One document matched: draft-wing-behave-dns64-config-03.xml


<?xml version="1.0" encoding="US-ASCII"?>
<!DOCTYPE rfc SYSTEM "rfc2629.dtd" [
<!ENTITY rfc2119 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.2119.xml">
<!ENTITY rfc3315 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.3315.xml">
<!ENTITY rfc2131 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.2131.xml">
<!ENTITY rfc4361 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.4361.xml">
<!ENTITY I-D.ietf-behave-dns64 SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.ietf-behave-dns64.xml">
<!ENTITY I-D.boucadair-behave-dns-a64 SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.boucadair-behave-dns-a64.xml">
<!ENTITY rfc4242 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.4242.xml">
<!ENTITY rfc3646 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.3646.xml">
<!ENTITY rfc3484 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.3484.xml">
<!ENTITY rfc3513 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.3513.xml">
<!ENTITY rfc3056 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.3056.xml">
<!ENTITY rfc6052 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.6052.xml">
<!ENTITY I-D.ietf-behave-ftp64 SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.ietf-behave-ftp64.xml">
<!ENTITY I-D.wing-behave-learn-prefix SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.wing-behave-learn-prefix.xml">
<!ENTITY I-D.savolainen-mif-dns-server-selection SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.savolainen-mif-dns-server-selection.xml">
<!ENTITY I-D.arifumi-6man-rfc3484-revise SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.arifumi-6man-rfc3484-revise.xml">
]>
<?xml-stylesheet type='text/xsl' href='rfc2629.xslt' ?>
<?rfc toc="yes" ?>
<?rfc rfcprocack="yes" ?>
<?rfc symrefs="yes" ?>
<?rfc iprnotified="yes" ?>
<?rfc strict="yes" ?>
<?rfc compact="yes" ?>
<?rfc subcompact="no" ?>
<?rfc sortrefs="yes" ?>
<?rfc colonspace='yes' ?>
<?rfc tocindent='yes' ?>
<rfc category="std" docName="draft-wing-behave-dns64-config-03"
     ipr="trust200902">
  <front>
    <title abbrev="IPv6-only & Dual-Stack with DNS64">IPv6-only and Dual
    Stack Hosts on the Same Network with DNS64</title>

    <author fullname="Dan Wing" initials="D." surname="Wing">
      <organization abbrev="Cisco">Cisco Systems, Inc.</organization>

      <address>
        <postal>
          <street>170 West Tasman Drive</street>

          <city>San Jose</city>

          <region>CA</region>

          <code>95134</code>

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

        <email>dwing@cisco.com</email>
      </address>
    </author>

    <date />

    <workgroup>BEHAVE Working Group</workgroup>

    <abstract>
      <t>Some networks are expected to support IPv4-only, dual-stack, and
      IPv6-only hosts at the same time. Such networks also want to IPv6/IPv4
      translation for the IPv6-only host so it can access servers on the IPv4
      Internet. On such a network, the synthesized AAAA responses from a DNS64
      can cause traffic to be translated. This document describes a solution
      to avoid that translation when the application uses DNS.</t>
    </abstract>
  </front>

  <middle>
    <section anchor="introduction" title="Introduction">
      <t>In order to access IPv4 servers, an IPv6-only host needs to use an
      IPv6/IPv4 translator. Typically, the IPv6-only host performs a DNS query
      to a DNS64 recursive resolver, which synthesizes an AAAA when necessary.
      However, if a dual-stack host uses that same DNS64 recursive resolver
      and normal <xref target="RFC3484">address selection rules</xref>, the
      dual-stack host will send traffic through the IPv6/IPv4 translator when
      such traffic could have been sent using IPv4. Thus, as an optimization,
      it is desirable that a dual-stack host avoid IPv6/IPv4 translation.</t>

      <t>Note: If the dual-stack host's IPv4 traffic is being NATted the
      difference is NAT44 versus NAT64, so the performance and saleability
      concern is nearly identical. However, at least one application breaks
      when translated between IP address families unless special measures are
      taken <xref target="I-D.ietf-behave-ftp64"></xref>. The IETF should
      decide if it is worthwhile to avoid NAT64 for dual-stack hosts that are
      connected to a network operating a DNS64.</t>

      <t><list style="empty">
          <t>Note: Windows XP can only be configured with IPv4 DNS servers
          <xref target="XP-DNS"></xref>. This means a Windows XP host is
          always dual-stack and requires an IPv4 address in order to send its
          DNS queries. While it is possible to work around this issue by
          running BIND on the Windows XP device itself, this is complex. Thus,
          Windows XP should not be considered a viable operating system to
          join an IPv6-only network.</t>
        </list></t>
    </section>

    <section title="Terminology">
      <t>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 <xref
      target="RFC2119"></xref>.</t>

      <t>"IPv4-only" means a host that has only IPv4 address(es) assigned to
      its interface(s). "Dual-stack" means a host that has an IPv4 address and
      an IPv6 address assigned to its interface(es). "IPv6-only" means a host
      that has only IPv6 address(es) assigned to its interface(s).</t>
    </section>

    <section title="Mechanism">
      <t>It has been observed that some common operating systems, when
      configured as dual-stack, will successfully use an IPv4-mapped address
      (and send an IPv4 packet). But when configured as IPv6-only, they will
      not successfully use an IPv4-mapped address (because they lack an IPv4
      address) <xref target="experiment"></xref>.</t>

      <t>We take advantage of this by configuring the 'normal' DNS server
      using an IPv4-mapped IPv6 address (that is, an IPv6 address starting
      with ::ffff:/96), and configuring the DNS64 server using a normal IPv6
      address.</t>

      <t>DNS servers are used in the order listed <xref
      target="RFC3646"></xref>, so a dual-stack host will use the 'normal' DNS
      server (which is accessible over IPv4) and an IPv6-only host will be
      unable to use that 'normal' DNS server and will use the next server on
      its list.<list style="empty">
          <t>Note: Non-compliant IPv6 stacks might send a packet to the
          IPv4-mapped IPv6 address (::ffff:c000:0201, using the example
          below). To deal with such non-compliant IPv6 implementations the
          network can filter (drop) traffic to that IPv6 address, which will
          force those stacks to timeout when attempting to contact the first
          DNS server and fall back to using the second DNS server.</t>
        </list></t>

      <figure>
        <preamble>For example, a dual-stack host and an IPv6-only host would
        be configured with the following DNS servers, in this order, where the
        first one is the normal DNS server (192.0.2.1) and the second one is
        the DNS64 server (2001:db8:dddd::1234)</preamble>

        <artwork align="center"><![CDATA[
::ffff:192.0.2.1       # 'normal' DNS server
2001:db8:dddd::1234    # DNS64 server
]]></artwork>

        <postamble></postamble>
      </figure>

      <t>This technique requires no change to host operating systems or host
      applications.</t>

      <t>When transitioning from dual-stack to IPv6-only, nothing needs to
      occur - the higher-priority DNS server (with the IPv4-mapped IPv6
      address) will become inaccessible and the DNS client will fail over to
      the next-higher priority DNS server (which is the DNS64 server). This
      does mean the host will take a few extra sections to fully initialize,
      as it will have to timeout its attempts to communicate with the first
      DNS server.</t>

      <t>When transitioning from IPv6-only to dual-stack, nothing
      automatically causes the host to start querying the 'normal' DNS server.
      Thus, a host that transitions from IPv6-only to dual-stack will continue
      to query the DNS64 until the host's stack re-initializes.</t>

      <t><list style="empty">
          <t>Operating System Note: On Linux systems, this technique is not
          effective if the sysctl net.ipv6.bindv6only is set, as setting this
          parameter causes dual-stack systems to not send packets to
          IPv4-mapped IPv6 addresses.</t>
        </list></t>

      <t>If the first DNS server is unavailable (e.g., link failure or DNS
      server failure) and the host's resolver times out, it will try the
      second DNS server, which is a DNS64 server. This is unavoidable with
      this technique. Thus, it is important that a robust infrastructure be
      used for the DNS servers, especially the first DNS server.</t>
    </section>

    <section anchor="securityconsiderations" title="Security Considerations">
      <t>TBD.</t>
    </section>

    <section title="Acknowledgements">
      <t>Thanks to Mohamed Boucadair, Marcelo Braun, Ralph Droms, Dave Thaler,
      Bernie Volz, and Andrew Yourtchenko for their review comments.</t>
    </section>

    <section anchor="iana" title="IANA Considerations">
      <t>This document has no actions for IANA.</t>
    </section>
  </middle>

  <back>
    <references title="Normative References">
      &rfc2119;

      &rfc3315;

      &rfc2131;

      &rfc4361;

      &I-D.ietf-behave-dns64;

      &rfc4242;

      &rfc3646;

      &rfc3484;
    </references>

    <references title="Informative References">
      &rfc6052;

      &I-D.ietf-behave-ftp64;

      &I-D.wing-behave-learn-prefix;

      &I-D.savolainen-mif-dns-server-selection;

      &rfc3513;

      &I-D.arifumi-6man-rfc3484-revise;

      <reference anchor="mif"
                 target="http://www.ietf.org/dyn/wg/charter/mif-charter">
        <front>
          <title>Multiple Interfaces Working Group</title>

          <author fullname="IETF" surname="IETF">
            <organization></organization>
          </author>

          <date year="2009" />
        </front>
      </reference>

      <reference anchor="6man"
                 target="http://www.ietf.org/dyn/wg/charter/6man-charter">
        <front>
          <title>IPv6 Maintenance Working Group</title>

          <author fullname="IETF" surname="IETF">
            <organization></organization>
          </author>

          <date year="2009" />
        </front>
      </reference>

      <reference anchor="XP-DNS"
                 target="http://www.microsoft.com/resources/documentation/windows/xp/all/proddocs/en-us/sag_ip_v6_imp_config_items.mspx">
        <front>
          <title>Windows XP: IPv6 configuration items</title>

          <author fullname="Microsoft" surname="Microsoft">
            <organization></organization>
          </author>

          <date month="5" year="2005" />
        </front>
      </reference>

      <reference anchor="experiment"
                 target="http://www.ietf.org/mail-archive/web/int-area/current/msg01476.html">
        <front>
          <title>practical issues with using v4-mapped addresses for
          nat64</title>

          <author fullname="Marcelo Bagnulo Braun" surname="Braun">
            <organization></organization>
          </author>

          <date month="Jul" year="2008" />
        </front>
      </reference>

      &I-D.boucadair-behave-dns-a64;
    </references>

    <section title="Other Techniques">
      <t><list style="empty">
          <t>[Editor's Note: This section will be removed in a later version
          of this document. It is kept, in this appendix, for reference.]</t>
        </list></t>

      <t>This section discusses other techniques which had been considered to
      avoid DNS64/NAT64 by dual-stack hosts.</t>

      <section title="New DHCP option for 'normal' DNS server">
        <t>Another approach, which requires modification of dual-stack hosts
        which want to avoid the DNS64, is to introduce a new DHCP option.</t>

        <t>This approach feels a little backwards at first. The idea is to
        support unmodified hosts (which might be dual-stack but might be
        IPv6-only) by placing DNS64 servers into the normal <xref
        target="RFC3646">DHCPv6 option for DNS servers</xref>. Then, place the
        'normal' DNS servers into a *new* DHCPv6 option.</t>

        <section title="Host Transition">
          <t>TBD.</t>
        </section>

        <section title="Advantages and Disadvantages">
          <t>Disadvantages:<list style="symbols">
              <t>If dual-stack hosts want to avoid NAT64, they need to be
              modified to understand this new DHCP option. If they aren't
              modified, they will use NAT64.</t>
            </list></t>
        </section>
      </section>

      <section title="Modify Host's Address Selection Rules">
        <t>The <xref target="RFC3484">default address selection rules</xref>
        prefer IPv6 over IPv4. This means, for a dual-stack host, that IPv6
        will be preferred (if available) over IPv4. If a dual-stack host is
        configured to use a DNS64 server, that DNS64 server will synthesize an
        AAAA response if there is an A record. Thus, the dual-stack host will
        always use IPv6 if a DNS lookup was involved, even if IPv4 could have
        been used more optimally. <list style="empty">
            <t>Note: If both a NAT44 and NAT64 are deployed on the same
            network, roughly the same inefficiency occurs (that is, NAT state
            is created). However, it is generally considered better to perform
            NAT44 than NAT64, because NAT64 translates between IP address
            families which can have side effects (e.g., FTP).</t>
          </list></t>

        <t>To avoid this, the host's <xref target="RFC3484">default address
        selection rules</xref> can be modified so that IPv4 is preferred over
        the IPv6/IPv4 translator's prefix. At the same time, native IPv6 can
        still be preferred over IPv4. This is accomplished by adding the
        network's IPv6/IPv4 translator's prefix as the lowest Precedence in
        the address selection rules.</t>

        <t>If the IPv6/IPv4 translator's prefix is the IANA-assigned
        well-known prefix (64:FF9B::/96, as assigned in <xref
        target="RFC6052"></xref>), this can be hard-coded or easily scripted
        into the system startup. However, if the IPv6/IPv4 translator's prefix
        is a network-specific prefix (NSP, as described in <xref
        target="RFC6052"></xref>), the default address selection rules can be
        modified only after the host learns its currently-connected network's
        IPv6/IPv4 translator's prefix (e.g., using <xref
        target="I-D.wing-behave-learn-prefix"></xref>).</t>

        <t>On some operating systems, the address selection rules can be
        configured using a command line utility (e.g., Windows, FreeBSD),
        without new software in the host's IP stack. Other operating systems
        are not as accommodating of this solution (see <xref
        target="address-limitations"></xref>).</t>

        <t><list style="empty">
            <t>Note: it may be desirable to create a standard to adjust a
            host's address selection rules based on the translator's prefix.
            This is a topic for the <xref target="6man">IPv6 maintenance
            working group</xref>. This automatic mechanism may involve
            modifications to the host's IP stack, depending on how the IETF
            chooses to standardize such a mechanism. FOR EXAMPLE, it may be
            useful to consider <xref
            target="I-D.wing-behave-learn-prefix"></xref> (which proposes
            using either DNS or DHCPv6) in conjunction with adjusting the
            host's address selection rules.</t>
          </list></t>

        <section title="Host Transition">
          <t>An IPv6-only and a dual-stack host can both be configured with
          the same address selection rules (namely, both can add the network's
          translator as the lowest Precedence). This is because the IPv6-only
          host will never use IPv4 (because it lacks an IPv4 address) and will
          thus fall through and use the IPv6 address synthesized by the DNS64
          containing the IPv6/IPv4 translator's prefix (that is, as shown in
          the examples, the IPv6-only host will use the Precedence 3 entry in
          the default policy table). The dual-stack host, if it receives an
          AAAA response, will prefer use IPv6; if it receives only an A
          response, it will prefer to use IPv4 (using Precedence 10 for
          IPv4-mapped addresses defined in Section 2.5.4 of <xref
          target="RFC3513"></xref>).</t>
        </section>

        <section anchor="address-limitations"
                 title="Limitations and Advantages">
          <t>The following limitations are observed: <list style="symbols">
              <t>OSX does not implement a <xref target="RFC3484"></xref> or
              <xref target="RFC3484"></xref>-like policy table.</t>

              <t>Some applications implement their own address selection
              rules, effectively ignoring the OS's address selection
              rules.</t>
            </list></t>

          <t>The following advantages are observed: <list style="symbols">
              <t>Causes IPv4 to be preferred over IPv6/IPv4 translator
              addresses, even if DNS was not used to obtain the IPv4 or IPv6
              address (e.g., applications which do not use DNS).</t>
            </list></t>
        </section>

        <section title="Examples">
          <figure>
            <preamble>For example, if a network is using the WKP 64:FF9B::/96
            <xref target="RFC6052"></xref> and a host is using the new default
            policy table from <xref
            target="I-D.arifumi-6man-rfc3484-revise"></xref> (which added
            Precedence 5 for Teredo), the host's new policy table would
            contain one new entry with Precedence 3, as shown
            below:</preamble>

            <artwork align="center"><![CDATA[
Prefix        Precedence Label
::1/128               50     0     # localhost
::/0                  30     2     # IPv6 native
2002::/16             20     3     # 6to4 
::ffff:0:0/96         10     4     # IPv4-mapped 
2001::/32              5     5     # Teredo 
64:FF9B::/96           3     6     # 6/4 translator's prefix
]]></artwork>

            <postamble></postamble>
          </figure>

          <figure>
            <preamble>As another example, if a network has the prefix
            2001:0DB8::/32 and the NAT64 is using the Network-Specific Prefix
            (NSP) 2001:0DB8:AAAA::/96, and the host is using the new default
            policy table from <xref
            target="I-D.arifumi-6man-rfc3484-revise"></xref> (which added
            Precedence 5 for Teredo), the host's new policy table would
            contain one new entry with Precedence 3, as shown
            below:</preamble>

            <artwork align="center"><![CDATA[
Prefix        Precedence Label
::1/128               50     0     # localhost
::/0                  30     2     # IPv6 native
2002::/16             20     3     # 6to4 
::ffff:0:0/96         10     4     # IPv4-mapped 
2001::/32              5     5     # Teredo 
2001:0DB8:AAAA::/96    3     6     # 6/4 translator's prefix
]]></artwork>

            <postamble></postamble>
          </figure>
        </section>
      </section>

      <section title="Indicating AAAA synthesis using DNS response flag">
        <t>Dacheng Zhang's idea.</t>
      </section>

      <section title="New A64 record">
        <t><xref target="I-D.boucadair-behave-dns-a64"></xref></t>
      </section>

      <section title="Use DHCP to Assign Appropriate DNS Server">
        <t><list style="empty">
            <t>Note: due to the limitations of this solution (see <xref
            target="dhcp_limitations"></xref>), it may have little or no
            value.</t>
          </list>To avoid unnecessary traffic through a translator, it is
        desirable to configure IPv4-only and dual-stack hosts with a 'normal'
        DNS recursive resolver.</t>

        <t>However, it is necessary to configure IPv6-only hosts with a <xref
        target="I-D.ietf-behave-dns64">DNS64</xref> recursive resolver so
        those hosts can use an IPv6/IPv4 translator and access servers on the
        IPv4 Internet.</t>

        <t>It is difficult to provide different DNS servers to those types of
        hosts, because there is no existing protocol that declares a host is
        IPv4-only, dual-stack, or IPv6-only.</t>

        <t>This document describes how a network's DHCPv4 and DHCPv6 servers,
        combined with a <xref target="RFC4361">client-identifiers</xref>
        chosen by the host, can determine if a host is IPv4-only, dual-stack,
        or IPv6-only, and assign the correct DNS server according to that
        determination.</t>

        <t><list style="empty">
            <t>Note: the DHCP mechanism described in this section have some
            overlap with the <xref target="mif">Multiple Interfaces Working
            Group</xref> and with <xref
            target="I-D.savolainen-mif-dns-server-selection">split-zone
            DNS</xref>.</t>
          </list></t>

        <t>Both an IPv4-only host and a dual-stack host obtain an IPv4 network
        address. Today, hosts most commonly obtain an IPv4 address using <xref
        target="RFC2131">DHCPv4</xref>. An IPv6-only host does not obtain an
        IPv4 address; however, it may be using DHCPv6 to obtain other
        information (e.g., NTP servers). The following procedure takes
        advantage of that difference to determine if a host is IPv4-only,
        dual-stack, or IPv6-only.</t>

        <section title="Host Requirements">
          <t>The host has the following requirements:<list style="numbers">
              <t>if the host uses IPv4, it MUST use DHCPv4 to learn its IPv4
              address and its DNS server address(es); and,</t>

              <t>if the host uses IPv6, it MUST use DHCPv6 to learn its IPv6
              DNS resolver, using the Information-Request message described in
              Section 18.1.5 of <xref target="RFC3315"></xref> and using <xref
              target="RFC3646"></xref>; and,</t>

              <t>the host MUST use <xref
              target="RFC4361">client-identifiers</xref> to identify itself to
              its DHCP server(s), and MUST use the same client-identifier for
              both DHCPv4 and DHCPv6<list style="empty">
                  <t>Note: This last requirement is stronger than the SHOULD
                  in Section 6.2 of <xref target="RFC4361"></xref></t>
                </list>If the host does not support DHCP authentication, and
              acquires/releases its IPv4 address while keeping its IPv6
              address, it MUST support the procedure described in <xref
              target="host_transition"></xref>; and,</t>

              <t>the host MUST support the <xref target="RFC4242">DHCP
              Information Refresh Time Option</xref>.</t>
            </list></t>
        </section>

        <section anchor="server-req"
                 title="DHCPv4 and DHCPv6 Server Requirements">
          <t>The DHCPv4 and DHCPv6 servers have the following requirements:
          <list style="numbers">
              <t>the DHCPv4 and DHCPv6 servers MUST be able to communicate
              with each other both <xref
              target="RFC4361">client-identifiers</xref> and if an IPv4
              address is assigned to that client-identifier; and,</t>

              <t>If the DHCP server and the host support DHCP authentication,
              the DHCP server MUST support the procedure described in <xref
              target="host_transition"></xref>.</t>

              <t>MUST support the <xref target="RFC4242">DHCP Information
              Refresh Time Option</xref>.</t>
            </list></t>
        </section>

        <section title="DHCP Server Operation">
          <t>If the DHCP server first receives a DHCPv4 request for a
          particular client-identifier, it responds with the 'normal' DNS
          resolver. The DHCPv6 server remembers that RFC4361 client identity
          and if the DHCPv6 server sees a DHCPv6 request from that same client
          identity, it responds to the DHCPv6 request with a 'normal' DNS
          resolver.</t>

          <t>If the DHCP server first receives a DHCPv6 request for a
          particular client-identifier, it responds with a short <xref
          target="RFC4242">information refresh time</xref> (e.g., 30 seconds)
          and a DNS64 recursive resolver. <list style="empty">
              <t>Note-1: This means that during the short information refresh
              time, both a dual-stack host and an IPv6-only will have their
              DNS queries processed by the DNS64 recursive resolver. During
              that time, both the dual-stack host and the IPv6-only host will
              get connectivity to IPv4 servers, but the dual-stack host will
              use the IPv6/IPv4 translator until the information refresh time
              expires.</t>

              <t>Note-2: for discussion: Consider have DHCP server slightly
              delay (e.g., 100ms) responding to a DHCPv6 request. This gives a
              chance for the DHCPv4 request to be received, thus avoiding the
              issue described in Note-1.</t>
            </list></t>

          <t>After the short information refresh time, the DHCPv6 client will
          send a new request. By that time, the DHCPv6 server will have
          either: <list style="letters">
              <t>have seen a DHCPv4 request from the same RFC4361 host. This
              indicates the host supports dual-stack. The DHCP server should
              extend the DHCPv6 lease, and provide a 'normal' DNS server
              (instead of the DNS64 server).</t>

              <t>have not seen a DHCPv4 request from the same RFC4361 host.
              This indicates the host is IPv6-only. The DHCP server should
              extend the DHCPv6 lease and continue providing the same DNS64
              server.</t>
            </list></t>
        </section>

        <section anchor="host_transition" title="Host Transition">
          <t>During natural evolution of a network or because of
          debugging/troubleshooting, a host might transition between
          IPv4-only, dual-stack, or IPv6-only. When the host acquires or
          releases its IPv4 address it transitions to needing a different DNS
          server; if the host has an IPv4 address, it needs a 'normal' DNS
          server and if it does not have an IPv4 address it needs a DNS64
          server.</t>

          <t>There are two transitions considered, where the host
          transitions:<list style="numbers">
              <t>from IPv6-only to IPv4-supporting (that is, IPv4-only or
              dual-stack),</t>

              <t>from IPv4-supporting (that is, IPv4-only or dual-stack) to
              IPv6-only.</t>
            </list></t>

          <t>When doing (1), the DHCPv4 server will provide a 'normal' DNS
          server (because the DHCPv4 server sees the same client-identifier as
          seen by the DHCPv6 server). So case (1) is solved.</t>

          <t>However, when doing (2), the host is giving up its IPv4 address
          and is currently using a normal DNS server, but needs to be told to
          use a DNS64 server instead. There are two mechanisms to provide that
          function, based on the network and host's support of DHCP
          authentication (Section 19.1.1 of <xref target="RFC3315"></xref>)
          <list style="numbers">
              <t>with DHCP authentication: When a certain client identifier
              loses or acquires its IPv4 address and also has an IPv6 address,
              the DHCPv6 server MUST send a <xref target="RFC3315">DHCP
              RECONFIGURE message</xref> to the host and SHOULD include the
              Option Request option indicating the DNS server information has
              changed. The RECONFIGURE message triggers the host to send a new
              Information-Request message to the DHCPv6 server.</t>

              <t>without DHCP authentication: the host, when keeping its IPv6
              address and releasing its IPv4 address, MUST also issue a new
              DHCPv6 Information-Request message to the DHCPv6 server.</t>
            </list>In both cases, the Information-Request message causes the
          DHCPv6 server to reply with a DNS64 recursive resolver, as discussed
          in <xref target="server-req"></xref>.</t>
        </section>

        <section anchor="dhcp_limitations"
                 title="Advantages and Disadvantages">
          <t>Advantages: <list style="symbols">
              <t>Dual-stack applications, which perform DNS lookups, will
              effectively avoid NAT64 when using the 'normal' DNS server.</t>
            </list></t>

          <t>Disadvantages: <list style="symbols">
              <t>A network with mixed IPv4-only/dual-stack hosts and IPv6-only
              hosts needs to have a mix of DNS configurations for those hosts.
              Thus, mechanisms that advertise the same DNS servers to all
              hosts cannot be used on such networks (e.g., IPv6 router
              advertisements).</t>

              <t>If separate networks operate DHCPv4 and DHCPv6 (e.g., as with
              Dual-Stack Lite where the ISP operates DHCPv4 and the customer
              premise router operates DHCPv6), it is likely impossible for the
              DHCPv4 and DHCPv6 servers to communicate necessary information
              with each other.</t>

              <t>Windows does not support <xref target="RFC4361"></xref>.</t>

              <t>OSX does not support DHCPv6.</t>
            </list></t>
        </section>
      </section>

      <section title="New DHCP option for DNS64 server">
        <t>Another approach, which requires modification of IPv6-only hosts
        which need to use the DNS64, is to introduce a new DHCP option.</t>

        <t>The idea is to support unmodified dual-stack hosts (which use the
        normal DNS server provided via <xref target="RFC3646"></xref>), but to
        modify IPv6-only hosts to look for the DNS64 server in a newly-defined
        DHCPv6 option.</t>

        <section title="Advantages and Disadvantages">
          <t>Disadvantages:<list style="symbols">
              <t>Requires modifying IPv6-only hosts, and without this
              modification they won't work at all with a DNS64.</t>
            </list></t>
        </section>
      </section>

      <section title="New DHCP option to identify dual-stack host">
        <t>Dacheng Zhang's idea.</t>
      </section>
    </section>
  </back>
</rfc>

PAFTECH AB 2003-20262026-04-23 19:43:11