|
            +-----+
            |     | Other (default) Router
            |     |
            +-----+ 
               | /64
               |      Backbone Link
         +-------------------+-------------------+
         | /64               | /64               | /64
      +-----+             +-----+             +-----+
      |     | Backbone    |     | Backbone    |     | Backbone
      |     | router      |     | router      |     | router
      +-----+             +-----+             +-----+
         o N*/128       o o  o M*/128          o o P*/128
     o o   o  o       o o   o  o  o         o  o  o  o o
    o  o o  o o       o   o  o  o  o        o  o  o o o
    o   o  o  o          o    o  o           o  o   o
      o   o o               o  o                 o o

      LLN              LLN              LLN

Figure 3: Routing Configuration in the ML Subnet

6.1. Over the Backbone Link

The Backbone Router is a specific kind of Border Router that performs proxy Neighbor Discovery on its backbone interface on behalf of the nodes that it has discovered on its LLN interfaces.

The backbone is expected to be a high speed, reliable Backbone link, with affordable and reliable multicast capabilities, such as a bridged Ethernet Network, and to allow a full support of classical ND as specified in [RFC4861] and subsequent RFCs. In other words, the backbone is not a LLN.

Still, some restrictions of the attached LLNs will apply to the backbone. In particular, it is expected that the MTU is set to the same value on the backbone and all attached LLNs, and the scalability of the whole subnet requires that broadcast operations are avoided as much as possible on the backbone as well. Unless configured otherwise, the Backbone Router MUST echo the MTU that it learns in RAs over the backbone in the RAs that it sends towards the LLN links.

As a router, the Backbone Router behaves like any other IPv6 router on the backbone side. It has a connected route installed towards the backbone for the prefixes that are present on that backbone and that it proxies for on the LLN interfaces.

As a proxy, the 6BBR uses an EARO option in the NS-DAD and the multicast NA messages that it generates on behalf of a Registered Node, and it places an EARO in its unicast NA messages if and only if the NS/NA that stimulates it had an EARO in it.

When possible, the 6BBR SHOULD use unicast or solicited-node multicast address (SNMA) [RFC4291] to defend its Registered Addresses over the backbone. In particular, the 6BBR MUST join the SNMA group that corresponds to a Registered Address as soon as it creates an entry for that address and as long as it maintains that entry, whatever the state of the entry. The expectation is that it is possible to get a message delivered to all the nodes on the backbone that listen to a particular address and support this specification - which includes all the 6BBRs in the MultiLink Subnet - by sending a multicast message to the associated SNMA over the backbone.

The support of Optimistic DAD (ODAD) [RFC4429] is recommended for all nodes in the backbone and followed by the 6BBRs in their proxy activity over the backbone. With ODAD, any optimistic node MUST join the SNMA of a Tentative address, which interacts better with this specification.

This specification allows the 6BBR in Routing Proxy mode to advertise the Registered IPv6 Address with the 6BBR Link Layer Address, and attempts to update Neighbor Cache Entries (NCE) in correspondent nodes over the backbone, using gratuitous NA(Override). This method may fail of the multicast message is not properly received, and correspondent nodes may maintain an incorrect neighbor state, which they will eventually discover through Neighbor Unreachability Detection (NUD). Because mobility may be slow, the NUD procedure defined in [RFC4861] may be too impatient, and the support of [RFC7048] is recommended in all nodes in the network.

Since the MultiLink Subnet may grow very large in terms of individual IPv6 addresses, multicasts should be avoided as much as possible even on the backbone. Though it is possible for plain hosts to participate with legacy IPv6 ND support, the support by all nodes connected to the backbone of [I-D.nordmark-6man-rs-refresh] is recommended, and this implies the support of [RFC7559] as well.

6.2. Over the LLN Link

As a router, the Nodes and Backbone Router operation on the LLN follows [RFC6775]. Per that specification, LLN Hosts generally do not depend on multicast RAs to discover routers. It is still generally required for LLN nodes to accept multicast RAs [I-D.ietf-v6ops-reducing-ra-energy-consumption], but those are rare on the LLN link. Nodes are expected to follow the Simple Procedures for Detecting Network Attachment in IPv6 [RFC6059] (DNA procedures) to assert movements, and to support the Packet-Loss Resiliency for Router Solicitations [RFC7559] to make the unicast RS more reliable.

The Backbone Router acquires its states about the addresses on the LLN side through a registration process from either the nodes themselves, or from a node such as a RPL root / 6LBR (the Registering Node) that performs the registration on behalf of the address owner (the Registered Node).

When operating as a Routing Proxy, the router installs hosts routes (/128) to the Registered Addresses over the LLN links, via the Registering Node as identified by the Source Address and the SLLAO option in the NS(EARO) messages.

In that mode, the 6BBR handles the ND protocol over the backbone on behalf of the Registered Nodes, using its own MAC address in the TLLA and SLLA options in proxyed NS and NA messages. It results that for each Registered Address, a number of peer Nodes on the backbone have resolved the address with the 6BBR MAC address and keep that mapping stored in their Neighbor cache.

The 6BBR SHOULD maintain, per Registered Address, the list of the peers on the backbone to which it answered with it MAC address, and when a binding moves to a different 6BBR, it SHOULD send a unicast gratuitous NA(O) individually to each of them to inform them that the address has moved and pass the MAC address of the new 6BBR in the TLLAO option. If the 6BBR can not maintain that list, then it SHOULD remember whether that list is empty or not and if not, send a multicast NA(O) to all nodes to update the impacted Neighbor Caches with the information from the new 6BBR.

The Bridging Proxy is a variation where the BBR function is implemented in a Layer-3 switch or an wireless Access Point that acts as a Host from the IPv6 standpoint, and, in particular, does not operate the routing of IPv6 packets. In that case, the SLLAO in the proxied NA messages is that of the Registering Node and classical bridging operations take place on data frames.

If a registration moves from one 6BBR to the next, but the Registering Node does not change, as indicated by the S/TLLAO option in the ND exchanges, there is no need to update the Neighbor Caches in the peers Nodes on the backbone. On the other hand, if the LLAO changes, the 6BBR SHOULD inform all the relevant peers as described above, to update the impacted Neighbor Caches. In the same fashion, if the Registering Node changes with a new registration, the 6BBR SHOULD also update the impacted Neighbor Caches over the backbone.

7. BackBone Router Proxy Operations

This specification enables a Backbone Router to proxy Neighbor Discovery operations over the backbone on behalf of the nodes that are registered to it, allowing any node on the backbone to reach a Registered Node as if it was on-link. The backbone and the LLNs are considered different Links in a MultiLink subnet but the prefix that is used may still be advertised as on-link on the backbone to support legacy nodes; multicast ND messages are link-scoped and not forwarded across the backbone routers.

ND Messages on the backbone side that do not match to a registration on the LLN side are not acted upon on the LLN side, which stands protected. On the LLN side, the prefixes associated to the MultiLink Subnet are presented as not on-link, so address resolution for other hosts do not occur.

The default operation in this specification is Sleeping proxy which means:

• creating a new entry in an abstract Binding Table for a new Registered Address and validating that the address is not a duplicate over the backbone
• defending a Registered Address over the backbone using NA messages with the Override bit set on behalf of the sleeping node whenever possible
• advertising a Registered Address over the backbone using NA messages, asynchronously or as a response to a Neighbor Solicitation messages.
• Looking up a destination over the backbone in order to deliver packets arriving from the LLN using Neighbor Solicitation messages.
• Forwarding packets from the LLN over the backbone, and the other way around.
• Eventually triggering a liveliness verification of a stale registration.

A 6BBR may act as a Sleeping Proxy only if the state of the binding entry is REACHABLE, or TENTATIVE in which case the answer is delayed. In any other state, the Sleeping Proxy operates as a Unicasting Proxy.

As a Unicasting Proxy, the 6BBR forwards NS messages to the Registering Node, transforming Layer-2 multicast into unicast whenever possible. This is not possible in UNREACHABLE state, so the NS messages are multicasted, and rate-limited to protect the medium with an exponential back-off. In other states, The messages are forwarded to the Registering Node as unicast Layer-2 messages. In TENTATIVE state, the NS message is either held till DAD completes, or dropped.

The draft introduces the optional concept of primary and secondary BBRs. The primary is the backbone router that has the highest EUI-64 address of all the 6BBRs that share a registration for a same Registered Address, with the same Owner Unique ID and same Transaction ID, the EUI-64 address being considered as an unsigned 64bit integer. The concept is defined with the granularity of an address, that is a given 6BBR can be primary for a given address and secondary or another one, regardless on whether the addresses belong to the same node or not. The primary Backbone Router is in charge of protecting the address for DAD over the Backbone. Any of the Primary and Secondary 6BBR may claim the address over the backbone, since they are all capable to route from the backbone to the LLN node, and the address appears on the backbone as an anycast address.

The Backbone Routers maintain a distributed binding table, using classical ND over the backbone to detect duplication. This specification requires that:

1. All addresses that can be reachable from the backbone, including IPv6 addresses based on burn-in EUI64 addresses MUST be registered to the 6BBR.
2. A Registered Node MUST include the EARO option in an NS message that used to register an addresses to a 6LR; the 6LR MUST propagate that option unchanged to the 6LBR in the DAR/DAC exchange, and the 6LBR MUST propagate that option unchanged in proxy registrations.
3. The 6LR MUST echo the same EARO option in the NA that it uses to respond, but for the status filed which is not used in NS messages, and significant in NA.

A false positive duplicate detection may arise over the backbone, for instance if the Registered Address is registered to more than one LBR, or if the node has moved. Both situations are handled gracefully unbeknownst to the node. In the former case, one LBR becomes primary to defend the address over the backbone while the others become secondary and may still forward packets back and forth. In the latter case the LBR that receives the newest registration wins and becomes primary.

The expectation in this specification is that there is a single Registering Node at a time per Backbone Router for a given Registered Address, but that a Registered Address may be registered to Multiple 6BBRs for higher availability.

Over the LLN, and for any given Registered Address, it is REQUIRED that:

• de-registrations (newer TID, same OUID, null Lifetime) are accepted and responded immediately with a status of 4; the entry is deleted;
• newer registrations (newer TID, same OUID, non-null Lifetime) are accepted and responded with a status of 0 (success); the entry is updated with the new TID, the new Registration Lifetime and the new Registering Node, if any has changed; in TENTATIVE state the response is held and may be overwritten; in other states the Registration-Lifetime timer is restarted and the entry is placed in REACHABLE state.
• identical registrations (same TID, same OUID) from a same Registering Node are not processed but responded with a status of 0 (success); they are expected to be identical and an error may be logged if not; in TENTATIVE state, the response is held and may be overwritten, but it MUST be eventually produced and it carries the result of the DAD process;
• older registrations (not(newer or equal) TID, same OUID) from a same Registering Node are ignored;
• identical and older registrations (not-newer TID, same OUID) from a different Registering Node are responded immediately with a status of 3 (moved); this may be rate limited to protect the medium;
• and any registration for a different Registered Node (different OUID) are responded immediately with a status of 1 (duplicate).

7.1. Registration and Binding State Creation

Upon a registration for a new address with an NS(EARO), the 6BBR performs a DAD operation over the backbone placing the new address as target in the NS-DAD message. The EARO from the registration MUST be placed unchanged in the NS-DAD message, and an entry is created in TENTATIVE state for a duration of TENTATIVE_DURATION. The NS-DAD message is sent multicast over the backbone to the SNMA address associated with the registered address. If that operation is known to be costly, and the 6BBR has an indication from another source (such as a NCE) that the Registered Address was present on the backbone, that information may be leveraged to send the NS-DAD message as a Layer-2 unicast to the MAC that was associated with the Registered Address.

In TENTATIVE state:

• the entry is removed if an NA is received over the backbone for the Registered Address with no EARO option, or with an EARO option with a status of 1 (duplicate) that indicates an existing registration for another LLN node. The OUID and TID fields in the EARO option received over the backbone are ignored. A status of 1 is returned in the EARO option of the NA back to the Registering Node;
• the entry is also removed if an NA with an ARO option with a status of 3 (moved), or a NS-DAD with an ARO option that indicates a newer registration for the same Registered Node, is received over the backbone for the Registered Address. A status of 3 is returned in the NA(EARO) back to the Registering Node;
• when a registration is updated but not deleted, e.g. from a newer registration, the DAD process on the backbone continues and the running timers are not restarted;
• Other NS (including DAD with no EARO option) and NA from the backbone are not responded in TENTATIVE state, but the list of their origins may be kept in memory and if so, the 6BBR may send them each a unicast NA with eventually an EARO option when the TENTATIVE_DURATION timer elapses, so as to cover legacy nodes that do not support ODAD.
• When the TENTATIVE_DURATION timer elapses, a status 0 (success) is returned in a NA(EARO) back to the Registering Node(s), and the entry goes to REACHABLE state for the Registration Lifetime; the DAD process is successful and the 6BBR MUST send a multicast NA(EARO) to the SNMA associated to the Registered Address over the backbone with the Override bit set so as to take over the binding from other 6BBRs.

7.2. Defending Addresses

If a 6BBR has an entry in REACHABLE state for a Registered Address:

• If the 6BBR is primary, or does not support the concept, it MUST defend that address over the backbone upon an incoming NS-DAD, either if the NS does not carry an EARO, or if an EARO is present that indicates a different Registering Node (different OUID). The 6BBR sends a NA message with the Override bit set and the NA carries an EARO option if and only if the NS-DAD did so. When present, the EARO in the NA(O) that is sent in response to the NS-DAD(EARO) carries a status of 1 (duplicate), and the OUID and TID fields in the EARO option are obfuscated with null or random values to avoid network scanning and impersonation attacks.
• If the 6BBR receives an NS-DAD(EARO) that reflect a newer registration, the 6BBR updates the entry and the routing state to forward packets to the new 6BBR, but keeps the entry REACHABLE. In that phase, it MAY use REDIRECT messages to reroute traffic for the Registered Address to the new 6BBR.
• If the 6BBR receives an NA(EARO) that reflect a newer registration, the 6BBR removes its entry and sends a NA(AERO) with a status of 3 (moved) to the Registering Node, if the Registering Node is different from the Registered Node. If necessary, the 6BBR cleans up ND cache in peers nodes as discussed in Section 6.1, by sending a series of unicast to the impacted nodes, or one broadcast NA(O) to all-nodes.
• If the 6BBR received a NS(LOOKUP) for a Registered Address, it answers immediately with an NA on behalf of the Registered Node, without polling it. There is no need of an EARO in that exchange.
• When the Registration-Lifetime timer elapses, the entry goes to STALE state for a duration of STABLE_STALE_DURATION in LLNs that keep stable addresses such as LWPANs, and UNSTABLE_STALE_DURATION in LLNs where addresses are renewed rapidly, e.g. for privacy reasons.

The STALE state is a chance to keep track of the backbone peers that may have an ND cache pointing on this 6BBR in case the Registered Address shows back up on this or a different 6BBR at a later time. In STALE state:

• If the Registered Address is claimed by another node on the backbone, with an NS-DAD or an NA, the 6BBR does not defend the address. Upon an NA(O), or the stale time elapses, the 6BBR removes its entry and sends a NA(AERO) with a status of 4 (removed) to the Registering Node.
• If the 6BBR received a NS(LOOKUP) for a Registered Address, the 6BBR MUST send an NS(NUD) following rules in [RFC7048] to the registering Node targeting the Registered Address prior to answering. If the NUD succeeds, the operation in REACHABLE state applies. If the NUD fails, the 6BBR refrains from answering the lookup. The NUD expected to be mapped by the Registering Node into a liveliness validation of the Registered Node if they are in fact different nodes.

8. Security Considerations

This specification expects that the link layer is sufficiently protected, either by means of physical or IP security for the Backbone Link or MAC sublayer cryptography. In particular, it is expected that the LLN MAC provides secure unicast to/from the Backbone Router and secure Broadcast from the Backbone Router in a way that prevents tempering with or replaying the RA messages.

The use of EUI-64 for forming the Interface ID in the link local address prevents the usage of Secure ND ([RFC3971] and [RFC3972]) and address privacy techniques. This specification RECOMMENDS the use of additional protection against address theft such as provided by [I-D.sarikaya-6lo-ap-nd], which guarantees the ownership of the OUID.

When the ownership of the OUID cannot be assessed, this specification limits the cases where the OUID and the TID are multicasted, and obfuscates them in responses to attempts to take over an address.

9. Protocol Constants

This Specification uses the following constants:

TENTATIVE_DURATION:

800 milliseconds

STABLE_STALE_DURATION:

24 hours

UNSTABLE_STALE_DURATION:

5 minutes

DEFAULT_NS_POLLING:

3 times

10. IANA Considerations

This document requires the following additions: