Now I'd like to (re)start a bigger discussion about the M/O flags: how
we should describe the relationship between the M/O flags and the
related protocols.
So far, many people seem to have (somehow) agreed on what Christian
said (attached below): the M/O flags should just be hints of
availability of the corresponding services (or protocols) rather than
a trigger to invoke the protocols under a certain level of requirement.
While some part of his interpretation (i.e., "do DHCP, and only if it
fails do auto-config") was not really accurate according to the
specification as others already pointed out, I still think he made
important points, in that RFC2462 has "supposedly mandatory nature"
with ambiguity on details.
An example of the "mandatory nature" is the following sentence of
Section 5.5.3:
If the value of ManagedFlag changes from FALSE to
TRUE, ..., the host should invoke the stateful
address autoconfiguration protocol, ...
The "ambiguity" includes:
- the level of requirements (e.g., should the above "should" be
actually SHOULD?, etc)
- what is the protocol to be invoked by this statement (we are now
trying to clarify this point)
- what hosts should do if the M/O flags keep being cleared; e.g.,
does this mean the hosts should not invoke the protocols?
- what if the hosts do not implement the stateful protocol? (recall
that in the node requirements document it is basically optional to
implement DHCPv6 for the stateful address configuration)
I interpreted Christian's suggestion as one that tries to remove the
"mandatory nature" considering the current deployment/implementation
status, and in that sense, I tend to agree. The suggestion won't
fully solve the ambiguity issues, but it will make the document
well-balanced based on the current status.
Although his suggestion was apparently supported by several key
players in this group, I'm afraid details on actual changes for
rfc2462bis may still vary. So I'll soon throw concrete idea of the
changes based on my own interpretation of his suggestion in a separate
message.
JINMEI, Tatuya
Communication Platform Lab.
Corporate R&D Center, Toshiba Corp.
[EMAIL PROTECTED]
--- Begin Message ---
I think a whole lot of the issue has to do with the supposedly mandatory nature of the
M flag, which leads to phrases like "do DHCP, and only if it fails do auto-config." It
would be much simpler to simply define the flags as "announcing an available service",
as in:
1) The "M" flag is set to indicate that a DHCPv6 address configuration service is
available on this link, as specified in RFC3315.
2) The "O" flag is set to indicate that a DHCPv6 information service is available on
this link, as specified in RFC3736.
We should then leave it at that, and leave it to nodes to decide whether they want to
use these services or not. For example, a server with a configured address will never
use DHCPv6 address configuration; an appliance that never has to resolve DNS names
will never use the information service. By setting the flags to indicate service
availability, we will reduce the amount of useless chatter on the link when the
services are not in fact available.
We should note that, from a protocol point of view, there is no need to use the M bit
to control stateless address configuration. This function is already achieved by the
"Autonomous flag" in the prefix information option. If the flag is not set, the hosts
will not configure information from the prefix.
-- Christian Huitema
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[EMAIL PROTECTED]
Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
--- End Message ---
