[IPv6]Re: I-D Action: draft-gont-6man-rfc8028-update-00.txt

[Brian E Carpenter <[e-mail]>]

Hi Fernando,
On 19-Dec-24 18:36, Fernando Gont wrote:
> Hi, Brian,
> 
> Thanks for your comments! -- In-line....
> 
> 
> On 14/12/2024 7:57 PM, Brian E Carpenter wrote:
>> Hi,
>>
>> I think that draft-gont-v6ops-multi-ipv6 makes it clear that we do need
>> to update host behaviour in the way that gont-6man-rfc8028-update
>> suggests. I can see quite a few points in the draft that willl need
>> editorial fixes, but the main thing is:
>>
>> "SLAAC hosts SHOULD record which router (as identified by its IPv6 link-
>> local address and the associated zone index) advertised each piece of
>> SLAAC network configuration information."
>>
>> I believe that is necessary.
> 
> Ack!
> 
> 
> 
>> "... if two SLAAC routers advertise the same IPv6 prefix via PIOs, a
>> host will maintain a (Prefix, remaining valid lifetime, remaining
>> preferred lifetime) for each router."
>>
>> Yes, but the draft needs to explain what the host will do when the two
>> advertisements are in conflict.
> 
> Working on the clarification right now. But, TLDR;, once one accepts
> that SLAAC information is per-router information, that essentially means
> that the possible conflict disappears.
> 
> 
> e.g.
> 
> * At time 0, both ROUTER_A and ROUTER_B advertise 2001:db8::/64 (via a
>     PIO) for address configuration.
> 
>     In response, hosts will configure addresses in that prefix, and may
>     send packets sourced from such addresses to either ROUTER_A or
>     ROUTER_B.
> 
> 
> * At time 0 + n  (i.e., later :-) ), ROUTER_A advertises 2001:db8::/64
>     with, say, a valid lifetime of 0. This should only affect the prefix
>     with respect to ROUTER_A.
> 
>     In other wards: there's at least one local router for which
>     2001:db8::/64 is valid, and hence hosts shouldn't remove the
>     corresponding addresses.

Right, "at least one" is the important point there. Next-hop selection
may change when one of the routers drops the prefix, but the addresses
themselves remain valid unless all routers drop the prefix.

Thanks
    Brian

> 
>     However, packets sourced from addresses in the 2001:db8::/64 prefix
>     should now be routed ONLY via ROUTER_B (rather than via ROUTER_A OR
>     ROUTER_B).
> 
>     -- i.e., each router speaks for itself, rather than for all routers
>     in the local subnet.
> 
> 
> Thoughts?
> 
> Thanks,
> Fernando
> 
> 
> 
>