On Tue, Aug 02, 2005 at 11:20:11PM +0200, Iljitsch van Beijnum wrote:
> On 2-aug-2005, at 15:41, Tony Hain wrote:
> 
> >I choose not to try to get to the mic, but on the debated  
> >requirement point
> >3; why is there a belief that the DHCP relay information is correctly
> >configured if the simpler set of 2 bits is improperly configured?
> 
> I think a case can be made for the notion that most routers and most  
> networks don't implement the bits today, so seeing MO==00 doesn't  
> really authoratively say there is no DHCP service, but just that the  
> routers operate in a default configuration.

This sounds like a good suggestion to me.

> So it would be useful to redefine MO==00 to mean "no opinion" and  
> come up with a new value that indicates authoritative information  
> that DHCPv6 is unavailable and/or shouldn't be used.

I believe they should indicate whether DHCPv6 is available or not
(possibly also stateful/stateless). Clients should then generally not
do DHCP when they are told it isn't available (in particular with the
"no opinion" option, there shouldn't be a need to try). If clients are
told DHCP is available they could still choose not to use DHCP based
on the client configuration, type of client etc.

I don't want the bits to tell clients to do or not do DHCP request,
because on a link there may be some clients that should not do DHCP
at all, while others that need DHCP to function.

> >I agree that point 1 & 3 are in direct conflict and that 1 is the real
> >requirement.
> 
> In the scenario above that wouldn't be the case.
> 
> (Having DHCPv6=Y/unknown/DHCPv6=N also increases the likelihood that  
> OS vendors will be prepared to enable DHCPv6 when MO==00 by default,  
> they may / hopefully will not want to do this if there is no other  
> way to indicate the absence of DHCPv6 servers.)

Without the "no info", vendors may have to set the default to say DHCP
is available. I would rather see "no info".

Stig

> >Hosts should never try to outguess the specific instructions from the
> >router...
> 
> Agree, especially in this case where it leads to continuous extra  
> multicast traffic (don't forget that multicast takes up much more  
> bandwidth than unicast on wifi, a reason why we should think twice  
> about draft-pashby-ipv6-network-discovery-00.txt).
> 
> >This means that even the case of M & O clear without a prefix
> >should not result in a DHCP request. This only invites rouge DHCP  
> >servers as
> >a DOS attack. One valid case for M & O clear without a prefix would  
> >be a
> >router retracting the delegated prefix when the link associated  
> >with that
> >prefix fails. When all such upstream links fail there should be no  
> >prefix
> >unless the network manager has configured a ULA prefix. The lack of  
> >a prefix
> >is not a configuration error,
> 
> I believe it is: the router is still advertising itself as a router,  
> implying it can reach the rest of the world. This is harmful. (I  
> think there is a need to study the permutations of different prefix  
> and RA timeouts and their effects, I'll bring that up in v6ops.)
> 
> --------------------------------------------------------------------
> IETF IPv6 working group mailing list
> [email protected]
> Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
> --------------------------------------------------------------------

--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------

Reply via email to