Internet Engineering Task Force (IETF)                 T. Henderson, Ed.
Request for Comments: 8047                      University of Washington
Category: Standards Track                                        C. Vogt
ISSN: 2070-1721                                              Independent
                                                                J. Arkko
                                                                Ericsson
                                                           December 2016
                                                           February 2017

            Host Multihoming with the Host Identity Protocol

Abstract

   This document defines host multihoming extensions to the Host
   Identity Protocol (HIP), by leveraging protocol components defined
   for host mobility.

Status of This Memo

   This is an Internet Standards Track document.

   This document is a product of the Internet Engineering Task Force
   (IETF).  It represents the consensus of the IETF community.  It has
   received public review and has been approved for publication by the
   Internet Engineering Steering Group (IESG).  Further information on
   Internet Standards is available in Section 2 of RFC 7841.

   Information about the current status of this document, any errata,
   and how to provide feedback on it may be obtained at
   http://www.rfc-editor.org/info/rfc8047.

Copyright Notice

   Copyright (c) 2016 2017 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents
   (http://trustee.ietf.org/license-info) in effect on the date of
   publication of this document.  Please review these documents
   carefully, as they describe your rights and restrictions with respect
   to this document.  Code Components extracted from this document must
   include Simplified BSD License text as described in Section 4.e of
   the Trust Legal Provisions and are provided without warranty as
   described in the Simplified BSD License.

Table of Contents

   1.  Introduction and Scope  . . . . . . . . . . . . . . . . . . .   2
   2.  Terminology and Conventions . . . . . . . . . . . . . . . . .   3
   3.  Protocol Model  . . . . . . . . . . . . . . . . . . . . . . .   4
   4.  Protocol Overview . . . . . . . . . . . . . . . . . . . . . .   4
     4.1.  Background  . . . . . . . . . . . . . . . . . . . . . . .   4
     4.2.  Usage Scenarios . . . . . . . . . . . . . . . . . . . . .   5
       4.2.1.  Multiple Addresses  . . . . . . . . . . . . . . . . .   6
       4.2.2.  Multiple Security Associations  . . . . . . . . . . .   6
       4.2.3.  Host Multihoming for Fault Tolerance  . . . . . . . .   7
       4.2.4.  Host Multihoming for Load Balancing . . . . . . . . .   9
       4.2.5.  Site Multihoming  . . . . . . . . . . . . . . . . . .  10
       4.2.6.  Dual-Host Multihoming . . . . . . . . . . . . . . . .  10
       4.2.7.  Combined Mobility and Multihoming . . . . . . . . . .  11
       4.2.8.  Initiating the Protocol in R1, I2, or R2  . . . . . .  11
       4.2.9.  Using LOCATOR_SETs across Addressing Realms . . . . .  13
     4.3.  Interaction with Security Associations  . . . . . . . . .  13
   5.  Processing Rules  . . . . . . . . . . . . . . . . . . . . . .  14
     5.1.  Sending LOCATOR_SETs  . . . . . . . . . . . . . . . . . .  14
     5.2.  Handling Received LOCATOR_SETs  . . . . . . . . . . . . .  16
     5.3.  Verifying Address Reachability  . . . . . . . . . . . . .  18
     5.4.  Changing the Preferred Locator  . . . . . . . . . . . . .  18
   6.  Security Considerations . . . . . . . . . . . . . . . . . . .  19
   7.  References  . . . . . . . . . . . . . . . . . . . . . . . . .  21
     7.1.  Normative References  . . . . . . . . . . . . . . . . . .  21
     7.2.  Informative References  . . . . . . . . . . . . . . . . .  21
   Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . .  22
   Authors' Addresses  . . . . . . . . . . . . . . . . . . . . . . .  22

1.  Introduction and Scope

   The Host Identity Protocol (HIP) [RFC7401] supports an architecture
   that decouples the transport layer (TCP, UDP, etc.) from the
   internetworking layer (IPv4 and IPv6) by using public/private key
   pairs, instead of IP addresses, as host identities.  When a host uses
   HIP, the overlying protocol sublayers (e.g., transport-layer sockets
   and Encapsulating Security Payload (ESP) Security Associations (SAs))
   are instead bound to representations of these host identities, and
   the IP addresses are only used for packet forwarding.  However, each
   host must also know at least one IP address at which its peers are
   reachable.  Initially, these IP addresses are the ones used during
   the HIP base exchange.

   One consequence of such a decoupling is that new solutions to
   network-layer mobility and host multihoming are possible.  Basic host
   mobility is defined in [RFC8046] and covers the case in which a host
   has a single address and changes its network point of attachment
   while desiring to preserve the HIP-enabled security association.
   Host multihoming is somewhat of a dual case to host mobility, in
   that, a host may simultaneously have more than one network point of
   attachment.  There are potentially many variations of host
   multihoming possible.  [RFC8046] specifies the format of the HIP
   parameter (LOCATOR_SET parameter) used to convey IP addressing
   information between peers, the procedures for sending and processing
   this parameter to enable basic host mobility, and procedures for an
   address verification mechanism.  The scope of this document
   encompasses messaging and elements of procedure for some basic host
   multihoming scenarios of interest.

   Another variation of multihoming that has been heavily studied is
   site multihoming.  Solutions for host multihoming in multihomed IPv6
   networks have been specified by the IETF shim6 working group.  The
   Shim6 protocol [RFC5533] bears many architectural similarities to
   HIP, but there are differences in the security model and in the
   protocol.

   While HIP can potentially be used with transports other than the ESP
   transport format [RFC7402], this document largely assumes the use of
   ESP and leaves other transport formats for further study.

   Finally, making underlying IP multihoming transparent to the
   transport layer has implications on the proper response of transport
   congestion control, path MTU selection, and Quality of Service (QoS).
   Transport-layer mobility triggers, and the proper transport response
   to a HIP multihoming address change, are outside the scope of this
   document.

   This specification relies on implementing Sections 4 ("LOCATOR_SET
   Parameter Format") and 5 ("Processing Rules") of [RFC8046] as a
   starting point for this implementation.

2.  Terminology and Conventions

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
   document are to be interpreted as described in RFC 2119 [RFC2119].

   The following terms used in this document are defined in [RFC8046]:
   LOCATOR_SET, Locator, locator, Address, Preferred preferred locator, and
   Credit-Based Authorization.

3.  Protocol Model

   The protocol model for HIP support of host multihoming extends the
   model for host mobility described in Section 3 of [RFC8046].  This
   section only highlights the differences.

   In host multihoming, a host has multiple locators simultaneously
   rather than sequentially, as in the case of mobility.  By using the
   LOCATOR_SET parameter defined in [RFC8046], a host can inform its
   peers of additional (multiple) locators at which it can be reached.
   When multiple locators are available and announced to the peer, a
   host can designate a particular locator as a "preferred" locator,
   meaning that the host prefers that its peer send packets to the
   designated address before trying an alternative address.  Although
   this document defines a basic mechanism for multihoming, it does not
   define all possible policies and procedures, such as which locators
   to choose when more than one is available, the operation of
   simultaneous mobility and multihoming, source address selection
   policies (beyond those specified in [RFC6724]), and the implications
   of multihoming on transport protocols.

4.  Protocol Overview

   In this section, we briefly introduce a number of usage scenarios for
   HIP multihoming.  These scenarios assume that HIP is being used with
   the ESP transport [RFC7402], although other scenarios may be defined
   in the future.  To understand these usage scenarios, the reader
   should be at least minimally familiar with the HIP protocol
   specification [RFC7401], the use of the ESP transport format
   [RFC7402], and the HIP mobility specification [RFC8046].  However,
   for the (relatively) uninitiated reader, it is most important to keep
   in mind that in HIP, the actual payload traffic is protected with
   ESP, and that the ESP Security Parameter Index (SPI) acts as an index
   to the right host-to-host context.

4.1.  Background

   The multihoming scenarios can be explained in contrast to the non-
   multihoming case described in the base protocol specification. specification
   [RFC7401].  We review the pertinent details here.  In the base
   specification, when used with the ESP transport format, the HIP base
   exchange will set up a single SA in each direction.  The IP addresses
   associated with the SAs are the same as those used to convey the HIP
   packets.  For data traffic, a security policy database (SPD) and
   security association database (SAD) will likely exist, following the
   IPsec architecture.  One distinction between HIP and IPsec, however,
   is that the host IDs, and not the IP addresses, are conceptually used
   as selectors in the SPD.  In the outbound direction, as a result of
   SPD processing, when an outbound SA is selected, the correct IP
   destination address for the peer must also be assigned.  Therefore,
   outbound SAs are conceptually associated with the peer IP address
   that must be used as the destination IP address below the HIP layer.
   In the inbound direction, the IP addresses may be used as selectors
   in the SAD to look up the SA, but they are not strictly required; the
   ESP SPI may be used alone.  To summarize, in the non-multihoming
   case, there is only one source IP address, one destination IP
   address, one inbound SA, and one outbound SA.

   The HIP readdressing protocol [RFC8046] is an asymmetric protocol in
   which a mobile or multihomed host informs a peer host about changes
   of IP addresses on affected SPIs.  IP address and ESP SPI information
   is carried in Locator data structures fields in a HIP parameter called a LOCATOR_SET.
   The HIP mobility specification [RFC8046] describes how the
   LOCATOR_SET is carried in a HIP UPDATE packet.

   To summarize the mobility elements of procedure, as background for
   multihoming, the basic idea of host mobility is to communicate a
   local IP address change to the peer when active HIP-maintained SAs
   are in use.  To do so, the IP address must be conveyed, any
   association between the IP address and an inbound SA (via the SPI
   index) may be conveyed, and protection against flooding attacks must
   be ensured.  The association of an IP address with an SPI is
   performed by a Locator Type of type 1, "1", which is a concatenation of an
   ESP SPI with an IP address.

   An address verification method is specified in [RFC8046].  It is
   expected that addresses learned in multihoming scenarios also are
   subject to the same verification rules.  At times, the scenarios
   describe addresses as being in either an ACTIVE, VERIFIED, or
   DEPRECATED state.  From the perspective of a host, newly learned
   addresses of the peer must be verified before put into active
   service, and addresses removed by the peer are put into a deprecated
   state.  Under limited conditions described in [RFC8046], an
   UNVERIFIED address may be used.

   With this background, we next describe an additional protocol to
   facilitate scenarios in which one or both hosts have multiple IP
   addresses available.  Increasingly, this is the common case with
   network-connected hosts on the Internet.

4.2.  Usage Scenarios

4.2.1.  Multiple Addresses

   Hosts may have multiple IP addresses within different address
   families (IPv4 and IPv6) and scopes available to support HIP
   messaging and HIP-enabled SAs.  The multiple addresses may be on a
   single network interface or multiple network interfaces.  It is
   outside of the scope of this document to specify how a host decides
   which of possibly multiple addresses may be used to support a HIP
   association.  Some IP addresses may be held back from usage due to
   privacy, security, or cost considerations.

   When multiple IP addresses are shared with a peer, the procedures
   described in the HIP mobility specification [RFC8046] allow for a
   host to set a Preferred preferred locator ("P") bit, requesting that one of the
   multiple addresses be preferred for control- or data-plane traffic.
   It is also permitted to leave the Preferred preferred bit unset for all
   addresses, allowing the peer to make address selection decisions.

   Hosts that use link-local addresses as source addresses in their HIP
   handshakes may not be reachable by a mobile peer.  Such hosts SHOULD
   provide a globally routable address either in the initial handshake
   or via the LOCATOR_SET parameter.

   To support mobility, as described in the HIP mobility specification
   [RFC8046], the LOCATOR_SET may be sent in a HIP UPDATE packet.  To
   support multihoming, the LOCATOR_SET may also be sent in R1, I2, or
   R2 packets defined in the HIP protocol specification [RFC7401].  The
   reason to consider sending LOCATOR_SET parameters in base exchange
   packets is to convey all usable addresses for fault-tolerance or
   load-balancing considerations.

4.2.2.  Multiple Security Associations

   When multiple addresses are available between peer hosts, a question
   that arises is whether to use one or multiple SAs.  The intent of
   this specification is to support different use cases but to leave the
   policy decision to the hosts.

   When one host has n addresses and the other host has m addresses, it
   is possible to set up as many as (n * m) SAs in each direction.  In
   such a case, every combination of source and destination IP addresses
   would have a unique SA, and the possibility of the reordering of
   datagrams on each SA will be lessened (ESP SAs may have an anti-
   replay window [RFC4303] sensitive to reordering).  However, the
   downside to creating a mesh of SAs is the signaling overhead required
   (for exchanging UPDATE messages conveying ESP_INFO parameters) and
   the state maintenance required in the SPD/SAD.

   For load balancing, when multiple paths are to be used in parallel,
   it may make sense to create different SAs for different paths.  In
   this use case, while a full mesh of 2 * (n * m) SAs may not be
   required, it may be beneficial to create one SA pair per load-
   balanced path to avoid anti-replay window issues.

   For fault tolerance, it is more likely that a single SA and multiple
   IP addresses associated with that SA can be used, and the alternative
   addresses can be used only upon failure detection of the addresses in
   use.  Techniques for path failure detection are outside the scope of
   this specification.  An implementation may use ICMP interactions,
   reachability checks, or other means to detect the failure of a
   locator.

   In summary, whether and how a host decides to leverage additional
   addresses in a load-balancing or fault-tolerant manner is outside the
   scope of the specification (although the academic literature on
   multipath TCP schedulers may provide guidance on how to design such a
   policy).  However, in general, this document recommends that for
   fault tolerance, it is likely sufficient to use a single SA pair for
   all addresses, and for load balancing, to support a different SA pair
   for all active paths being balanced across.

4.2.3.  Host Multihoming for Fault Tolerance

   A (mobile or stationary) host may have more than one interface or
   global address.  The host may choose to notify the peer host of the
   additional interface or address by using the LOCATOR_SET parameter.
   The LOCATOR_SET parameter may be included in an I2, R1, or R2 packet,
   or it may be conveyed, after the base exchange completes in an UPDATE
   packet.

   When more than one locator is provided to the peer host, the host MAY
   indicate which locator is preferred (the locator on which the host
   prefers to receive traffic).  By default, the address that a host
   uses in the base exchange is its preferred locator (for the address
   family and address scope in use during the base exchange) until
   indicated otherwise.  It may be the case that the host does not
   express any preferred locators.

   In the multihoming case, the sender may also have multiple valid
   locators from which to source traffic.  In practice, a HIP
   association in a multihoming configuration may have both a preferred
   peer locator and a preferred local locator.  The host should try to
   use the peer's preferred locator unless policy or other circumstances
   prevent such usage.  A preferred local locator may be overridden if
   source address selection rules on the destination address (peer's
   preferred locator) suggest the use of a different source address.

   Although the protocol may allow for configurations in which there is
   an asymmetric number of SAs between the hosts (e.g., one host has two
   interfaces and two inbound SAs, while the peer has one interface and
   one inbound SA), it is suggested that inbound and outbound SAs be
   created pairwise between hosts.  When an ESP_INFO arrives to rekey a
   particular outbound SA, the corresponding inbound SA should also be
   rekeyed at that time.  Section 4.3 discusses the interaction between
   addresses and security associations in more detail.

   Consider the case of two hosts, one single-homed and one multihomed.
   The multihomed host may decide to inform the single-homed host about
   its other address(es).  It may choose to do so as follows.

   If the multihomed host wishes to convey the additional address(es)
   for fault tolerance, it should include all of its addresses in
   Locator records, fields, indicating the Traffic Type, Locator Type, and
   Preferred Locator for each address.
   whether the locator is a preferred locator.  If it wishes to bind any
   particular address to an existing SPI, it may do so by using a
   Locator of Type 1 of "1" as specified in the HIP mobility specification
   [RFC8046].  It does not need to rekey the existing SA or request
   additional SAs at this time.

   Figure 1 illustrates this scenario.  Note that the conventions for
   message parameter notations in figures (use of parentheses and
   brackets) is defined in Section 2.2 of [RFC7401].

     Multihomed Host                     Peer Host

              UPDATE(LOCATOR_SET, SEQ)
        ----------------------------------->
              UPDATE(ACK)
        <-----------------------------------

                   Figure 1: Basic Multihoming Scenario

   In this scenario, the peer host associates the multiple addresses
   with the SA pair between it and the multihomed host.  It may also
   undergo address verification procedures to transition the addresses
   to ACTIVE state.  For inbound data traffic, it may choose to use the
   addresses along with the SPI as selectors.  For outbound data
   traffic, it must choose among the available addresses of the
   multihomed host, considering the state of address verification
   [RFC8046] of each address, and also considering available information
   about whether an address is in a working state.

4.2.4.  Host Multihoming for Load Balancing

   A multihomed host may decide to set up new SA pairs corresponding to
   new addresses, for the purpose of load balancing.  The decision to
   load balance and the mechanism for splitting load across multiple SAs
   is out of scope of this document.  The scenario can be supported by
   sending the LOCATOR_SET parameter with one or more ESP_INFO
   parameters to initiate new ESP SAs.  To do this, the multihomed host
   sends a LOCATOR_SET with an ESP_INFO, indicating the request for a
   new SA by setting the OLD SPI value to zero and the NEW SPI value to
   the newly created incoming SPI.  A Locator Type of "1" is used to
   associate the new address with the new SPI.  The LOCATOR_SET
   parameter also contains a second Type "1" locator, Locator, that of the
   original address and SPI.  To simplify parameter processing and avoid
   explicit protocol extensions to remove locators, each LOCATOR_SET
   parameter MUST list all locators in use on a connection (a complete
   listing of inbound locators and SPIs for the host).  The multihomed
   host waits for a corresponding ESP_INFO (new outbound SA) from the
   peer and an ACK of its own UPDATE.  As in the mobility case, the peer
   host must perform an address verification before actively using the
   new address.

   Figure 2 illustrates this scenario.

     Multihomed Host                     Peer Host

              UPDATE(ESP_INFO, LOCATOR_SET, SEQ, [DIFFIE_HELLMAN])
        ----------------------------------->
              UPDATE(ESP_INFO, SEQ, ACK, [DIFFIE_HELLMAN,] ECHO_REQUEST)
        <-----------------------------------
              UPDATE(ACK, ECHO_RESPONSE)
        ----------------------------------->

               Figure 2: Host Multihoming for Load Balancing

   In multihoming scenarios, it is important that hosts receiving
   UPDATEs associate them correctly with the destination address used in
   the packet carrying the UPDATE.  When processing inbound LOCATOR_SETs
   that establish new security associations on an interface with
   multiple addresses, a host uses the destination address of the UPDATE
   containing the LOCATOR_SET as the local address to which the
   LOCATOR_SET plus ESP_INFO is targeted.  This is because hosts may
   send UPDATEs with the same (locator) IP address to different peer
   addresses -- this has the effect of creating multiple inbound SAs
   implicitly affiliated with different peer source addresses.

4.2.5.  Site Multihoming

   A host may have an interface that has multiple globally routable IP
   addresses.  Such a situation may be a result of the site having
   multiple upper Internet Service Providers, or just because the site
   provides all hosts with both IPv4 and IPv6 addresses.  The host
   should stay reachable at all or any subset of the currently available
   global routable addresses, independent of how they are provided.

   This case is handled the same as if there were different IP
   addresses, described above in Sections 4.2.3 and 4.2.4.  Note that a
   single interface may have addresses corresponding to site multihoming
   while the host itself may also have multiple network interfaces.

   Note that a host may be multihomed and mobile simultaneously, and
   that a multihomed host may want to protect the location of some of
   its interfaces while revealing the real IP address of some others.

   This document does not present additional site multihoming extensions
   to HIP; such extensions are for further study.

4.2.6.  Dual-Host Multihoming

   Consider the case in which both hosts are multihomed and would like
   to notify the peer of an additional address after the base exchange
   completes.  It may be the case that both hosts choose to simply
   announce the second address in a LOCATOR_SET parameter using an
   UPDATE message exchange.  It may also be the case that one or both
   hosts decide to ask for new SA pairs to be created using the newly
   announced address.  In the case that both hosts request this, the
   result will be a full mesh of SAs as depicted in Figure 3.  In such a
   scenario, consider that host1, which used address addr1a in the base
   exchange to set up SPI1a and SPI2a, wants to add address addr1b.  It
   would send an UPDATE with LOCATOR_SET (containing the address addr1b)
   to host2, using destination address addr2a, and a new ESP_INFO, and a
   new set of SPIs would be added between hosts 1 and 2 (call them SPI1b
   and SPI2b; not shown in the figure).  Next, consider host2 deciding
   to add addr2b to the relationship.  Host2 must select one of host1's
   addresses towards which to initiate an UPDATE.  It may choose to
   initiate an UPDATE to addr1a, addr1b, or both.  If it chooses to send
   to both, then a full mesh (four SA pairs) of SAs would exist between
   the two hosts.  This is the most general case; the protocol is
   flexible enough to accommodate this choice.

              -<- SPI1a --                         -- SPI2a ->-
      host1 <              > addr1a <---> addr2a <              > host2
              ->- SPI2a --                         -- SPI1a -<-

                             addr1b <---> addr2a  (second SA pair)
                             addr1a <---> addr2b  (third SA pair)
                             addr1b <---> addr2b  (fourth SA pair)

    Figure 3: Dual-Multihoming Case in which Each Host Uses LOCATOR_SET
                          to Add a Second Address

4.2.7.  Combined Mobility and Multihoming

   Mobile hosts may be simultaneously mobile and multihomed, i.e., have
   multiple mobile interfaces.  Furthermore, if the interfaces use
   different access technologies, it is fairly likely that one of the
   interfaces may appear stable (retain its current IP address) while
   some other(s) others may experience mobility (undergo IP address change).

   The use of LOCATOR_SET plus ESP_INFO should be flexible enough to
   handle most such scenarios, although more complicated scenarios have
   not been studied so far.

4.2.8.  Initiating the Protocol in R1, I2, or R2

   A Responder host MAY include a LOCATOR_SET parameter in the R1 packet
   that it sends to the Initiator.  This parameter MUST be protected by
   the R1 signature.  If the R1 packet contains LOCATOR_SET parameters
   with a new Preferred preferred locator, the Initiator SHOULD directly set the
   new Preferred preferred locator to status ACTIVE without performing address
   verification first, and it MUST send the I2 packet to the new
   Preferred
   preferred locator.  The I1 destination address and the new Preferred preferred
   locator may be identical.  All new non-preferred locators must still
   undergo address verification once the base exchange completes.  It is
   also possible for the host to send the LOCATOR_SET without any
   Preferred
   preferred bits set, in which case the exchange will continue as
   normal and the newly learned addresses will be in an UNVERIFIED state
   at the initiator.

            Initiator                                Responder

                              R1 with LOCATOR_SET
                  <-----------------------------------
   record additional addresses
   change Responder address
                     I2 sent to newly indicated preferred address
                  ----------------------------------->
                                                     (process normally)
                                  R2
                  <-----------------------------------
   (process normally, later verification of non-preferred locators)

                   Figure 4: LOCATOR_SET Inclusion in R1

   An Initiator MAY include one or more LOCATOR_SET parameters in the I2
   packet, independent of whether or not there was a LOCATOR_SET
   parameter in the R1.  These parameters MUST be protected by the I2
   signature.  Even if the I2 packet contains LOCATOR_SET parameters,
   the Responder MUST still send the R2 packet to the source address of
   the I2.  The new Preferred preferred locator, if set, SHOULD be identical to
   the I2 source address.  If the I2 packet contains LOCATOR_SET
   parameters, all new locators must undergo address verification as
   usual, and the ESP traffic that subsequently follows should use the
   Preferred
   preferred locator.

            Initiator                                Responder

                             I2 with LOCATOR_SET
                  ----------------------------------->
                                                     (process normally)
                                             record additional addresses
                       R2 sent to source address of I2
                  <-----------------------------------
   (process normally)

                   Figure 5: LOCATOR_SET Inclusion in I2

   The I1 and I2 may be arriving from different source addresses if the
   LOCATOR_SET parameter is present in R1.  In this case,
   implementations simultaneously using multiple pre-created R1s,
   indexed by Initiator IP addresses, may inadvertently fail the puzzle
   solution of I2 packets due to a perceived puzzle mismatch.  See, for
   instance, the example in Appendix A of [RFC7401].  As a solution, the
   Responder's puzzle indexing mechanism must be flexible enough to
   accommodate the situation when R1 includes a LOCATOR_SET parameter.

   Finally, the R2 may be used to carry the LOCATOR_SET parameter.  In
   this case, the LOCATOR_SET is covered by the HIP_MAC_2 and
   HIP_SIGNATURE.  Including LOCATOR_SET in R2 as opposed to R1 may have
   some advantages when a host prefers not to divulge additional
   locators until after the I2 is successfully processed.

   When the LOCATOR_SET parameter is sent in an UPDATE packet, the
   receiver will respond with an UPDATE acknowledgment.  When the
   LOCATOR_SET parameter is sent in an R1, I2, or R2 packet, the base
   exchange retransmission mechanism will confirm its successful
   delivery.

4.2.9.  Using LOCATOR_SETs across Addressing Realms

   It is possible for HIP associations to use these mechanisms to
   migrate their HIP associations and security associations from
   addresses in the IPv4 addressing realm to IPv6, or vice versa.  It
   may be possible for a state to arise in which both hosts are only
   using locators in different addressing realms, but in such a case,
   some type of mechanism for interworking between the different realms
   must be employed; such techniques are outside the scope of the
   present text.

4.3.  Interaction with Security Associations

   A host may establish any number of security associations (or SPIs)
   with a peer.  The main purpose of having multiple SPIs with a peer is
   to group the addresses into collections that are likely to experience
   fate sharing, or to perform load balancing.

   A basic property of HIP SAs is that the inbound IP address is not
   used to look up the incoming SA.  However, the use of different
   source and destination addresses typically leads to different paths,
   with different latencies in the network, and if packets were to
   arrive via an arbitrary destination IP address (or path) for a given
   SPI, the reordering due to different latencies may cause some packets
   to fall outside of the ESP anti-replay window.  For this reason, HIP
   provides a mechanism to affiliate destination addresses with inbound
   SPIs, when there is a concern that anti-replay windows might be
   violated.  In this sense, we can say that a given inbound SPI has an
   "affinity" for certain inbound IP addresses, and this affinity is
   communicated to the peer host.  Each physical interface SHOULD have a
   separate SA, unless the ESP anti-replay window is extended or
   disabled.

   Moreover, even when the destination addresses used for a particular
   SPI are held constant, the use of different source interfaces may
   also cause packets to fall outside of the ESP anti-replay window,
   since the path traversed is often affected by the source address or
   interface used.  A host has no way to influence the source interface
   on which a peer sends its packets on a given SPI.  A host SHOULD
   consistently use the same source interface and address when sending
   to a particular destination IP address and SPI.  For this reason, a
   host may find it useful to change its SPI or at least reset its ESP
   anti-replay window when the peer host readdresses.

5.  Processing Rules

   Basic processing rules for the LOCATOR_SET parameter are specified in
   [RFC8046].  This document focuses on multihoming-specific rules.

5.1.  Sending LOCATOR_SETs

   The decision of when to send a LOCATOR_SET, and which addresses to
   include, is a local policy issue.  [RFC8046] recommends that a host
   "send a LOCATOR_SET whenever it recognizes a change of its IP
   addresses in use on an active HIP association and [when it] assumes
   that the change is going to last at least for a few seconds."  It is
   possible to delay the exposure of additional locators to the peer,
   and to send data from previously unannounced locators, as might arise
   in certain mobility or multihoming situations.

   When a host decides to inform its peers about changes in its IP
   addresses, it has to decide how to group the various addresses with
   SPIs.  If hosts are deployed in an operational environment in which
   HIP-aware NATs and firewalls (that may perform parameter inspection)
   exist, and different such devices may exist on different paths, hosts
   may take that knowledge into consideration about how addresses are
   grouped, and may send the same LOCATOR_SET in separate UPDATEs on the
   different paths.  However, more detailed guidelines about how to
   operate in the presence of such HIP-aware NATs and firewalls are a
   topic for further study.  Since each SPI is associated with a
   different security association, the grouping policy may also be based
   on ESP anti-replay protection considerations.  In the typical case,
   simply basing the grouping on actual kernel-level physical and
   logical interfaces may be the best policy.  The grouping policy is
   outside of the scope of this document.

   Locators corresponding to tunnel interfaces (e.g., IPsec tunnel
   interfaces or Mobile IP home addresses) or other virtual interfaces
   MAY be announced in a LOCATOR_SET, but implementations SHOULD avoid
   announcing such locators as preferred locators if more direct paths
   may be obtained by instead preferring locators from non-tunneling
   interfaces if such locators provide a more direct path to the HIP
   peer.

   [RFC8046] specifies that hosts MUST NOT announce broadcast or
   multicast addresses in LOCATOR_SETs.  Link-local addresses MAY be
   announced to peers that are known to be neighbors on the same link,
   such as when the IP destination address of a peer is also link local.
   The announcement of link-local addresses in this case is a policy
   decision; link-local addresses used as Preferred preferred locators will create
   reachability problems when the host moves to another link.  In any
   case, link-local addresses MUST NOT be announced to a peer unless
   that peer is known to be on the same link.

   Once the host has decided on the groups and assignment of addresses
   to the SPIs, it creates a LOCATOR_SET parameter that serves as a
   complete representation of the addresses and associated SPIs intended
   for active use.  We now describe a few cases introduced in Section 4.
   We assume that the Traffic Type for each locator is set to "0" (other
   values for Traffic Type may be specified in documents that separate
   the HIP control plane from data-plane traffic).  Other mobility and
   multihoming cases are possible but are left for further
   experimentation.

   1.  Host multihoming (addition of an address).  We only describe the
       simple case of adding an additional address to a (previously)
       single-homed, non-mobile host.  The host MAY choose to simply
       announce this address to the peer, for fault tolerance.  To do
       this, the multihomed host creates a LOCATOR_SET parameter
       including the existing address and SPI as a Type "1" Locator, and
       the new address as a Type "0" Locator.  The host sends this in an
       UPDATE message with the SEQ parameter, which is acknowledged by
       the peer.

   2.  The host MAY set up a new SA pair between this new address and an
       address of the peer host.  To do this, the multihomed host
       creates a new inbound SA and creates a new SPI.  For the outgoing
       UPDATE message, it inserts an ESP_INFO parameter with an OLD SPI
       field of "0", a NEW SPI field corresponding to the new SPI, and a
       KEYMAT Index as selected by local policy.  The host adds to the
       UPDATE message a LOCATOR_SET with two Type "1" Locators: the
       original address and SPI active on the association, and the new
       address and new SPI being added (with the SPI matching the NEW
       SPI contained in the ESP_INFO).  The Preferred preferred bit SHOULD be set
       depending on the policy to tell the peer host which of the two
       locators is preferred.  The UPDATE also contains a SEQ parameter
       and optionally a DIFFIE_HELLMAN parameter and follows rekeying
       procedures with respect to this new address.  The UPDATE message
       SHOULD be sent to the peer's Preferred preferred address with a source
       address corresponding to the new locator.

   The sending of multiple LOCATOR_SETs is unsupported.  Note that the
   inclusion of LOCATOR_SET in an R1 packet requires the use of Type "0"
   locators
   Locators since no SAs are set up at that point.

5.2.  Handling Received LOCATOR_SETs

   A host SHOULD be prepared to receive a LOCATOR_SET parameter in the
   following HIP packets: R1, I2, R2, and UPDATE.

   This document describes sending both ESP_INFO and LOCATOR_SET
   parameters in an UPDATE.  The ESP_INFO parameter is included when
   there is a need to rekey or key a new SPI and can otherwise be
   included for the possible benefit of HIP-aware middleboxes.  The
   LOCATOR_SET parameter contains a complete map of the locators that
   the host wishes to make or keep active for the HIP association.

   In general, the processing of a LOCATOR_SET depends upon the packet
   type in which it is included.  Here, we describe only the case in
   which ESP_INFO is present and a single LOCATOR_SET and ESP_INFO are
   sent in an UPDATE message; other cases are for further study.  The
   steps below cover each of the cases described in Section 5.1.

   The processing of ESP_INFO and LOCATOR_SET parameters is intended to
   be modular and support future generalization to the inclusion of
   multiple ESP_INFO and/or multiple LOCATOR_SET parameters.  A host
   SHOULD first process the ESP_INFO before the LOCATOR_SET, since the
   ESP_INFO may contain a new SPI value mapped to an existing SPI, while
   a Type "1" locator Locator will only contain a reference to the new SPI.

   When a host receives a validated HIP UPDATE with a LOCATOR_SET and
   ESP_INFO parameter, it processes the ESP_INFO as follows.  The
   ESP_INFO parameter indicates whether an SA is being rekeyed, created,
   deprecated, or just identified for the benefit of middleboxes.  The
   host examines the OLD SPI and NEW SPI values in the ESP_INFO
   parameter:

   1.  (no rekeying) If the OLD SPI is equal to the NEW SPI and both
       correspond to an existing SPI, the ESP_INFO is gratuitous
       (provided for middleboxes), and no rekeying is necessary.

   2.  (rekeying) If the OLD SPI indicates an existing SPI and the NEW
       SPI is a different non-zero value, the existing SA is being
       rekeyed and the host follows HIP ESP rekeying procedures by
       creating a new outbound SA with an SPI corresponding to the NEW
       SPI, with no addresses bound to this SPI.  Note that locators in
       the LOCATOR_SET parameter will reference this new SPI instead of
       the old SPI.

   3.  (new SA) If the OLD SPI value is zero and the NEW SPI is a new
       non-zero value, then a new SA is being requested by the peer.
       This case is also treated like a rekeying event; the receiving
       host must create a new SA and respond with an UPDATE ACK.

   4.  (deprecating the SA) If the OLD SPI indicates an existing SPI and
       the NEW SPI is zero, the SA is being deprecated and all locators
       uniquely bound to the SPI are put into the DEPRECATED state.

   If none of the above cases apply, a protocol error has occurred and
   the processing of the UPDATE is stopped.

   Next, the locators in the LOCATOR_SET parameter are processed.  For
   each locator listed in the LOCATOR_SET parameter, check that the
   address therein is a legal unicast or anycast address.  That is, the
   address MUST NOT be a broadcast or multicast address.  Note that some
   implementations MAY accept addresses that indicate the local host,
   since it may be allowed that the host runs HIP with itself.

   For each Type "1" address listed in the LOCATOR_SET parameter, the
   host checks whether the address is already bound to the SPI
   indicated.  If the address is already bound, its lifetime is updated.
   If the status of the address is DEPRECATED, the status is changed to
   UNVERIFIED.  If the address is not already bound, the address is
   added, and its status is set to UNVERIFIED.  If there exist remaining
   addresses corresponding to the SPI that were NOT listed in the
   LOCATOR_SET parameter, the host sets the status of such addresses to
   DEPRECATED.

   For each Type "0" address listed in the LOCATOR_SET parameter, if the
   status of the address is DEPRECATED, or the address was not
   previously known, the status is changed to UNVERIFIED.  The host MAY
   choose to associate this address with one or more SAs.  The
   association with different SAs is a local policy decision, unless the
   peer has indicated that the address is Preferred, preferred, in which case the
   address should be put into use on an SA that is prioritized in the
   security policy database.

   As a result, at the end of processing, the addresses listed in the
   LOCATOR_SET parameter have a state of either UNVERIFIED or ACTIVE,
   and any old addresses on the old SA not listed in the LOCATOR_SET
   parameter have a state of DEPRECATED.

   Once the host has processed the locators, if the LOCATOR_SET
   parameter contains a new Preferred preferred locator, the host SHOULD initiate
   a change of the Preferred preferred locator.  This requires that the host first
   verifies reachability of the associated address and only then changes
   the Preferred preferred locator; see Section 5.4.

   If a host receives a locator with an unsupported Locator Type, and
   when such a locator is also declared to be the Preferred preferred locator for
   the peer, the host SHOULD send a NOTIFY error with a Notify Message
   Type of LOCATOR_TYPE_UNSUPPORTED, with the Notification Data field
   containing the locator(s) that the receiver failed to process.
   Otherwise, a host MAY send a NOTIFY error if a (non-preferred)
   locator with an unsupported Locator Type is received in a LOCATOR_SET
   parameter.

5.3.  Verifying Address Reachability

   Address verification is defined in [RFC8046].

   When address verification is in progress for a new Preferred preferred locator,
   the host SHOULD select a different locator listed as ACTIVE, if one
   such locator is available, to continue communications until address
   verification completes.  Alternatively, the host MAY use the new
   Preferred
   preferred locator while in UNVERIFIED status to the extent Credit-
   Based Authorization permits.  Credit-Based Authorization is explained
   in [RFC8046].  Once address verification succeeds, the status of the
   new Preferred preferred locator changes to ACTIVE.

5.4.  Changing the Preferred Locator

   A host MAY want to change the Preferred preferred outgoing locator for
   different reasons, e.g., because traffic information or ICMP error
   messages indicate that the currently used preferred address may have
   become unreachable.  Another reason may be due to receiving a
   LOCATOR_SET parameter that has the Preferred preferred bit set.

   To change the Preferred preferred locator, the host initiates the following
   procedure:

   1.  If the new Preferred preferred locator has ACTIVE status, the Preferred preferred
       locator is changed and the procedure succeeds.

   2.  If the new Preferred preferred locator has UNVERIFIED status, the host
       starts to verify its reachability.  The host SHOULD use a
       different locator listed as ACTIVE until address verification
       completes if one such locator is available.  Alternatively, the
       host MAY use the new Preferred preferred locator, even though in UNVERIFIED
       status, to the extent Credit-Based Authorization permits.  Once
       address verification succeeds, the status of the new Preferred preferred
       locator changes to ACTIVE, and its use is no longer governed by
       Credit-Based Authorization.

   3.  If the peer host has not indicated a preference for any address,
       then the host picks one of the peer's ACTIVE addresses randomly
       or according to policy.  This case may arise if, for example,
       ICMP error messages that deprecate the Preferred preferred locator arrive,
       but the peer has not yet indicated a new Preferred preferred locator.

   4.  If the new Preferred preferred locator has DEPRECATED status and there is
       at least one non-deprecated address, the host selects one of the
       non-deprecated addresses as a new Preferred preferred locator and
       continues.  If the selected address is UNVERIFIED, the address
       verification procedure described above will apply.

6.  Security Considerations

   This document extends the scope of host mobility solutions defined in
   [RFC8046] to also include host multihoming, and as a result, many of
   the same security considerations for mobility also pertain to
   multihoming.  In particular, [RFC8046] describes how HIP host
   mobility is resistant to different types of impersonation attacks and
   denial-of-service (DoS) attacks.

   The security considerations for this document are similar to those of
   [RFC8046] because the strong authentication capabilities for mobility
   also carry over to end-host multihoming.  [RFC4218] provides a threat
   analysis for IPv6 multihoming, and the remainder of this section
   first describes how HIP host multihoming addresses those previously
   described threats, and then it discusses some additional security
   considerations.

   The high-level threats discussed in [RFC4218] involve redirection
   attacks for the purposes of packet recording, data manipulation, and
   availability.  There are a few types of attackers to consider: on-
   path attackers, off-path attackers, and malicious hosts.

   [RFC4218] also makes the comment that in identifier/locator split
   solutions such as HIP, application security mechanisms should be tied
   to the identifier, not the locator, and attacks on the identifier
   mechanism and on the mechanism binding locators to the identifier are
   of concern.  This document does not consider the former issue
   (application-layer security bindings) to be within scope.  The latter
   issue (locator bindings to identifier) is directly addressed by the
   cryptographic protections of the HIP protocol, in that locators
   associated to an identifier are listed in HIP packets that are signed
   using the identifier key.

   Section 3.1 of [RFC4218] lists several classes of security
   configurations in use in the Internet.  HIP maps to the fourth
   (strong identifier) and fifth ("leap-of-faith") categories, the
   latter being associated with the optional opportunistic mode of HIP
   operation.  The remainder of Section 3 describes existing security
   problems in the Internet and comments that the goal of a multihoming
   solution is not to solve them specifically but rather not to make any
   of them worse.  HIP multihoming should not increase the severity of
   the identified risks.  One concern for both HIP mobility and
   multihoming is the susceptibility of the mechanisms to misuse
   flooding-based redirections due to a malicious host.  The mechanisms
   described in [RFC8046] for address verification are important in this
   regard.

   Regarding the new types of threats introduced by multihoming
   (Section 4 of [RFC4218]), HIP multihoming should not introduce new
   concerns.  Classic and premeditated redirection are prevented by the
   strong authentication in HIP messages.  Third-party DoS attacks are
   prevented by the address verification mechanism.  Replay attacks can
   be avoided via use of replay protection in ESP SAs.  In addition,
   accepting packets from unknown locators is protected by either the
   strong authentication in the HIP control packets or by the ESP-based
   encryption in use for data packets.

   The HIP mechanisms are designed to limit the ability to introduce DoS
   on the mechanisms themselves (Section 7 of [RFC4218]).  Care is taken
   in the HIP base exchange to avoid creating state or performing much
   work before hosts can authenticate one another.  A malicious host
   involved in HIP multihoming with another host might attempt to misuse
   the mechanisms for multihoming by, for instance, increasing the state
   required or inducing a resource limitation attack by sending too many
   candidate locators to the peer host.  Therefore, implementations
   supporting the multihoming extensions should consider avoiding
   accepting large numbers of peer locators and rate limiting any UPDATE
   messages being exchanged.

   The exposure of a host's IP addresses through HIP mobility and
   multihoming extensions may raise the following privacy concern.  The
   administrator of a host may be trying to hide its location in some
   context through the use of a VPN or other virtual interfaces.
   Similar privacy issues also arise in other frameworks such as WebRTC
   and are not specific to HIP.  Implementations SHOULD provide a
   mechanism to allow the host administrator to block the exposure of
   selected addresses or address ranges.

   Finally, some implementations of VPN tunneling have experienced
   instances of 'leakage' of flows that were intended to have been
   protected by a security tunnel but are instead sent in the clear,
   perhaps because some of the addresses used fall outside of the range
   of addresses configured for the tunnel in the security policy or
   association database.  Implementors are advised to take steps to
   ensure that the usage of multiple addresses between hosts does not
   cause accidental leakage of some data session traffic outside of the
   ESP-protected envelope.

7.  References

7.1.  Normative References

   [RFC2119]  Bradner, S., "Key words for use in RFCs to Indicate
              Requirement Levels", BCP 14, RFC 2119,
              DOI 10.17487/RFC2119, March 1997,
              <http://www.rfc-editor.org/info/rfc2119>.

   [RFC6724]  Thaler, D., Ed., Draves, R., Matsumoto, A., and T. Chown,
              "Default Address Selection for Internet Protocol Version 6
              (IPv6)", RFC 6724, DOI 10.17487/RFC6724, September 2012,
              <http://www.rfc-editor.org/info/rfc6724>.

   [RFC7401]  Moskowitz, R., Ed., Heer, T., Jokela, P., and T.
              Henderson, "Host Identity Protocol Version 2 (HIPv2)",
              RFC 7401, DOI 10.17487/RFC7401, April 2015,
              <http://www.rfc-editor.org/info/rfc7401>.

   [RFC7402]  Jokela, P., Moskowitz, R., and J. Melen, "Using the
              Encapsulating Security Payload (ESP) Transport Format with
              the Host Identity Protocol (HIP)", RFC 7402,
              DOI 10.17487/RFC7402, April 2015,
              <http://www.rfc-editor.org/info/rfc7402>.

   [RFC8046]  Henderson, T., Vogt, C., and J. Arkko, "Host Mobility with
              the Host Identity Protocol", RFC 8046,
              DOI 10.17487/RFC8046, December 2016, February 2017,
              <http://www.rfc-editor.org/info/rfc8046>.

7.2.  Informative References

   [RFC4218]  Nordmark, E. and T. Li, "Threats Relating to IPv6
              Multihoming Solutions", RFC 4218, DOI 10.17487/RFC4218,
              October 2005, <http://www.rfc-editor.org/info/rfc4218>.

   [RFC4303]  Kent, S., "IP Encapsulating Security Payload (ESP)",
              RFC 4303, DOI 10.17487/RFC4303, December 2005,
              <http://www.rfc-editor.org/info/rfc4303>.

   [RFC5533]  Nordmark, E. and M. Bagnulo, "Shim6: Level 3 Multihoming
              Shim Protocol for IPv6", RFC 5533, DOI 10.17487/RFC5533,
              June 2009, <http://www.rfc-editor.org/info/rfc5533>.

Acknowledgments

   This document contains content that was originally included in RFC
   5206.  Pekka Nikander and Jari Arkko originated RFC 5206, and
   Christian Vogt and Thomas Henderson (editor) later joined as
   coauthors.  Also in RFC 5206, Greg Perkins contributed the initial
   draft of the security section, and Petri Jokela was a coauthor of the
   initial individual submission.

   The authors thank Miika Komu, Mika Kousa, Jeff Ahrenholz, and Jan
   Melen for many improvements to the document.  Concepts from a paper
   on host multihoming across address families, by Samu Varjonen, Miika
   Komu, and Andrei Gurtov, contributed to this revised specification.

Authors' Addresses

   Thomas R. Henderson (editor)
   University of Washington
   Campus Box 352500
   Seattle, WA
   United States of America

   Email: tomhend@u.washington.edu

   Christian Vogt
   Independent
   3473 North First Street
   San Jose, CA  95134
   United States of America

   Email: mail@christianvogt.net

   Jari Arkko
   Ericsson
   Jorvas,  FIN-02420
   Finland

   Phone: +358 40 5079256
   Email: jari.arkko@piuha.net