One document matched: draft-ietf-v6ops-tunnel-loops-03.xml


<?xml version="1.0" encoding="US-ASCII"?>
<?rfc toc="yes"?>
<?rfc tocompact="yes"?>
<?rfc tocdepth="3"?>
<?rfc tocindent="yes"?>
<?rfc symrefs="yes"?>
<?rfc sortrefs="yes"?>
<?rfc comments="yes"?>
<?rfc inline="yes"?>
<?rfc compact="yes"?>
<?rfc subcompact="no"?>
<?rfc strict='yes'?>
<rfc category="info" docName="draft-ietf-v6ops-tunnel-loops-03.txt"
     ipr="trust200902">
  <front>
    <title abbrev="Routing Loop Attack">Routing Loop Attack using IPv6
    Automatic Tunnels: Problem Statement and Proposed Mitigations</title>

    <author fullname="Gabi Nakibly" initials="G." surname="Nakibly">
      <organization
      abbrev="National EW Research & Simulation Center">National EW
      Research & Simulation Center</organization>

      <address>
        <postal>
          <street>P.O. Box 2250 (630)</street>

          <city>Haifa</city>

          <code>31021</code>

          <country>Israel</country>
        </postal>

        <email>gnakibly@yahoo.com</email>
      </address>
    </author>

    <author fullname="Fred L. Templin" initials="F." surname="Templin">
      <organization>Boeing Research & Technology</organization>

      <address>
        <postal>
          <street>P.O. Box 3707 MC 7L-49</street>

          <city>Seattle</city>

          <region>WA</region>

          <code>98124</code>

          <country>USA</country>
        </postal>

        <email>fltemplin@acm.org</email>
      </address>
    </author>

    <date day="04" month="February" year="2011" />

    <keyword>I-D</keyword>

    <keyword>Internet-Draft</keyword>

    <abstract>
      <t>This document is concerned with security vulnerabilities in
      IPv6-in-IPv4 automatic tunnels. These vulnerabilities allow an attacker
      to take advantage of inconsistencies between the IPv4 routing state and
      the IPv6 routing state. The attack forms a routing loop which can be
      abused as a vehicle for traffic amplification to facilitate DoS attacks.
      The first aim of this document is to inform on this attack and its root
      causes. The second aim is to present some possible mitigation
      measures.</t>
    </abstract>
  </front>

  <middle>
    <section title="Introduction">
      <t>IPv6-in-IPv4 tunnels are an essential part of many migration plans
      for IPv6. They allow two IPv6 nodes to communicate over an IPv4-only
      network. Automatic tunnels that use stateless address mapping (hereafter
      called "automatic tunnels") are a category of tunnels in which a
      tunneled packet's egress IPv4 address is embedded within the destination
      IPv6 address of the packet. An automatic tunnel's router is a router
      that respectively encapsulates and decapsulates the IPv6 packets into
      and out of the tunnel.</t>

      <t>Ref. <xref target="USENIX09"></xref> pointed out the existence of a
      vulnerability in the design of IPv6 automatic tunnels. Tunnel routers
      operate on the implicit assumption that the destination address of an
      incoming IPv6 packet is always an address of a valid node that can be
      reached via the tunnel. The assumption of path validity poses a denial
      of service risk as inconsistency between the IPv4 routing state and the
      IPv6 routing state allows a routing loop to be formed.</t>

      <t>An attacker can exploit this vulnerability by crafting a packet which
      is routed over a tunnel to a node that is not participating in that
      tunnel. This node may forward the packet out of the tunnel to the native
      IPv6 network. There the packet is routed back to the ingress point that
      forwards it back into the tunnel. Consequently, the packet loops in and
      out of the tunnel. The loop terminates only when the Hop Limit field in
      the IPv6 header of the packet is decremented to zero. This vulnerability
      can be abused as a vehicle for traffic amplification to facilitate DoS
      attacks <xref target="RFC4732"></xref>.</t>

      <t>Without compensating security measures in place, all IPv6 automatic
      tunnels that are based on protocol-41 encapsulation <xref
      target="RFC4213"></xref> are vulnerable to such an attack including
      ISATAP <xref target="RFC5214"></xref>, 6to4 <xref
      target="RFC3056"></xref> and 6rd <xref target="RFC5969"></xref>. It
      should be noted that this document does not consider non-protocol-41
      encapsulation attacks. In particular, we do not address the Teredo <xref
      target="RFC4380"></xref> attacks described in <xref
      target="USENIX09"></xref>. These attacks are considered in <xref
      target="I-D.gont-6man-teredo-loops"></xref>.</t>

      <t>The aim of this document is to shed light on the routing loop attack
      and describe possible mitigation measures that should be considered by
      operators of current IPv6 automatic tunnels and by designers of future
      ones. We note that tunnels may be deployed in various operational
      environments, e.g. service provider network, enterprise network, etc.
      Specific issues related to the attack which are derived from the
      operational environment are not considered in this document.</t>
    </section>

    <section title="A Detailed Description of the Attack">
      <t>In this section we shall denote an IPv6 address of a node reached via
      a given tunnel by the prefix of the tunnel and an IPv4 address of the
      tunnel end point, i.e., Addr(Prefix, IPv4). Note that the IPv4 address
      may or may not be part of the prefix (depending on the specification of
      the tunnel's protocol). The IPv6 address may be dependent on additional
      bits in the interface ID, however for our discussion their exact value
      is not important.</t>

      <t>The two victims of this attack are routers - R1 and R2 - of two
      different tunnels - T1 and T2. Both routers have the capability to
      forward IPv6 packets in and out of their respective tunnels. The two
      tunnels need not be based on the same tunnel protocol. The only
      condition is that the two tunnel protocols be based on protocol-41
      encapsulation. The IPv4 address of R1 is IP1, while the prefix of its
      tunnel is Prf1. IP2 and Prf2 are the respective values for R2. We assume
      that IP1 and IP2 belong to the same address realm, i.e., they are either
      both public, or both private and belong to the same internal network.
      The following network diagram depicts the locations of the two
      routers.</t>

      <figure anchor="network" title="The network setting of the attack">
        <artwork><![CDATA[
                               #######
                               # R1  #
                               #######
                              //      \
                    T1       //        \
                 interface  //          \  
            _______________//_         __\________________
           |                  |       |                   |
           |  IPv4 Network    |       |   IPv6 Network    |
           |__________________|       |___________________|
                          \\             /
                           \\           / 
                     T2     \\         /
                  interface  \\       /
                               #######
                               # R2  #
                               #######


]]></artwork>
      </figure>

      <t>The attack is depicted in <xref target="figattack"></xref>. It is
      initiated by sending an IPv6 packet (packet 0 in <xref
      target="figattack"></xref>) destined to a fictitious end point that
      appears to be reached via T2 and has IP1 as its IPv4 address, i.e.,
      Addr(Prf2, IP1). The source address of the packet is a T1 address with
      Prf1 as the prefix and IP2 as the embedded IPv4 address, i.e.,
      Addr(Prf1, IP2). As the prefix of the destination address is Prf2, the
      packet will be routed over the IPv6 network to T2.</t>

      <t>We assume that R2 is the packet's entry point to T2. R2 receives the
      packet through its IPv6 interface and forwards it over its T2 interface
      encapsulated with an IPv4 header having a destination address derived
      from the IPv6 destination, i.e., IP1. The source address is the address
      of R2, i.e., IP2. The packet (packet 1 in <xref
      target="figattack"></xref>.) is routed over the IPv4 network to R1,
      which receives the packet on its IPv4 interface. It processes the packet
      as a packet that originates from one of the end nodes of T1.</t>

      <t>Since the IPv4 source address corresponds to the IPv6 source address,
      R1 will decapsulate the packet. Since the packet's IPv6 destination is
      outside of T1, R1 will forward the packet onto a native IPv6 interface.
      The forwarded packet (packet 2 in <xref target="figattack"></xref>) is
      identical to the original attack packet. Hence, it is routed back to R2,
      in which the loop starts again. Note that the packet may not necessarily
      be transported from R1 over native IPv6 network. R1 may be connected to
      the IPv6 network through another tunnel.</t>

      <figure anchor="figattack"
              title="Routing loop attack between two tunnels' routers">
        <artwork><![CDATA[
                       R1               R2                
                       |                |   0                
                       |        1       |<------                
                       |<===============|                
                       |        2       |                
                       |--------------->|                
                       |        .       |                
                       |        .       |                   

             1  - IPv4: IP2 --> IP1           
                  IPv6: Addr(Prf1,IP2) --> Addr(Prf2,IP1)         
             0,2- IPv6: Addr(Prf1,IP2) --> Addr(Prf2,IP1)
             
           Legend: ====> - tunneled IPv6, ---> - native IPv6
]]></artwork>
      </figure>

      <t>The crux of the attack is as follows. The attacker exploits the fact
      that R2 does not know that R1 does not take part of T2 and that R1 does
      not know that R2 does not take part of T1. The IPv4 network acts as a
      shared link layer for the two tunnels. Hence, the packet is repeatedly
      forwarded by both routers. It is noted that the attack will fail when
      the IPv4 network can not transport packets between the tunnels. For
      example, when the two routers belong to different IPv4 address realms or
      when ingress/egress filtering is exercised between the routes.</t>

      <t>The loop will stop when the Hop Limit field of the packet reaches
      zero. After a single loop the Hop Limit field is decreased by the number
      of IPv6 routers on path from R1 and R2. Therefore, the number of loops
      is inversely proportional to the number of IPv6 hops between R1 and
      R2.</t>

      <t>The tunnel pair T1 and T2 may be any combination of automatic tunnel
      types, e.g., ISATAP, 6to4 and 6rd. This has the exception that both
      tunnels can not be of type 6to4, since two 6to4 routers can not belong
      to different tunnels (there is only one 6to4 tunnel in the Internet).
      For example, if the attack were to be launched on an ISATAP router (R1)
      and 6to4 relay (R2), then the destination and source addresses of the
      attack packet would be 2002:IP1:* and Prf1::0200:5EFE:IP2,
      respectively.</t>
    </section>

    <section title="Proposed Mitigation Measures">
      <t>This section presents some possible mitigation measures for the
      attack described above. For each measure we shall discuss its advantages
      and disadvantages.</t>

      <t>The proposed measures fall under the following three categories:</t>

      <t><list style="symbols">
          <t>Verification of end point existence</t>

          <t>Operational measures</t>

          <t>Destination and source addresses checks</t>
        </list></t>

      <section title="Verification of end point existence">
        <t>The routing loop attack relies on the fact that a router does not
        know whether there is an end point that can reached via its tunnel
        that has the source or destination address of the packet. This
        category includes mitigation measures which aim to verify that there
        is a node which participate in the tunnel and its address corresponds
        to the packet's destination or source addresses, as appropriate.</t>

        <section anchor="neighbor-check" title="Neighbor Cache Check">
          <t>One way that the router can verify that an end host exists and
          can be reached via the tunnel is by checking whether a valid entry
          exists for it in the neighbor cache of the corresponding tunnel
          interface. The neighbor cache entry can be populated through, e.g.,
          an initial reachability check, receipt of neighbor discovery
          messages, administrative configuration, etc.</t>

          <t>When the router has a packet to send to a potential tunnel host
          for which there is no neighbor cache entry, it can perform an
          initial reachability check on the packet's destination address,
          e.g., as specified in the second paragraph of Section 8.4 of <xref
          target="RFC5214"></xref>. (The router can similarly perform a
          "reverse reachability" check on the packet's source address when it
          receives a packet from a potential tunnel host for which there is no
          neighbor cache entry.) This reachability check parallels the address
          resolution specifications in Section 7.2 of <xref
          target="RFC4861"></xref>, i.e., the router maintains a small queue
          of packets waiting for reachability confirmation to complete. If
          confirmation succeeds, the router discovers that a legitimate tunnel
          host responds to the address. Otherwise, the router discards
          subseqent packets and returns ICMP destination unreachable
          indications as specified in Section 7.2.2 of <xref
          target="RFC4861"></xref>.</t>

          <t>Note that this approach assumes that the neighbor cache will
          remain coherent and not subject to malicious attack, which must be
          confirmed based on specific deployment scenarios. One possible way
          for an attacker to subvert the neighbor cache is to send false
          neighbor discovery messages with a spoofed source address.</t>
        </section>

        <section anchor="known-ipv4-check" title="Known IPv4 Address Check">
          <t>Another approach that enables a router to verify that an end host
          exists and can be reached via the tunnel is simply by
          pre-configuring the router with the set of IPv4 addresses that are
          authorized to use the tunnel. Upon this configuration the router can
          perform the following simple checks:</t>

          <t><list style="symbols">
              <t>When the router forwards an IPv6 packet into the tunnel
              interface with a destination address that matches an on-link
              prefix and that embeds the IPv4 address IP1, it discards the
              packet if IP1 does not belong to the configured list of IPv4
              addresses.</t>

              <t>When the router receives an IPv6 packet on the tunnel's
              interface with a source address that matches a on-link prefix
              and that embeds the IPv4 address IP2, it discards the packet if
              IP2 does not belong to the configured list of IPv4
              addresses.</t>
            </list></t>
        </section>
      </section>

      <section anchor="operational" title="Operational Measures">
        <t>The following measures can be taken by the network operator. Their
        aim is to configure the network in such a way that the attacks can not
        take place.</t>

        <section anchor="Avoiding-shared-IPv4-link"
                 title="Avoiding a Shared IPv4 Link">
          <t>As noted above, the attack relies on having an IPv4 network as a
          shared link-layer between more than one tunnel. From this the
          following two mitigation measures arise:</t>

          <section anchor="filtering"
                   title="Filtering IPv4 Protocol-41 Packets">
            <t>In this measure a tunnel router may drop all IPv4 protocol-41
            packets received or sent over interfaces that are attached to an
            untrusted IPv4 network. This will cut-off any IPv4 network as a
            shared link. This measure has the advantage of simplicity.
            However, such a measure may not always be suitable for scenarios
            where IPv4 connectivity is essential on all interfaces.</t>
          </section>

          <section anchor="single-tunnel"
                   title="Operational Avoidance of Multiple Tunnels">
            <t>This measure mitigates the attack by simply allowing for a
            single IPv6 tunnel to operate in a bounded IPv4 network. For
            example, the attack can not take place in broadband home networks.
            In such cases there is a small home network having a single
            residential gateway which serves as a tunnel router. A tunnel
            router is vulnerable to the attack only if it has at least two
            interfaces with a path to the Internet: a tunnel interface and a
            native IPv6 interface (as depicted in <xref
            target="network"></xref>). However, a residential gateway usually
            has only a single interface to the Internet, therefore the attack
            can not take place. Moreover, if there are only one or a few
            tunnel routers in the IPv4 network and all participate in the same
            tunnel then there is no opportunity for perpetuating the loop.</t>

            <t>This approach has the advantage that it avoids the attack
            profile altogether without need for explicit mitigations. However,
            it requires careful configuration management which may not be
            tenable in large and/or unbounded IPv4 networks.</t>
          </section>
        </section>

        <section anchor="single-border-router" title="A Single Border Router">
          <t>It is reasonable to assume that a tunnel router shall accept or
          forward tunneled packets only over its tunnel interface. It is also
          reasonable to assume that a tunnel router shall accept or forward
          IPv6 packets only over its IPv6 interface. If these two interfaces
          are physically different then the network operator can mitigate the
          attack by ensuring that the following condition holds: there is no
          path between these two interfaces that does not go through the
          tunnel router.</t>

          <t>The above condition ensures that an encapsulated packet which is
          transmitted over the tunnel interface will not get to another tunnel
          router and from there to the IPv6 interface of the first router. The
          condition also ensures the reverse direction, i.e., an IPv6 packet
          which is transmitted over the IPv6 interface will not get to another
          tunnel router and from there to the tunnel interface of the first
          router. This condition is essentially translated to a scenario in
          which the tunnel router is the only border router between the IPv6
          network and the IPv4 network to which it is attached (as in
          broadband home network scenario mentioned above).</t>
        </section>

        <section anchor="multi-border-router"
                 title="A Comprehensive List of Tunnel Routers">
          <t>If a tunnel router can be configured with a comprehensive list of
          IPv4 addresses of all other tunnel routers in the network, then the
          router can use the list as a filter to discard any tunneled packets
          coming from other routers. For example, a tunnel router can use the
          network's ISATAP Potential Router List (PRL) <xref
          target="RFC5214"></xref> as a filter as long as there is operational
          assurance that all ISATAP routers are listed and that no other types
          of tunnel routers are present in the network.</t>

          <t>This measure parallels the one proposed for 6rd in <xref
          target="RFC5969"></xref> where the 6rd BR filters all known relay
          addresses of other tunnels inside the ISP's network.</t>

          <t>This measure is especially useful for intra-site tunneling
          mechanisms, such as ISATAP and 6rd, since filtering can be exercised
          on well-defined site borders.</t>
        </section>
      </section>

      <section anchor="address-check"
               title="Destination and Source Address Checks">
        <t>Tunnel routers can use a source address check mitigation when they
        forward an IPv6 packet into a tunnel interface with an IPv6 source
        address that embeds one of the router's configured IPv4 addresses.
        Similarly, tunnel routers can use a destination address check
        mitigation when they receive an IPv6 packet on a tunnel interface with
        an IPv6 destination address that embeds one of the router's configured
        IPv4 addresses. These checks should correspond to both tunnels' IPv6
        address formats, regardless of the type of tunnel the router
        employs.</t>

        <t>For example, if tunnel router R1 (of any tunnel protocol) forwards
        a packet into a tunnel interface with an IPv6 source address that
        matches the 6to4 prefix 2002:IP1::/48, the router discards the packet
        if IP1 is one of its own IPv4 addresses. In a second example, if
        tunnel router R2 receives an IPv6 packet on a tunnel interface with an
        IPv6 destination address with an off-link prefix but with an interface
        identifier that matches the ISATAP address suffix ::0200:5EFE:IP2, the
        router discards the packet if IP2 is one of its own IPv4
        addresses.</t>

        <t>Hence a tunnel router can avoid the attack by performing the
        following checks:</t>

        <t><list style="symbols">
            <t>When the router forwards an IPv6 packet into a tunnel
            interface, it discards the packet if the IPv6 source address has
            an off-link prefix but embeds one of the router's configured IPv4
            addresses.</t>

            <t>When the router receives an IPv6 packet on a tunnel interface,
            it discards the packet if the IPv6 destination address has an
            off-link prefix but embeds one of the router's configured IPv4
            addresses.</t>
          </list></t>

        <t>This approach has the advantage that that no ancillary state is
        required, since checking is through static lookup in the lists of IPv4
        and IPv6 addresses belonging to the router. However, this approach has
        some inherent limitations</t>

        <t><list style="symbols">
            <t>The checks incur an overhead which is proportional to the
            number of IPv4 addresses assigned to the router. If a router is
            assigned many addresses, the additional processing overhead for
            each packet may be considerable. Note that an unmitigated attack
            packet would be repetitively processed by the router until the Hop
            Limit expires, which may require as many as 255 iterations. Hence,
            an unmitigated attack will consume far more aggregate processing
            overhead than per-packet address checks even if the router assigns
            a large number of addresses.</t>

            <t>The checks should be performed for the IPv6 address formats of
            every existing automatic IPv6 tunnel protocol (which uses
            protocol-41 encapsulation). Hence, the checks must be updated as
            new protocols are defined.</t>

            <t>Before the checks can be performed the format of the address
            must be recognized. There is no guarantee that this can be
            generally done. For example, one can not determine if an IPv6
            address is a 6rd one, hence the router would need to be configured
            with a list of all applicable 6rd prefixes (which may be
            prohibitively large) in order to unambiguously apply the
            checks.</t>

            <t>The checks cannot be performed if the embedded IPv4 address is
            a private one <xref target="RFC1918"></xref> since it is ambiguous
            in scope. Namely, the private address may be legitimately
            allocated to another node in another routing region.</t>
          </list></t>

        <t>The last limitation may be relieved if the router has some
        information that allows it to unambiguously determine the scope of the
        address. The check in the following subsection is one example for
        this.</t>

        <section anchor="known-ipv6-check" title="Known IPv6 Prefix Check">
          <t>A router may be configured with the full list of IPv6 subnet
          prefixes assigned to the tunnels attached to its current IPv4
          routing region. In such a case it can use the list to determine when
          static destination and source address checks are possible. By
          keeping track of the list of IPv6 prefixes assigned to the tunnels
          in the IPv4 routing region, a router can perform the following
          checks on an address which embeds a private IPv4 address:</t>

          <t><list style="symbols">
              <t>When the router forwards an IPv6 packet into its tunnel with
              a source address that embeds a private IPv4 address and matches
              an IPv6 prefix in the prefix list, it determines whether the
              packet should be discarded or forwarded by performing the source
              address check specified in <xref target="address-check"></xref>.
              Otherwise, the router forwards the packet.</t>

              <t>When the router receives an IPv6 packet on its tunnel
              interface with a destination address that embeds a private IPv4
              address and matches an IPv6 prefix in the prefix list, it
              determines whether the packet should be discarded or forwarded
              by performing the destination address check specified in <xref
              target="address-check"></xref>. Otherwise, the router forwards
              the packet.</t>
            </list> The disadvantage of this approach is the administrative
          overhead for maintaining the list of IPv6 subnet prefixes associated
          with an IPv4 routing region may become unwieldy should that list be
          long and/or frequently updated.</t>
        </section>
      </section>
    </section>

    <section anchor="recommendations" title="Recommendations">
      <t>In light of the mitigation measures proposed above we make the
      following recommendations in decreasing order:</t>

      <t><list style="numbers">
          <t>When possible, it is recommended that the attacks are
          operationally eliminated (as per one of the measures proposed in
          <xref target="operational"></xref>).</t>

          <t>For tunnel routers that keep a coherent and trusted neighbor
          cache which includes all legitimate end-points of the tunnel, we
          recommend exercising the Neighbor Cache Check.</t>

          <t>For tunnel routers that can implement the Neighbor Reachability
          Check, we recommend exercising it.</t>

          <t>For tunnels having small and static list of end-points we
          recommend exercising Known IPv4 Address Check.</t>

          <t>We generally do not recommend using the Destination and Source
          Address Checks since they can not mitigate routing loops with 6rd
          routers. Therefore, these checks should not be used alone unless
          there is operational assurance that other measures are exercised to
          prevent routing loops with 6rd routers.</t>
        </list></t>

      <t>As noted earlier, tunnels may be deployed in various operational
      environments. There is a possibility that other mitigations may be
      feasible in specific deployment scenarios. The above recommendations are
      general and do not attempt to cover such scenarios.</t>
    </section>

    <section title="IANA Considerations">
      <t>This document has no IANA considerations.</t>
    </section>

    <section anchor="security" title="Security Considerations">
      <t>This document aims at presenting possible solutions to the routing
      loop attack which involves automatic tunnels' routers. It contains
      various checks that aim to recognize and drop specific packets that have
      strong potential to cause a routing loop. These checks do not introduce
      new security threats.</t>
    </section>

    <section anchor="acknowledge" title="Acknowledgments">
      <t>This work has benefited from discussions on the V6OPS, 6MAN and
      SECDIR mailing lists. Remi Despres, Christian Huitema, Dmitry Anipko,
      Dave Thaler and Fernando Gont are acknowledged for their
      contributions.</t>
    </section>
  </middle>

  <back>
    <references title="Normative References">
      <?rfc include="reference.RFC.3056"?>

      <?rfc include="reference.RFC.5214"?>

      <?rfc include="reference.RFC.5969"?>

      <?rfc include="reference.RFC.1918"?>

      <?rfc include="reference.RFC.4861"?>

      <?rfc include="reference.RFC.4213"?>
    </references>

    <references title="Informative References">
      <reference anchor="USENIX09">
        <front>
          <title>Routing Loop Attacks using IPv6 Tunnels</title>

          <author fullname="Gabi Nakibly" initials="G." surname="Nakibly">
            <organization></organization>
          </author>

          <author fullname="Michael Arov" initials="M." surname="Arov">
            <organization></organization>
          </author>

          <date month="USENIX WOOT, August" year="2009" />
        </front>
      </reference>

      <?rfc include="reference.RFC.4732"?>

      <?rfc include="reference.RFC.4380"?>

      <?rfc include="reference.I-D.draft-gont-6man-teredo-loops-00"?>
    </references>
  </back>
</rfc>

PAFTECH AB 2003-20262026-04-23 23:05:32