RFC 6027






Internet Engineering Task Force (IETF)                            Y. Nir
Request for Comments: 6027                                   Check Point
Category: Informational                                     October 2010
ISSN: 2070-1721


                    IPsec Cluster Problem Statement

Abstract



   This document defines the terminology, problem statement, and
   requirements for implementing Internet Key Exchange (IKE) and IPsec
   on clusters.  It also describes gaps in existing standards and their
   implementation that need to be filled in order to allow peers to
   interoperate with clusters from different vendors.  Agreed upon
   terminology, problem statement, and requirements will allow IETF
   working groups to consider development of IPsec/IKEv2 mechanisms to
   simplify cluster implementations.

Status of This Memo



   This document is not an Internet Standards Track specification; it is
   published for informational purposes.

   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).  Not all documents
   approved by the IESG are a candidate for any level of Internet
   Standard; see Section 2 of RFC 5741.

   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/rfc6027.

















Nir                           Informational                     [Page 1]

RFC 6027             IPsec Cluster Problem Statement        October 2010


Copyright Notice



   Copyright (c) 2010 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 ....................................................3
      1.1. Conventions Used in This Document ..........................3
   2. Terminology .....................................................3
   3. The Problem Statement ...........................................5
      3.1. Scope ......................................................5
      3.2. A Lot of Long-Lived State ..................................6
      3.3. IKE Counters ...............................................6
      3.4. Outbound SA Counters .......................................6
      3.5. Inbound SA Counters ........................................7
      3.6. Missing Synch Messages .....................................8
      3.7. Simultaneous Use of IKE and IPsec SAs by Different
           Members ....................................................8
           3.7.1. Outbound SAs Using Counter Modes ....................9
      3.8. Different IP Addresses for IKE and IPsec ..................10
      3.9. Allocation of SPIs ........................................10
   4. Security Considerations ........................................10
   5. Acknowledgements ...............................................11
   6. References .....................................................11
      6.1. Normative References ......................................11
      6.2. Informative References ....................................11














Nir                           Informational                     [Page 2]

RFC 6027             IPsec Cluster Problem Statement        October 2010


1.  Introduction



   IKEv2, as described in [RFC5996], and IPsec, as described in
   [RFC4301] and others, allows deployment of VPNs between different
   sites as well as from VPN clients to protected networks.

   As VPNs become increasingly important to the organizations deploying
   them, there is a demand to make IPsec solutions more scalable and
   less prone to down time, by using more than one physical gateway to
   either share the load or back each other up, forming a "cluster" (see
   Section 2).  Similar demands have been made in the past for other
   critical pieces of an organization's infrastructure, such as DHCP and
   DNS servers, Web servers, databases, and others.

   IKE and IPsec are, in particular, less friendly to clustering than
   these other protocols, because they store more state, and that state
   is more volatile.  Section 2 defines terminology for use in this
   document and in the envisioned solution documents.

   In general, deploying IKE and IPsec in a cluster requires such a
   large amount of information to be synchronized among the members of
   the cluster that it becomes impractical.  Alternatively, if less
   information is synchronized, failover would mean a prolonged and
   intensive recovery phase, which negates the scalability and
   availability promises of using clusters.  In Section 3, we will
   describe this in more detail.

1.1.  Conventions Used in This Document



   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 [RFC2119].

2.  Terminology



   "Single Gateway" is an implementation of IKE and IPsec enforcing a
   certain policy, as described in [RFC4301].

   "Cluster" is a set of two or more gateways, implementing the same
   security policy, and protecting the same domain.  Clusters exist to
   provide both high availability through redundancy and scalability
   through load sharing.

   "Member" is one gateway in a cluster.

   "Availability" is a measure of a system's ability to perform the
   service for which it was designed.  It is measured as the percentage
   of time a service is available from the time it is supposed to be



Nir                           Informational                     [Page 3]

RFC 6027             IPsec Cluster Problem Statement        October 2010


   available.  Colloquially, availability is sometimes expressed in
   "nines" rather than percentage, with 3 "nines" meaning 99.9%
   availability, 4 "nines" meaning 99.99% availability, etc.

   "High Availability" is a property of a system, not a configuration
   type.  A system is said to have high availability if its expected
   down time is low.  High availability can be achieved in various ways,
   one of which is clustering.  All the clusters described in this
   document achieve high availability.  What "high" means depends on the
   application, but usually is 4 to 6 "nines" (at most 0.5-50 minutes of
   down time per year in a system that is supposed to be available all
   the time.

   "Fault Tolerance" is a property related to high availability, where a
   system maintains service availability, even when a specified set of
   fault conditions occur.  In clusters, we expect the system to
   maintain service availability, when one or more of the cluster
   members fails.

   "Completely Transparent Cluster" is a cluster where the occurrence of
   a fault is never visible to the peers.

   "Partially Transparent Cluster" is a cluster where the occurrence of
   a fault may be visible to the peers.

   "Hot Standby Cluster", or "HS Cluster" is a cluster where only one of
   the members is active at any one time.  This member is also referred
   to as the "active" member, whereas the other(s) are referred to as
   "standbys".  The Virtual Router Redundancy Protocol (VRRP)
   ([RFC5798]) is one method of building such a cluster.

   "Load Sharing Cluster", or "LS Cluster" is a cluster where more than
   one of the members may be active at the same time.  The term "load
   balancing" is also common, but it implies that the load is actually
   balanced between the members, and this is not a requirement.

   "Failover" is the event where one member takes over some load from
   some other member.  In a hot standby cluster, this happens when a
   standby member becomes active due to a failure of the former active
   member, or because of an administrator command.  In a load sharing
   cluster, this usually happens because of a failure of one of the
   members, but certain load-balancing technologies may allow a
   particular load (such as all the flows associated with a particular
   child Security Association (SA)) to move from one member to another
   to even out the load, even without any failures.






Nir                           Informational                     [Page 4]

RFC 6027             IPsec Cluster Problem Statement        October 2010


   "Tight Cluster" is a cluster where all the members share an IP
   address.  This could be accomplished using configured interfaces with
   specialized protocols or hardware, such as VRRP, or through the use
   of multicast addresses, but in any case, peers need only be
   configured with one IP address in the Peer Authentication Database.

   "Loose Cluster" is a cluster where each member has a different IP
   address.  Peers find the correct member using some method such as DNS
   queries or the IKEv2 redirect mechanism ([RFC5685]).  In some cases,
   a member's IP address(es) may be allocated to another member at
   failover.

   "Synch Channel" is a communications channel among the cluster
   members, which is used to transfer state information.  The synch
   channel may or may not be IP based, may or may not be encrypted, and
   may work over short or long distances.  The security and physical
   characteristics of this channel are out of scope for this document,
   but it is a requirement that its use be minimized for scalability.

3.  The Problem Statement



   This section starts by scoping the problem, and goes on to list each
   of the issues encountered while setting up a cluster of IPsec VPN
   gateways.

3.1.  Scope



   This document will make no attempt to describe the problems in
   setting up a generic cluster.  It describes only problems related to
   the IKE/IPsec protocols.

   The problem of synchronizing the policy between cluster members is
   out of scope, as this is an administrative issue that is not
   particular to either clusters or to IPsec.

   The interesting scenario here is VPN, whether inter-domain or remote
   access.  Host-to-host transport mode is not expected to benefit from
   this work.

   We do not describe in full the problems of the communication channel
   between cluster members (the Synch Channel), nor do we intend to
   specify anything in this space later.  Specifically, mixed-vendor
   clusters are out of scope.

   The problem statement anticipates possible protocol-level solutions
   between IKE/IPsec peers in order to improve the availability and/or
   performance of VPN clusters.  One vendor's IPsec endpoint should be
   able to work, optimally, with another vendor's cluster.



Nir                           Informational                     [Page 5]

RFC 6027             IPsec Cluster Problem Statement        October 2010


3.2.  A Lot of Long-Lived State



   IKE and IPsec have a lot of long-lived state:

   o  IKE SAs last for minutes, hours, or days, and carry keys and other
      information.  Some gateways may carry thousands to hundreds of
      thousands of IKE SAs.

   o  IPsec SAs last for minutes or hours, and carry keys, selectors,
      and other information.  Some gateways may carry hundreds of
      thousands of such IPsec SAs.

   o  SPD (Security Policy Database) cache entries.  While the SPD is
      unchanging, the SPD cache changes on the fly due to narrowing.
      Entries last at least as long as the SAD (Security Association
      Database) entries, but tend to last even longer than that.

   A naive implementation of a cluster would have no synchronized state,
   and a failover would produce an effect similar to that of a rebooted
   gateway.  [RFC5723] describes how new IKE and IPsec SAs can be
   recreated in such a case.

3.3.  IKE Counters



   We can overcome the first problem described in Section 3.2, by
   synchronizing states -- whenever an SA is created, we can synch this
   new state to all other members.  However, those states are not only
   long lived, they are also ever changing.

   IKE has message counters.  A peer MUST NOT process message n until
   after it has processed message n-1.  Skipping message IDs is not
   allowed.  So a newly active member needs to know the last message IDs
   both received and transmitted.

   One possible solution is to synchronize information about the IKE
   message counters after every IKE exchange.  This way, the newly
   active member knows what messages it is allowed to process, and what
   message IDs to use on IKE requests, so that peers process them.  This
   solution may be appropriate in some cases, but may be too onerous in
   systems with a lot of SAs.  It also has the drawback that it never
   recovers from the missing synch message problem, which is described
   in Section 3.6.

3.4.  Outbound SA Counters



   The Encapsulating Security Payload (ESP) and Authentication Header
   (AH) have an optional anti-replay feature, where every protected
   packet carries a counter number.  Repeating counter numbers is



Nir                           Informational                     [Page 6]

RFC 6027             IPsec Cluster Problem Statement        October 2010


   considered an attack, so the newly active member MUST NOT use a
   replay counter number that has already been used.  The peer will drop
   those packets as duplicates and/or warn of an attack.

   Though it may be feasible to synchronize the IKE message counters, it
   is almost never feasible to synchronize the IPsec packet counters for
   every IPsec packet transmitted.  So we have to assume that at least
   for IPsec, the replay counter will not be up to date on the newly
   active member, and the newly active member may repeat a counter.

   A possible solution is to synch replay counter information, not for
   each packet emitted, but only at regular intervals, say, every 10,000
   packets or every 0.5 seconds.  After a failover, the newly active
   member advances the counters for outbound IPsec SAs by 10,000
   packets.  To the peer, this looks like up to 10,000 packets were
   lost, but this should be acceptable, as neither ESP nor AH guarantee
   reliable delivery.

3.5.  Inbound SA Counters



   An even tougher issue is the synchronization of packet counters for
   inbound IPsec SAs.  If a packet arrives at a newly active member,
   there is no way to determine whether or not this packet is a replay.
   The periodic synch does not solve this problem at all, because
   suppose we synchronize every 10,000 packets, and the last synch
   before the failover had the counter at 170,000.  It is probable,
   though not certain, that packet number 180,000 has not yet been
   processed, but if packet 175,000 arrives at the newly active member,
   it has no way of determining whether or not that packet has already
   been processed.  The synchronization does prevent the processing of
   really old packets, such as those with counter number 165,000.
   Ignoring all counters below 180,000 won't work either, because that's
   up to 10,000 dropped packets, which may be very noticeable.

   The easiest solution is to learn the replay counter from the incoming
   traffic.  This is allowed by the standards, because replay counter
   verification is an optional feature (see Section 3.2 in [RFC4301]).
   The case can even be made that it is relatively secure, because non-
   attack traffic will reset the counters to what they should be, so an
   attacker faces the dual challenge of a very narrow window for attack,
   and the need to time the attack to a failover event.  Unless the
   attacker can actually cause the failover, this would be very
   difficult.  It should be noted, though, that although this solution
   is acceptable as far as RFC 4301 goes, it is a matter of policy
   whether this is acceptable.






Nir                           Informational                     [Page 7]

RFC 6027             IPsec Cluster Problem Statement        October 2010


   Another possible solution to the inbound IPsec SA problem is to rekey
   all child SAs following a failover.  This may or may not be feasible
   depending on the implementation and the configuration.

3.6.  Missing Synch Messages



   The synch channel is very likely not to be infallible.  Before
   failover is detected, some synchronization messages may have been
   missed.  For example, the active member may have created a new child
   SA using message n.  The new information (entry in the SAD and update
   to counters of the IKE SA) is sent on the synch channel.  Still, with
   every possible technology, the update may be missed before the
   failover.

   This is a bad situation, because the IKE SA is doomed.  The newly
   active member has two problems:

   o  It does not have the new IPsec SA pair.  It will drop all incoming
      packets protected with such an SA.  This could be fixed by sending
      some DELETEs and INVALID_SPI notifications, if it wasn't for the
      other problem.

   o  The counters for the IKE SA show that only request n-1 has been
      sent.  The next request will get the message ID n, but that will
      be rejected by the peer.  After a sufficient number of
      retransmissions and rejections, the whole IKE SA with all
      associated IPsec SAs will get dropped.

   The above scenario may be rare enough that it is acceptable that on a
   configuration with thousands of IKE SAs, a few will need to be
   recreated from scratch or using session resumption techniques.
   However, detecting this may take a long time (several minutes) and
   this negates the goal of creating a cluster in the first place.

3.7.  Simultaneous Use of IKE and IPsec SAs by Different Members



   For load sharing clusters, all active members may need to use the
   same SAs, both IKE and IPsec.  This is an even greater problem than
   in the case of hot standby clusters, because consecutive packets may
   need to be sent by different members to the same peer gateway.

   The solution to the IKE SA issue is up to the implementation.  It's
   possible to create some locking mechanism over the synch channel, or
   else have one member "own" the IKE SA and manage the child SAs for
   all other members.  For IPsec, solutions fall into two broad
   categories.





Nir                           Informational                     [Page 8]

RFC 6027             IPsec Cluster Problem Statement        October 2010


   The first is the "sticky" category, where all communications with a
   single peer, or all communications involving a certain SPD cache
   entry go through a single peer.  In this case, all packets that match
   any particular SA go through the same member, so no synchronization
   of the replay counter needs to be done.  Inbound processing is a
   "sticky" issue (no pun intended), because the packets have to be
   processed by the correct member based on peer and the Security
   Parameter Index (SPI), and most load balancers will not be able to
   match the SPIs to the correct member, unless stickiness extends to
   all traffic with a particular peer.  Another disadvantage of sticky
   solutions is that the load tends to not distribute evenly, especially
   if one SA covers a significant portion of IPsec traffic.

   The second is the "duplicate" category, where the child SA is
   duplicated for each pair of IPsec SAs for each active member.
   Different packets for the same peer go through different members, and
   get protected using different SAs with the same selectors and
   matching the same entries in the SPD cache.  This has some
   shortcomings:

   o  It requires multiple parallel SAs, for which the peer has no use.
      Section 2.8 of [RFC5996] specifically allows this, but some
      implementation might have a policy against long-term maintenance
      of redundant SAs.

   o  Different packets that belong to the same flow may be protected by
      different SAs, which may seem "weird" to the peer gateway,
      especially if it is integrated with some deep-inspection
      middleware such as a firewall.  It is not known whether this will
      cause problems with current gateways.  It is also impossible to
      mandate against this, because the definition of "flow" varies from
      one implementation to another.

   o  Reply packets may arrive with an IPsec SA that is not "matched" to
      the one used for the outgoing packets.  Also, they might arrive at
      a different member.  This problem is beyond the scope of this
      document and should be solved by the application, perhaps by
      forwarding misdirected packets to the correct gateway for deep
      inspection.

3.7.1.  Outbound SAs Using Counter Modes



   For SAs involving counter mode ciphers such as Counter Mode (CTR)
   ([RFC3686]) or Galois/Counter Mode (GCM) ([RFC4106]) there is yet
   another complication.  The initial vector for such modes MUST NOT be
   repeated, and senders use methods such as counters or linear feedback
   shift registers (LFSRs) to ensure this.  For an SA shared between
   more than one active member, or even failing over from one member to



Nir                           Informational                     [Page 9]

RFC 6027             IPsec Cluster Problem Statement        October 2010


   another, the cluster members need to make sure that they do not
   generate the same initial vector.  See [COUNTER_MODES] for a
   discussion of this problem in another context.

3.8.  Different IP Addresses for IKE and IPsec



   In many implementations there are separate IP addresses for the
   cluster, and for each member.  While the packets protected by tunnel
   mode child SAs are encapsulated in IP headers with the cluster IP
   address, the IKE packets originate from a specific member, and carry
   that member's IP address.  This may be done so that IPsec traffic
   bypasses the load balancer for greater scalability.  For the peer,
   this looks weird, as the usual thing is for the IPsec packets to come
   from the same IP address as the IKE packets.  Unmodified peers may
   drop such packets.

   One obvious solution is to use some fancy capability of the IKE host
   to change things so that IKE packets also come out of the cluster IP
   address.  This can be achieved through NAT or through assigning
   multiple addresses to interfaces.  This is not, however, possible for
   all implementations, and will not reduce load on the balancer.

   [ARORA] discusses this problem in greater depth, and proposes another
   solution, that does involve protocol changes.

3.9.  Allocation of SPIs



   The SPI associated with each child SA, and with each IKE SA, MUST be
   unique relative to the peer of the SA.  Thus, in the context of a
   cluster, each cluster member MUST generate SPIs in a fashion that
   avoids collisions (with other cluster members) for these SPI values.
   The means by which cluster members achieve this requirement is a
   local matter, outside the scope of this document.

4.  Security Considerations



   Implementations running on clusters MUST be as secure as
   implementations running on single gateways.  In other words, no
   extension or interpretation used to allow operation in a cluster may
   facilitate attacks that are not possible for single gateways.

   Moreover, thought must be given to the synching requirements of any
   protocol extension to make sure that it does not create an
   opportunity for denial-of-service attacks on the cluster.







Nir                           Informational                    [Page 10]

RFC 6027             IPsec Cluster Problem Statement        October 2010


   As mentioned in Section 3.5, allowing an inbound child SA to failover
   to another member has the effect of disabling replay counter
   protection for a short time.  Though the threat is arguably low, it
   is a policy decision whether this is acceptable.

   Section 3.7 describes the problem of the two directions of a flow
   being protected by two SAs that are not part of a matched pair or
   that are not even being processed by the same cluster member.  This
   is not a security problem as far as IPsec is concerned because IPsec
   has policy at the IP, protocol and port level only.  However, many
   IPsec implementations are integrated with stateful firewalls, which
   need to see both sides of a flow.  Such implementations may have to
   forward packets to other members for the firewall to properly inspect
   the traffic.

5.  Acknowledgements



   This document is the collective work, and includes contribution from
   many people who participate in the IPsecME working group.

   The editor would particularly like to acknowledge the extensive
   contribution of the following people (in alphabetical order):
   Jitender Arora, Jean-Michel Combes, Dan Harkins, David Harrington,
   Steve Kent, Tero Kivinen, Alexey Melnikov, Yaron Sheffer, Melinda
   Shore, and Rodney Van Meter.

6.  References



6.1.  Normative References



   [RFC2119]  Bradner, S., "Key words for use in RFCs to Indicate
              Requirement Levels", BCP 14, RFC 2119, March 1997.

   [RFC4301]  Kent, S. and K. Seo, "Security Architecture for the
              Internet Protocol", RFC 4301, December 2005.

   [RFC5996]  Kaufman, C., Hoffman, P., Nir, Y., and P. Eronen,
              "Internet Key Exchange Protocol Version 2 (IKEv2)",
              September 2010.

6.2.  Informative References



   [ARORA]    Arora, J. and P. Kumar, "Alternate Tunnel Addresses for
              IKEv2", Work in Progress, April 2010.







Nir                           Informational                    [Page 11]

RFC 6027             IPsec Cluster Problem Statement        October 2010


   [COUNTER_MODES]
              McGrew, D. and B. Weis, "Using Counter Modes with
              Encapsulating Security Payload (ESP) and Authentication
              Header (AH) to Protect Group Traffic", Work in Progress,
              March 2010.

   [RFC3686]  Housley, R., "Using Advanced Encryption Standard (AES)
              Counter Mode", RFC 3686, January 2009.

   [RFC4106]  Viega, J. and D. McGrew, "The Use of Galois/Counter Mode
              (GCM) in IPsec Encapsulating Security Payload (ESP)",
              RFC 4106, June 2005.

   [RFC5685]  Devarapalli, V. and K. Weniger, "Redirect Mechanism for
              IKEv2", RFC 5685, November 2009.

   [RFC5723]  Sheffer, Y. and H. Tschofenig, "IKEv2 Session Resumption",
              RFC 5723, January 2010.

   [RFC5798]  Nadas, S., "Virtual Router Redundancy Protocol (VRRP)",
              RFC 5798, March 2010.

Author's Address



   Yoav Nir
   Check Point Software Technologies Ltd.
   5 Hasolelim st.
   Tel Aviv  67897
   Israel

   EMail: ynir@checkpoint.com




















Nir                           Informational                    [Page 12]