Re: [Roll] Making RUL draft normative reference in useofrplinfo

["Pascal Thubert (pthubert)" <[e-mail]>]

It's getting urgent to publish with the coming freeze. As for the RUL draft I'm suggesting a clearer language to address the below:

"
6.2.  External Routes and RPL Artifacts

   Section 4.1. of [USEofRPLinfo] provides a set of rules that MUST be
   followed when forwarding packets over an external route:

  ...

   Non-Storing Mode DAO messages are used to signal external routes to
   the Root, even if the DODAG is operated in Storing Mode.  This
   enables to advertise the 6LR that injects the route for use as tunnel
   endpoint in the data path.  For all external routes, the Root should
   use an IP-in-IP tunnel to that 6LR, with the RPL artifacts in the
   outer header to be stripped by the 6LR.  The IP-in-IP encapsulation
   may be avoided in storing mode if the path to the external
   destination beyond the 6LR is known to handle or ignore the RPL
   artifacts properly [RFC8200].  A RUL is an example of a destination
   that is reachable via an external (host) route for which IP-in-IP
   tunneling may be avoided as it ignores the RPI and the consumed SRH
   artifacts.  The use of Non-Storing mode signaling in Storing Mode and
   the associated IP-in-IP encapsulation are transparent to intermediate
   routers that only see packets back and forth between the Root and the
   6LR and do not need a special support for external routes.

...

"

I'll also make a PR for the change below in useofrplinfo.

All the best,

Pascal

From: Pascal Thubert (pthubert)
Sent: mercredi 30 octobre 2019 13:59
To: Routing Over Low power and Lossy networks <[e-mail]>
Cc: Michael Richardson <[e-mail]>
Subject: RE: [Roll] Making RUL draft normative reference in useofrplinfo

Hello Rahul, all

From: Roll <[e-mail]<mailto:[e-mail]>> On Behalf Of Rahul Jadhav
Sent: dimanche 27 octobre 2019 05:48
To: roll <[e-mail]<mailto:[e-mail]>>
Cc: Michael Richardson <[e-mail]<mailto:[e-mail]>>
Subject: Re: [Roll] Making RUL draft normative reference in useofrplinfo

Thank you for clarifying the thoughts in the mail. I am sure I would not have caught a drift if this would have been talked about directly (without a mail) during the 106-session.

Specifically, I want to talk about the following text:

"

This specification updates RPL [RFC6550] to RECOMMEND that external


targets are advertised using Non-Storing Mode signaling even in a


Storing-Mode network. This way, all packets to an external target


reach the Root that can encapsulate them, and the Root is informed of


the address of the 6LR that terminates the IP-in-IP tunnel.
"
I understand the reason why we need IP-in-IP tunnel terminating at the 6LR advertising the external-target. But when we say that only external targets are advertised using non-storing mode signaling even in a storing MOP, what we are asking is that all the 6LRs en-route (and not only the 6LR to which the external target is attaching) handle the external target in non-storing mode.
This means that all 6LRs in the network need to be capable of non-storing and storing mode even though the MOP advertised was for storing MOP only. This has implications for behaviour during parent switching. The text opens up a lot of room for interpretations and uncovered-scenarios. I am apprehensive of adding such text to useofrplinfo.


<Pascal>
( : Misunderstanding warning : )

The 'signaling' discussed in that text is the control plane, more specifically the DAO that is sent by the 6LR directly to the root as opposed to percolating child to parent like other internal DAO message in storing mode. In the data plane you have IP in IP to the 6LR and no RH, so the intermediate routers just see a packet to the 6LR and need not worry whether there's an IP packet inside.

We certainly need to clarify. What about:

"

This specification updates RPL [RFC6550] to RECOMMEND that external

targets are advertised using Non-Storing Mode DAO messaging even in a

Storing-Mode network. This way, external routes are not advertised within

the DODAG and all packets to an external target reach the Root like normal

non-storing mode traffic. The Non-Storing Mode DAO informs the Root of the

address of the 6LR that injects the external route, and the root uses IP-in-IP

encapsulation to that 6LR, which terminates the IP-in-IP tunnel and forwards

the original packet outside the RPL domain.

"
</Pascal>




Regarding following, "Do we limit to the ground common with the RUL draft for which there is a defined RPL behavior, or do we define a new different thing where Use-Of-RPL-Info handles an unclear superset of the nodes that support the RUL draft and for which there is a defined RPL operation? "

I certainly feel that we should limit the ground common for which there is a defined RPL behavior.

In general, unaware-leaves seem to have a much broader impact than I initially thought.


<Pascal>
Cool, we agree : )
</Pascal>

Regards,
Rahul


________________________________
Then that text was stripped off the RUL draft and placed it in the Use-Of-RPL-Info because it was really text on the forwarding behavior, which is different if some assumptions cannot be made on the external destination (need to encaps to the parent when it is not necessarily needed for a RUL). That new text also fills a gap in Use-Of-RPL-Info and is completely localized in a  new section. the  text was moved about routing from the RUL draft to that section. Since it is not published yet, please find that text inline below:



"



4.1. Updates to RFC6550: Advertise External Routes with Non-Storing


Mode Signaling.



Section 6.7.8. of [RFC6550] introduces the 'E' flag that is set to


indicate that the 6LR that generates the DAO redistributes external


targets into the RPL network. An external Target is a Target that


has been learned through an alternate protocol, for instance a route


to a prefix that is outside the RPL domain but reachable via a 6LR.


Being outside of the RPL domain, a node that is reached via an


external target cannot be guaranteed to ignore the RPL artifacts and


cannot be expected to process the [RFC8138] compression correctly.


This means that the RPL artifacts should be contained in an IP-in-IP


encapsulation that is removed by the 6LR, and that any remaining


compression should be expanded by the 6LR before it forwards a packet


outside the RPL domain.



This specification updates RPL [RFC6550] to RECOMMEND that external


targets are advertised using Non-Storing Mode signaling even in a


Storing-Mode network. This way, all packets to an external target


reach the Root that can encapsulate them, and the Root is informed of


the address of the 6LR that terminates the IP-in-IP tunnel. In case


of an external target, the Root SHOULD use the same IP-in-IP


encapsulation for packets that it generates or that are originated


within the RPL domain as if the packets were coming from the


Internet.



A RUL is a special case of external target when the target is


actually a host and it is known to support a consumed Routing Header


and to ignore a HbH header as prescribed by [RFC8200].. The target


may have been learned through as a host route or may have been


registered to the 6LR using [RFC8505]. IP-in-IP encapsulation MAY be


avoided for Root to RUL communication if the RUL is known to process


the packets as forwarded by the parent 6LR without decapsulation.



In order to enable IP-in-IP all the way to a 6LN, it is beneficial


that the 6LN supports decapsulating IP-in-IP, but that is not assumed


by [RFC8504]. If the 6LN is a RUL, the Root that encapsulates a


packet SHOULD terminate the tunnel at a parent 6LR unless it is aware


that the RUL supports IP-in-IP decapsulation.



A node that is reachable over an external route is not expected to


support [RFC8138]. Whether a decapsulation took place or not and


even when the 6LR is delivering the packet to a RUL, the 6LR that


injected an external route MUST uncompress the packet before


forwarding over that external route.

"



So where are we?



We need to agree on what we place in the definition of a RUL for which Use-Of-RPL-Info presents a special behavior. Do we limit to the ground common with the RUL draft for which there is a defined RPL behavior, or do we define a new different thing where Use-Of-RPL-Info handles an unclear superset of the nodes that support the RUL draft and for which there is a defined RPL operation? I do not see the need because if (ever, though doubtful) a new draft defines the RPL operation for one version of "that guy" then is can also say whether "that guy" should be handled as a RUL or as an "external destination" in Use-Of-RPL-Info.



Thus the proposal to that that the operation of a "RUL" in Use-Of-RPL-Info is for a node that supports the RUL draft, and make the RUL draft a normative reference to define what that exactly mean. If we do so, MISSREF will play its guardian role to ensure that Use-Of-RPL-Info is consistent with the RUL draft till the last minute because it is not published too hastly.



Looking forward to your reactions,



Ines, Pascal and Michael.