>> allow simultaneous usage of DHCPv6 and SLAAC by the hosts.
>> 
>> At high level, it makes sense to require to allow simultaneous DHCP
>> and SLAAC usage on a same subnet; a private email suggests the
>> same. With that, Android and Windows would live ok side-by-side on
>> same subnet.
>> 
>> Further, thinking about how to implement the req, one would wonder 
>> whether the prefix in PIO with A flag set in an RA with M set,
>> would be the same as the prefix used by the DHCP Server to form and
>> deliver addresses?
>> 
>> (if yes, I think that is difficult to achieve: (1) difficult to put
>> same prefix in the software implementing RA sending, and in the
>> DHCP server connfig files and (2) difficult to make sure the Server
>> does not form an address for a Client, address that a Host has
>> already formed in the same prefix).
>> 
>> (because of that reason, I think that trying to implement that 
>> requirement would lead to designating a prefix for SLAAC and
>> another prefix for DHCP; that may sound a little bit as a waste).
>> 
>> Alex
>> 
>>> 
>>> About time to update RFC7084 - 
>>> https://tools.ietf.org/html/rfc7084#section-4.3
>>> 
>>> 
>>> 
>>> Cheers, Rajiv
>>> 
>>> 
>>>> On Oct 29, 2019, at 7:22 AM, Ted Lemon <mellon@fugue.com>
>>>> wrote:
>>>> 
>>>> On Oct 29, 2019, at 6:37 AM, Alexandre Petrescu 
>>>> <alexandre.petrescu@gmail.com 
>>>> <mailto:alexandre.petrescu@gmail.com>> wrote:
>>>>> Well no.  After  activating DHCPv6 on CPE the CPE sent three
>>>>> RAs changing the Lifetime and flipping the M(anaged) and 
>>>>> A(utonomous) flags.
>>>>> 
>>>>> Packet dumps available upon request.
>>>> 
>>>> That’s the problem.   It should turn on the managed bit but
>>>> not turn off the autonomous bit.  The two can validly be on at
>>>> the same time.
>>>> 
>>>> Of course, since their meanings are deprecated, it’s not too 
>>>> surprising that implementations get this wrong.
>>>> 
>>>> _______________________________________________ v6ops mailing
>>>> list v6ops@ietf.org
>>>> https://www.ietf.org/mailman/listinfo/v6ops
>> 
>> 
>> _______________________________________________ v6ops mailing list 
>> v6ops@ietf.org https://www.ietf.org/mailman/listinfo/v6ops
> 
> 
> 