Re: [v6ops] M/O flags and PD

[Mark Andrews <[e-mail]>]

In message <[e-mail]>, Brian E Carpenter writes:
> On 31/10/2015 04:24, Sander Steffann wrote:
> > Hi Tore,
> > 
> > Op 30 okt. 2015, om 09:10 heeft Tore Anderson <[e-mail]> het volgende geschreven:
> >> * Gert Doering <[e-mail]>
> >>
> >>> I do not have RFCs to counter that, but my interpretation is a bit
> >>> different.  I see that section of 4861 as traditionally applied to 
> >>> *hosts*, while IA_PD traditionally is a *router* function, so out of 
> >>> scope for the "host part" of 4861.
> >>
> >> Section 4.2 of RFC 4861 simply describes the RA message format, it's
> >> not in a "host part" of the RFC as far as I can tell. Keep in mind that
> >> RAs are typically multicast to all IPv6 nodes, so the originating router
> >> does not know whether it is a host or a router (or both, or none) that
> >> will be receiving the RA.
> > 
> > I always understood this bit from RFC 3633 to mean that any host configuration flags (such as M/O) are not relevant to DHCPv6-PD: "Note that this use of DHCP is 
> not bound to the assignment of IP addresses or other configuration information to hosts".
> 
> The phrase "is not bound to" is rather unclear, don't you think? It neither says
> nor implies that routers are thereby exempt from issuing correct RAs.

Read the whole paragraph.  This has nothing to do with RA bits.

   "Note that this use of DHCP is not bound to the assignment of IP
   addresses or other configuration information to hosts, and that no
   mechanism is currently available to communicate delegated prefixes to
   a DHCP server that serves such a function.  This may be an item of
   future work, should usage warrant."

I see that as saying that there is no mechanism defined to inform
a DHCPv6 server of the prefix obtained and that the IETF may need
to do so in the future.  In most cases of the use cases DHCP server
is in the router so the router vendor just figures this out for
themselves (e.g. the dhcp client updates the dhcp server's configuration
and tells the dhcp server to reload).  In case where there is a
seperate DHCP server and the router is just the DHCP relay is left
unspecified how the obtained addresses are communicated.

Additional there is no requirement for there to be a DHCPv6 server
at all.  SLAAC or something else is sufficient.  This is the first
part of the paragraph.

This document is not a complete solution.  There are missing bits that
need to be specified to make it a complete solution.

> However, being realistic, note that
> On 30/10/2015 21:55, Tore Anderson wrote:
> 
> > It boils down to this: If you are about to originate an RA to a link, and
> > know that there is information available in DHCPv6 on said link, you
> > should set set the M or O flag.
> 
> Even if you accept that, in the case that you *don't* know that such information
> is available, you physically cannot set the M or O flag. So in the real world,
> a node that is interested in IA_PD had better go fishing for it, whatever the
> MO bits say. So I retract all the if statements in the pseudocode that I sent.
> The only robust implementation is one that always looks for DHCPv6 information.
> 
>     Brian
> 
> 
> _______________________________________________
> v6ops mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/v6ops
-- 
Mark Andrews, ISC
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: [phone]                 INTERNET: [e-mail]