[Search] [txt|pdf|bibtex] [Tracker] [Email] [Nits]

Versions: 00 01 02 03 04 05                                             
Internet Engineering Task Force                         Francis Dupont
INTERNET DRAFT                                           ENST Bretagne
Expires in August 2002                                   February 2002

              How to make IPsec more mobile IPv6 friendly


Status of this Memo

   This document is an Internet Draft and is in full conformance with
   all provisions of Section 10 of RFC 2026.

   This document is an Internet-Draft.  Internet-Drafts are working
   documents of the Internet Engineering Task Force (IETF), its
   areas, and its working groups.  Note that other groups may also
   distribute working documents as Internet-Drafts.

   Internet-Drafts are draft documents valid for a maximum of six
   months and may be updated, replaced, or obsoleted by other
   documents at any time.  It is inappropriate to use Internet-
   Drafts as reference material or to cite them other than as
   "work in progress."

   The list of current Internet Drafts can be accessed at

   The list of Internet-Draft Shadow Directories can be accessed at

   Distribution of this memo is unlimited.


   IPsec specifications [1-6] does not work well with Mobile IPv6 [7]
   and with any mobility device based on addresses [8] even if HIP [9]
   should change this regrettable situation.

   But many problems come directly from bad or dubious interpretations
   of IPsec specifications, so this document presents many points
   where many current implementations can be improved (i.e. become
   more Mobile IPv6 friendly).

draft-dupont-ipsec-mipv6-00.txt                               [Page 1]

INTERNET-DRAFT          IPsec more MIPv6 friendly             Feb 2002

1. Introduction

   This document assumes the reader knows IPsec (i.e. [1-6]) but not
   Mobile IPv6, so this section explains how Mobile IPv6 works.

   In Mobile IPv6, each Mobile Node (MN) is always identified by
   its Home Address (H@), regardless of its current point of
   attachment to the Internet. While situated away from its home,
   a MN is also associated with a Care-of Address (Co@), which
   provides information about the mobile node's current location.
   IPv6 packets from a Correspondent Node (CN) addressed to a MN's H@
   are transparently routed to its Co@.

   A MN has a special CN, the Home Agent (HA), which is used to
   intercept packets addressed to the MN's H@ on the home link and to
   forward them to the MN at its current Co@ through a tunnel.

   End-to-end communications between a MN and a CN can use one of
   these 3 modes:
    - bidirectional tunneling with the HA:
       MN => HA -> CN (=> denotes an encapsulation)
       CN -> HA => MN
      The CN knows only the MN's H@ and in fact even does not know
      the MN is mobile. This mode is very safe but not optimized
      at all.
    - triangular routing:
       MN ~> CN (~> denotes the use of an extension header)
       CN -> HA => MN
      The MN uses a Home Address Option (HAO): it puts its Co@ which
      is topologically correct into the source address field of the
      IPv6 header, and puts its H@ in the HAO.
    - optimized routing:
       MN ~> CN
       CN ~> MN
      The MN uses a Home Address Option for packets to the CN,
      the CN uses a Routing Header (RH) and puts the MN's Co@ in the
      destination address field of the IPv6 header, and puts the
      MN's H@ in the RH. This mode raises many security concerns,
      mainly about the mobility signaling, but is very efficient.
   This use of HAO and RH are in fact degenerated tunnels as shown
   by the Deering/Zill tunneling document [10], i.e. they can be
   considered as IPv6 in IPv6 tunnels where two addresses in
   the outer and inner IPv6 headers are redundant so deleted.

draft-dupont-ipsec-mipv6-00.txt                               [Page 2]

INTERNET-DRAFT          IPsec more MIPv6 friendly             Feb 2002

   The term node in MN is a bit misleading: mobile routers are not
   considered by current Mobile IPv6 specifications and communications
   in this document are always end-to-end.

   The association between a Co@ and the H@ is named "binding" and
   is cached by CNs. Bindings are managed (i.e. created, changed
   and deleted) by signaling messages named Binding Updates (BUs).

2. IPsec Architecture and Mobile IPv6

   IPsec defines two modes, transport and tunnel modes.
   As mobile IPv6 itself is based on real or degenerated tunneling,
   there are three possible basic interactions: transport mode
   after Mobile IPv6 tunneling, transport mode before Mobile IPv6
   tunneling, combination between tunnel mode and Mobile IPv6

2.1 Transport Mode After Mobile IPv6

   This case is defined only when Mobile IPv6 uses real tunneling,
   i.e. in current specifications between the MN and its HA.
   As the IPv6 header (and addresses!) viewed by IPsec is the outer
   one, in general this case has no interest (the MN's outer address
   is a transient Co@, the interesting one, the H@, is in the inner

2.2 Transport Mode Before Mobile IPv6

   In this case the IPsec transform is applied to the payload of
   an ordinary packet between the MN and the CN, using for the MN
   its static/long term H@, and *after* Mobile IPv6 may add its
   extension header corresponding to its mode:
    - bidirectional tunneling is perfectly transparent: IPsec and
      Mobile IPv6 don't interfere. The same consideration
      applies to the HA=>MN tunnel.
    - triangular routing is more interesting and gives different
      results for the Authentication Header (AH [2]) and the
      Encapsulating Security Payload (ESP [3]):
       * AH authenticates both Co@ (in the IPv6 header) and the H@
         (in the HAO)
       * ESP with authentication will reject fake packets because
         the attacker can't know the authentication key.
      So with authentication the only possible attack is a denial
      of services from an attacker who knows the SPI and is likely
      be able to inject fake traffic without HAO. So this is an
      intrinsic IPsec thread.

draft-dupont-ipsec-mipv6-00.txt                               [Page 3]

INTERNET-DRAFT          IPsec more MIPv6 friendly             Feb 2002

      *RECOMMENDATION A: packets with a HAO matching an IPsec SA
      *providing authentication (i.e. AH or ESP with non-null
      *authentication) MUST be accepted (i.e. the HAO considered as
      *verifiable) and the HAO MUST be considered as verified [11]
      *after successful IPsec processing.
    - optimized routing is like triangular routing for the MN~>CN
      way with, in addition to recommendation A, the common way
      to verify HAO: binding cache entry check. Symmetrically
      IPsec adds nothing for the RH verification because the MN
      has already all the interesting informations.
    - optimized routing between two MNs is an interesting case
      not yet fully addressed by Mobile IPv6 specifications [7]:
      RH and HAO have more overhead than a plain tunnel so a
      combined tunnel mode with Mobile IPv6 becomes very
      interesting in this special case...

2.3 Tunnel Mode not combined with Mobile IPv6

   Not combined means there are two overhead, one for IPsec tunneling,
   another for Mobile IPv6 extra headers, when obviously one
   encapsulation provides enough addresses (two sources and two
   destinations) for Mobile IPv6!
   *RECOMMENDATION B: IPsec in tunnel mode and Mobile IPv6 SHOULD
   *be combined in order to avoid to add their overheads.

2.4 Addresses of SAs

   SAs can be characterized by:
    - one address (destination in IPsec inbound processing [1])
    - two addresses (source and destination selectors [1])
    - three addresses (source, destination and proxy in PF_KEY [12])
    - four addresses (in tunnel mode)
   This is a bit confusing and gives security holes and/or extra
   checks on addresses which are highly Mobile IPv6 unfriend.

   Transport mode is easy because there are always exactly two
   addresses. For instance in inbound processing the destination
   address is used for the SA lookup and the source address MUST
   be checked ([1], section 5.2.1). So this document will assume
   the tunnel mode is used.

draft-dupont-ipsec-mipv6-00.txt                               [Page 4]

INTERNET-DRAFT          IPsec more MIPv6 friendly             Feb 2002

   In general, the SADB is not designed to be managed directly
   and/or itself, i.e. without the SPD. Addresses are handled
   by the SPD with a pair of selectors which characterized the
   source and the destination of the traffic which receives IPsec
   protection. SPD entries specify the type of IPsec processing
   (for instance one type is the bypassing of IPsec: this is needed
   by IKE for its own messages [1]) and the parameters of SAs
   to use or to build (using SADB_ACQUIRE PF_KEY messages [12]).

   IKE (and its successor(s)) is designed to build SAs per pairs
   so IPsec implementations using PF_KEY follow (or should follow)
   for tunnel mode SAs this interpretation:
    - the source and the destination addresses are plain addresses
      in the general case and designed the end nodes of the tunnel
      (i.e. there are the outer header addresses).
    - the inner source address is the proxy address (and exists
      only in tunnel mode). This can be something else than a plain
      address (i.e. for instance a prefix) but not in the Mobile IPv6
      case. The check can be performed after each IPsec inbound
      processing [1] or at the SPD check (not the specified way
      by the final result is the same).
      *RECOMMENDATION C1: the source address checked after each
      *IPsec inbound processing against the SA selector MUST be
      *the inner header source address.
    - the selection of the traffic to be processed is handled
      by SPD entries. This includes the future inner addresses
      in outbound processing: guidelines are to use the inbound
      processing rules for SADB design, the outbound processing
      rules for SPD design and to complete by symmetry (with the
      funny (?) side-effect that source and destination roles can
      be reversed).

   RFC 2401 [1] has detailed rules about the outer source address
   but they are commonly misunderstood: to check it gives no
   extra security because when (where!) a attacker can get the SPI
   he can inject fake traffic too. But this check harm near all
   mobility mechanisms based on addresses, even nomadism aka
   the "road warrior" case.
   *RECOMMENDATION C2: the outer source address in tunnel mode
   *MUST NOT be checked after or before IPsec inbound processing.
   This recommendation doesn't apply to the SPD checking, i.e.
   step 4 of RFC 2401 [1] section 5.2.1.
   This recommendation by itself doesn't solve the problem of the
   other SA of the pair: the MN can change its Co@ and continue
   to use the SA to the CN but the other way/SA will work only
   when the other SA will change or rebuild with the new Co@
   as the destination.

draft-dupont-ipsec-mipv6-00.txt                               [Page 5]

INTERNET-DRAFT          IPsec more MIPv6 friendly             Feb 2002

   BTW RFC 2401 [1] specifies IPv6 in IPv4 and IPv4 in IPv6 tunnels
   and these tunnels are taken into account in PF_KEY [12] so:
   *RECOMMENDATION D: dual stack (i.e. IPv4 and IPv6) IPsec
   *implementations MUST support IPvX in IPvY tunnel modes with
   *any X and Y, including cases where X != Y.

2.5 Combined Tunnel Mode with Mobile IPv6 (Standard Case)

   When IPsec in tunnel mode is combined with Mobile IPv6 there
   is one encapsulation with fixed/H@ in the inner header and
   transient/Co@ in the outer header, the opposite of the
   common tunnel mode usage between two security gateways.
   The CN can be itself a MN: only one detail not directly
   related to IPsec changes: mobility has to support the
   simultaneous movement of both peers (handled by fallback
   through a HA for instance in Mobile IPv6, cf. [7]).

   Between two movements the IPsec tunnels are not very special,
   they look like end-to-end IPsec tunnels between two peers.
   The only unusual detail is the outer and inner addresses can be
   different (they are if a MN is not at home) but this is
   an issue for IKE.

   The interesting case is what happens when a Co@ changes:
   the MN should send a BU to the CN which, according to
   recommendation C, must not be filtered out because the
   Co@ is not the same.
   *RECOMMENDATION E1: BU protected by an IPsec SA providing
   *authentication MUST be considered as authenticated.
   *RECOMMENDATION E2: in the E1 case, all BU parameters MUST
   *be covered by the authentication, specially when the
   *authentication is provided by an ESP transform, the new Co@
   *MUST be covered, for instance using an alternate Co@ suboption.

   As explained before, to deal with the BU is not enough for
   the CN~>MN way. For instance, the Binding Acknowledgment (BA)
   can be protected only when the reversed SA is updated or rebuilt.
   *RECOMMENDATION F: mobility signaling and IPsec SA management
   *direct cooperation SHOULD be considered (i.e. development of
   *this kind of mechanisms encouraged).

draft-dupont-ipsec-mipv6-00.txt                               [Page 6]

INTERNET-DRAFT          IPsec more MIPv6 friendly             Feb 2002

2.6 Combined Tunnel Mode with Mobile IPv6 (Home Agent Case)

   This section doesn't consider traffic from the HA itself
   which is handled by routing optimization as for a standard CN,
   but is about the MN<=>HA tunnels. This document assumes that
   the tunnel is bidirectional even the Mobile IPv6 is not (yet)
   very clear about this.

   The addresses used for these tunnels are very simple:
    - on the MN side, inner header address is the H@, outer a Co@.
    - on the HA side, inner header address is any valid address
     (unicast address when used as a source address) and the
     outer address is the HA address [7].
   IPsec considerations are near the same than for standard CNs,
   in fact things are simpler because one can assume the HA is
   never mobile. Recommendation F applies but is not useful when
   no IPsec SA exists, i.e. when a MN boots in visit: this will
   be the special case in IKE considerations (next section).

3. IKE and Mobile IPv6

   There are three basic issues:
    - how to handle the multiple addresses of a MN? In the phase
      one? In a phase two?
    - how to establish SAs between a MN and a standard CN?
    - how to establish SAs between a MN from a foreign link
      and its HA the first time?
   This document uses the name IKE [6] for the IPsec DOI [4]
   and ISAKMP [5] framework too.

3.1 IKE and Identities (Phase One)

   In the phase one identities (IDii and IDir) design the peers so
   subnet and range identities can not be used. This document
   assumes phase one with digital signatures using a X.509 style
   of certificates but most of the considerations applies to
   public key authentications too.

draft-dupont-ipsec-mipv6-00.txt                               [Page 7]

INTERNET-DRAFT          IPsec more MIPv6 friendly             Feb 2002

   In phase one a peer is identified by its address used for the
   transport of IKE messages and its identity payload. Identities
   in the general meaning can be present in certificates too but
   all cases are not equivalent:
    - the Identity must be related to the certificate:
      *RECOMMENDATION G: the Identity payload presented by the peer
      *MUST be verified, for instance certificates are used, the
      *Identity and the subject or an alternative subject of the
      *certificate associated to the signature MUST match.
    - if the Identity is an address, it must be the right address.
      *RECOMMENDATION H: if the Identity payload presented by the
      *peer is an address, it MUST be the same address than the one
      *used for transport of IKE messages.
    - same for addresses used as (alternative) subjects:
      *RECOMMENDATION I: if an address is used as the subject or
      *an alternative subject of the certificate associated to the
      *signature, the address used for transport of IKE messages
      *SHOULD match the subject or an alternative subject.
      "SHOULD match" means the default policy is to perform the check.
    - last about Identity/address checks:
      *RECOMMENDATION J: the case where the certificate associated
      *to the signature has no address subject or alternative subject
      *SHOULD be considered as a special case by policies, for instance
      *the policy MAY specify particular constraints for this case.
    - of course, this can be a bit annoying for MNs which must use
      in some cases their Co@ for transport of IKE messages so:
      *RECOMMENDATION K: the addition of a Home Address Identity
      *which should allow strict application of previous I and J
      *recommendations in all cases for a MN's peer SHOULD be
      *considered in IKE and its successor(s).

3.2 IKE and Identities (Phase Two)

   In the phase two (aka quick mode) identities (IDci and IDcr) are
   optional and design the policy rule (in the SPD or in the IKE
   software configuration) to apply:
    - in tunnel mode the peer addresses will be the outer header
      addresses, and identities can denote the traffic selector part
      of the policy rule.

draft-dupont-ipsec-mipv6-00.txt                               [Page 8]

INTERNET-DRAFT          IPsec more MIPv6 friendly             Feb 2002

    - without identities the SA pairs will be applied to all traffic
      not matched by a more specific policy between the peers using
      the same addresses than in IKE messages. This is ambiguous so:
      *RECOMMENDATION L1: when the phase one and phases two are
      *allowed to use different addresses, identity payloads MUST be
      *used in phases two.
    - in tunnel mode there is an ambiguity about the peer addresses:
      *RECOMMENDATION L2: when the phase one and phases two are
      *allowed to use different addresses, the peer addresses used
      *in a phase 2 context MUST be the addresses used for the
      *transport of IKE messages of this phase 2.
    - junk identities are not useful:
      *RECOMMENDATION M: identity payloads used in phase two SHOULD
      *denote clearly address sets.
    - IKE [6] explicitly provides a "proxy case" usable for mobility:
      *RECOMMENDATION N: the policy MAY authorize the establishment of
      *a transport mode SA pair using an address identity payload
      *which doesn't match the address used for transport of IKE
      *messages. This authorization SHOULD be based on parameters
      *provided in the phase one authentication, for instance the
      *phase one peer identity and certificate.

3.3 IKE and Mobile IPv6 (Standard Case)

   If the MN has the choice between using its H@ or a Co@ for IKE
   exchanges, only the first choice makes sense with a standard CN:
    - to use the Co@ is more complex and can require a new phase one
      with each peer after a movement.
    - when the CN is the initiator it always use the H@.
   Note the IKE software should not even know the node is mobile...
   For the same reason, the SA pairs should use the H@ as the MN
   address, giving an IPsec transform before Mobility case:
   *RECOMMENDATION O: with a standard CN, the MN SHOULD ignore
   *it is a mobile node and SHOULD use its H@ for all IKE exchanges
   *and for its own address in the SA pairs. To avoid both IPsec
   *and Mobile IPv6 overheads, it SHOULD negotiate the transport mode.
   Transport mode was designed for end-to-end communications, IMHO
   this recommendation should be written with MUSTs!

draft-dupont-ipsec-mipv6-00.txt                               [Page 9]

INTERNET-DRAFT          IPsec more MIPv6 friendly             Feb 2002

3.4 IKE and Mobile IPv6 (Home Agent Case)

   Recommendation O doesn't work with the HA because when the MN
   boots in visit it can use its H@ only after processing of the
   first BU by the HA. This BU MUST be protected so it can not
   be protected by IPsec (trivial bootstrap problem).

   In fact there are two possible MN-HA SA pairs:
    - a SA pair for BU/BA exchange protection.
    - a SA pair for the MN<=>HA tunnels.

   The first SA pair is a bit hairy to establish because the MN
   can only use its Co@ in some circumstances:
   *RECOMMENDATION P: if BU/BA exchanges between the MN and the HA
   *are protected by an IPsec SA pairs, establishment of this SA
   *pair MUST be allowed using a Co@ for all IKE messages.
   The detailed requirements are:
    - an API should give a suitable Co@ for communications with
      the HA (i.e. something like getsockname() which returns
      the Co@ for a connected socket in place of the Ho@).
    - phase one and phase two messages are sent using this Co@.
    - phase one identity is not an address if no transient
      certificate for the Co@ is available.
    - the authentication with this identity must be allowed
      (including when recommendation J is enforced).
    - until the home address identity is defined and implemented,
      the phase two identity must be an address identity with the H@,
      and this must be allowed (according to recommendation O).
    - the result is a SA pair between the MN with its H@ and
      the HA with its HA Address, the favorite is AH transport mode
      (enough security and best efficiency).
   This SA pair establishment stresses the issue of relative lifetimes
   of the phase one and the SA pair so:
   *RECOMMENDATION Q: IKE implementations MUST support lifetimes
   *for the phase one which are far longer or far shorter than
   *the lifetime of SA pairs established by phases two.
   and with very long phase one lifetimes:
   *RECOMMENDATION R: IKE implementations SHOULD be able to lookup
   *still valid phase one state from a phase two message using
   *different addresses for transport, for instance using the
   *ISAKMP SA SPI aka cookies [5].

draft-dupont-ipsec-mipv6-00.txt                              [Page 10]

INTERNET-DRAFT          IPsec more MIPv6 friendly             Feb 2002

   The SA pair for tunnels is more easy: the only problem is with
    - a solution is dynamically create the proper policy from
      mobility information (i.e. BUs) and to establish the SA
      pair described in section 2.6 (combined tunnel mode with HA).
      One can make things a bit faster reusing the phase one
      (cf recommendations R and L2).
    - another solution is to create a SA pair using only the MN's H@
      (this can be done as soon as the first BU is processed because
      the HA can use the standard routing optimization mode for its
      own traffic, in fact this is the default behavior) and to
      use a to-be-defined direct cooperation between IPsec and
      mobility to update the outer MN address in the SA pair
      (a good use of the HIP readdress [9]).

4. Security Considerations

   At the exception of recommendations F and K, all recommendations
   made in this document are about interpretation of IPsec
   specification details. In fact, we are convinced these
   recommendations shall improve the security of both IPsec and
   Mobile IPv6.

5. Acknowledgments

   Some of the MN-HA ideas were developed in the authentication vs.
   authorization brainstorming, for instance the home address identity
   (unfortunately I don't remember who proposed this).

   Of course the current terrible interaction between IPsec and
   Mobile IPv6 was and still is discussed in the mobile-ip WG list
   and from time to time in the ipsec WG list too. So many thanks
   to the little number of persons who participate to both lists.

   I finish with authors of "open source" IKE implementations,
   particularly Shoichi Sakane who has written the IPsec
   implementation (including an IKE daemon, racoon) I use.

draft-dupont-ipsec-mipv6-00.txt                              [Page 11]

INTERNET-DRAFT          IPsec more MIPv6 friendly             Feb 2002

6. Normative References

   [1] S. Kent, R. Atkinson, "Security Architecture for the Internet
   Protocol", RFC 2401, November 1998.

   [2] S. Kent, R. Atkinson, "IP Authentication Header", RFC 2402,
   November 1998.

   [3] S. Kent, R. Atkinson, "IP Encapsulating Security Payload
   (ESP)", RFC 2406, November 1998.

   [4] D. Piper, "The Internet IP Security Domain of Interpretation
   for ISAKMP", RFC 2407, November 1998.

   [5] D. Maughan, M. Schertler, M. Schneider, J. Turner, "Internet
   Security Association and Key Management Protocol (ISAKMP)",
   RFC 2408, November 1998.

   [6] D. Harkins, D. Carrel, "The Internet Key Exchange (IKE)",
   RFC 2409, November 1998.

7. Informative References

   [7] D. Johnson, C. Perkins, "Mobility Support in IPv6",
   draft-ietf-mobileip-ipv6-15.txt, July 2001.

   [8] F. Dupont, "Mobility-aware IPsec ESP tunnels",
   draft-dupont-movesptun-00.txt, February 2001.

   [9] R. Moskowitz, "Host Identity Payload And Protocol",
   draft-moskowitz-hip-05.txt, November 2001.

   [10] S. Deering, B. Zill, "Redundant Address Deletion when
   Encapsulating IPv6 in IPv6",
   draft-deering-ipv6-encap-addr-deletion-00.txt, November 2001.

   [11] C. Perkins, "[mobile-ip] A new proposal for handling Home
   Address destination options", http://playground.sun.com/mobile-ip/,
   Message-ID: <3C6C7780.4CFAA7B4@iprg.nokia.com>, February 2002.

   [12] D. McDonald, C. Metz, B. Phan, "PF_KEY Key Management API,
   Version 2", RFC 2367, July 1998.

draft-dupont-ipsec-mipv6-00.txt                              [Page 12]

INTERNET-DRAFT          IPsec more MIPv6 friendly             Feb 2002

8. Author's Address

   Francis Dupont
   ENST Bretagne
   Campus de Rennes
   2, rue de la Chataigneraie
   BP 78
   35512 Cesson-Sevigne Cedex
   Fax: +33 2 99 12 70 30
   EMail: Francis.Dupont@enst-bretagne.fr

draft-dupont-ipsec-mipv6-00.txt                              [Page 13]