<?xml version="1.0"?>
<?xml-stylesheet type='text/xsl' href='./rfc2629.xslt' ?>

<?rfc toc="yes"?>
<?rfc symrefs="yes"?>
<?rfc compact="yes" ?>
<?rfc sortrefs="yes" ?>
<?rfc strict="yes" ?>
<?rfc linkmailto="yes" ?>
<!DOCTYPE rfc SYSTEM "rfc2629.dtd" >
    

<?xml-stylesheet type='text/xsl' href='./rfc2629.xslt' ?>

<?rfc symrefs="yes"?>
<?rfc compact="yes" ?>
<?rfc subcompact="yes" ?>
<?rfc sortrefs="yes" ?>
<rfc ipr="trust200902" docName="draft-nir-ipsecme-curve25519-00" category="std">
  <front>
    <title abbrev="Curve25519 for IKEv2">Using Curve25519 for IKEv2 Key Agreement</title>
    <author initials="Y." surname="Nir" fullname="Yoav Nir">
      <organization abbrev="Check Point">Check Point Software Technologies Ltd.</organization>
      <address>
        <postal>
          <street>5 Hasolelim st.</street>
          <city>Tel Aviv</city>
          <code>6789735</code>
          <country>Israel</country>
        </postal>
        <email>ynir.ietf@gmail.com</email>
      </address>
    </author>
    <author initials="S." surname="Josefsson" fullname="Simon Josefsson">
      <organization abbrev="SJD">SJD AB</organization>
      <address>
        <email>simon@josefsson.org</email>
      </address>
    </author>
    <date year="2015"/>
    <area>Security Area</area>
    <keyword>Internet-Draft</keyword>
    <abstract>
      <t> This document describes the use of Curve25519 for ephemeral key 
        exchange in the Internet Key Exchange (IKEv2) protocol.</t>
    </abstract>
  </front>
  <middle>
    <!-- ====================================================================== -->
    <section anchor="introduction" title="Introduction">
      <t> <xref target="CFRG-Curves" /> specifies a new elliptic curve function 
        for use in cryptographic applications. Curve25519 is a Diffie-Hellman 
        function designed with performance and security in mind.</t>
      <t> Almost ten years ago <xref target="RFC4753" /> specified the first 
        elliptic curve Diffie-Hellman groups for the Internet Key Exchange 
        protocol (IKEv2 - <xref target="RFC7296" />). These were the so-called 
        NIST curves. The state of the art has advanced since then. More modern 
        curves allow faster implementations while making it much easier to 
        write constant-time implementations free from side-channel attacks. 
        This document defines such a curve for use in IKE. See 
        <xref target="Curve25519" /> for details about the speed and security 
        of this curve.</t>
      <section anchor="mustshouldmay" title="Conventions Used in This Document">
        <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"/>.</t>
      </section>
    </section>
    <section anchor="crypto" title="Curve25519">
      <t> All cryptographic computations are done using the Curve25519 function 
        defined in <xref target="CFRG-Curves" />.  In this document, this 
        function is considered a black box that takes for input a (secret key, 
        public key) pair and outputs a public key.  Public keys are defined as 
        strings of 32 octets.  Secret keys are defined as 255-bit numbers such 
        that high-order bit (bit 254) is set, and the three lowest-order bits 
        are unset.  In addition, a common public key, denoted by G, is shared 
        by all users.</t>
      <t> An ephemeral Diffie-Hellman key exchange using Curve25519 goes as 
        follows: Each party picks a secret key d uniformly at random and 
        computes the corresponding public key:<figure>
      		<artwork><![CDATA[
   x_mine = Curve25519(d, G)
]]></artwork></figure></t>
      <t> Parties exchange their public keys (see <xref target="ke_format" />) 
        and compute a shared secret:<figure>
      		<artwork><![CDATA[
   SHARED_SECRET = Curve25519(d, x_peer).
]]></artwork></figure></t>
      <t> This shared secret is used directly as the value denoted g^ir in 
        section 2.14 of RFC 7296. It is always exactly 32 octets when 
        Curve25519 is used.</t>
      <t> A complete description of the Curve25519 function, as well as a few 
        implementation notes, are provided in <xref target="curve25519func" />.</t>
    </section> 
    <section anchor="in_ikev2" title="Use and Negotiation in IKEv2">
      <t> The use of Curve25519 in IKEv2 is negotiated using a Transform Type 4 
        (Diffie-Hellman group) in the SA payload of either an IKE_SA_INIT or a 
        CREATE_CHILD_SA exchange.</t>
      <section anchor="ke_format" title="Key Exchange Payload">
        <t> The diagram for the Key Exchange Payload from section 3.4 of RFC 
          7296 is copied below for convenience:<figure>
      		<artwork><![CDATA[
                        1                   2                   3
    0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   | Next Payload  |C|  RESERVED   |         Payload Length        |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |   Diffie-Hellman Group Num    |           RESERVED            |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                                                               |
   ~                       Key Exchange Data                       ~
   |                                                               |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
]]></artwork></figure></t>
        <t><list style="symbols">
          <t> Payload Length - Since a Curve25519 public key is 32 octets, the 
            Payload Length is always 40.</t>
          <t> The Diffie-Hellman Group Num is xx for Curve25519 (TBA by IANA)</t>
          <t> The Key Exchange Data is 32 octets encoded as an array of bytes 
            in little-endian order as described in section 8 of 
            <xref target="CFRG-Curves" /></t></list></t>
      </section> 
      <section anchor="rec_test" title="Recipient Tests">
        <t> This section describes the checks that a recipient of a public key 
          needs to perform. It is the equivalent of the tests described in 
          <xref target="RFC6989" /> for other Diffie-Hellman groups.</t>
        <t> Curve25519 was designed in a way that the result of Curve25519(x, d) 
          will never reveal information about d, provided is was chosen as 
          prescribed, for any value of x.</t>
        <t> Define legitimate values of x as the values that can be obtained as 
          x = Curve25519(G, d') for some d, and call the other values 
          illegitimate.  The definition of the Curve25519 function shows that 
          legitimate values all share the following property: the high-order 
          bit of the last byte is not set.</t>
        <t> Since there are some implementation of the Curve25519 function that 
          impose this restriction on their input and others that don't, 
          implementations of Curve25519 in IKE SHOULD reject public keys when 
          the high-order bit of the last byte is set (in other words, when the 
          value of the leftmost byte is greater than 0x7F) in order to prevent 
          implementation fingerprinting.</t>
        <t> Other than this recommended check, implementations do not need to 
          ensure that the public keys they receive are legitimate: this is not 
          necessary for security with Curve25519.</t>
      </section>
    </section> 
    <section anchor="security" title="Security Considerations">
      <t> Curve25519 is designed to facilitate the production of 
        high-performance constant-time implementations of the Curve25519 
        function. Implementors are encouraged to use a constant-time 
        implementation of the Curve25519 function.  This point is of crucial 
        importance if the implementation chooses to reuse its supposedly 
        ephemeral key pair for many key exchanges, which some implementations 
        do in order to improve performance.</t>
      <t> Curve25519 is believed to be at least as secure as the 256-bit random 
        ECP group (group 19) defined in RFC 4753, also known as NIST P-256.  
        While the NIST curves are advertised as being chosen verifiably at 
        random, there is no explanation for the seeds used to generate them. In 
        contrast, the process used to pick Curve25519 is fully documented and 
        rigid enough so that independent verification has been done.  This is 
        widely seen as a security advantage for Curve25519, since it prevents 
        the generating party from maliciously manipulating the parameters.</t>
      <t> Another family of curves available in IKE, generated in a fully 
        verifiable way, is the Brainpool curves <xref target="RFC6954" />.  
        Specifically, brainpoolP256 (group 28) is expected to provide a level 
        of security comparable to Curve25519 and NIST P-256.  However, due to 
        the use of pseudo-random prime, it is significantly slower than NIST 
        P-256, which is itself slower than Curve25519.</t>
    </section>
    <section anchor="iana" title="IANA Considerations">
      <t> IANA is requested to assign one value from the IKEv2 
        "Transform Type 4 - Diffie-Hellman Group Transform IDs" registry, with 
        name Curve25519, and this document as reference. The Recipient Tests 
        field should also point to this document.</t>
    </section>  
    <section anchor="ack" title="Acknowledgements">
      <t> Curve25519 was designed by D. J. Bernstein and Tanja Lange. The 
        specification of wire format is by Sean Turner, Rich Salz, and Watson 
        Ladd, with Adam Langley editing the current document. Much of the text 
        in this document is copied from Simon's draft for the TLS working group.</t>
    </section>    
  </middle>
  <!-- ====================================================================== -->
  <back>
    <references title="Normative References"> 
      <reference anchor='RFC2119'>
        <front>
          <title abbrev='RFC Key Words'>Key words for use in RFCs to Indicate Requirement Levels</title>
          <author initials='S.' surname='Bradner' fullname='Scott Bradner'>
            <organization>Harvard University</organization>
            <address>
              <postal>
                <street>1350 Mass. Ave.</street>
                <street>Cambridge</street>
                <street>MA 02138</street>
              </postal>
              <phone>- +1 617 495 3864</phone>
              <email>sob@harvard.edu</email>
            </address>
          </author>
          <date year='1997' month='March' />
          <area>General</area>
          <keyword>keyword</keyword>
        </front>
        <seriesInfo name='BCP' value='14' />
        <seriesInfo name='RFC' value='2119' />
        <format type='TXT' octets='4723' target='ftp://ftp.isi.edu/in-notes/rfc2119.txt' />
        <format type='HTML' octets='16553' target='http://tools.ietf.org/html/rfc2119' />
      </reference>
      <reference anchor="RFC7296">
        <front>
          <title>Internet Key Exchange Protocol Version 2 (IKEv2)</title>
          <author initials="T." surname="Kivinen" fullname="Tero Kivinen">
            <organization/>
          </author>
          <author initials="C." surname="Kaufman" fullname="C. Kaufman">
            <organization/>
          </author>
          <author initials="P." surname="Hoffman" fullname="P. Hoffman">
            <organization/>
          </author>
          <author initials="Y." surname="Nir" fullname="Y. Nir">
            <organization/>
          </author>
          <author initials="P." surname="Eronen" fullname="P. Eronen">
            <organization/>
          </author>
          <date year="2014" month="October"/>
        </front>
        <seriesInfo name="RFC" value="7296"/>
        <format type="HTML" target="https://tools.ietf.org/html/rfc5996"/>
      </reference>
      <reference anchor="CFRG-Curves">
        <front>
          <title>Elliptic Curves for Security</title>
            <author initials="A" surname="Langley" fullname="Adam Langley"></author>
            <date month="January" day="6" year="2015"/>
          </front>
        <seriesInfo name="Internet-Draft" value="draft-agl-cfrgcurve-00"/>
        <format type="TXT" target="http://www.ietf.org/internet-drafts/draft-agl-cfrgcurve-00.txt"/>
      </reference>
    </references>
    <references title="Informative References"> 
      <reference anchor="RFC4753">
        <front>
          <title>ECP Groups For IKE and IKEv2</title>
          <author initials="D." surname="Fu" fullname="D. Fu">
          </author>
          <author initials="J." surname="Solinas" fullname="J. Solinas">
          </author>
          <date year="2007" month="January"/>
        </front>
        <seriesInfo name="RFC" value="4753"/>
        <format type="TXT" octets="28760" target="http://www.rfc-editor.org/rfc/rfc4753.txt"/>
      </reference>
      <reference anchor="Curve25519" target="http://dx.doi.org/10.1007/11745853_14">
        <front>
          <title>Curve25519: New Diffie-Hellman Speed Records</title>
          <author initials="J." surname="Bernstein"/>
          <date year="2006" month="February" />
        </front>
        <seriesInfo name="LNCS" value="3958"/>
      </reference>
      <reference anchor="EFD"
        target="http://www.hyperelliptic.org/EFD/g1p/auto-montgom-xz.html">
         <front>
            <title>Explicit-Formulas Database: XZ coordinates for
               Montgomery curves</title>
            <author initials="D.J." surname="Bernstein"
               fullname="Daniel J. bernstein">
               <organization />
            </author>
            <author initials="T." surname="Lange" fullname="Tanja Lange">
               <organization />
            </author>
            <date month="January" year="2014"/>
         </front>
      </reference>
      <reference anchor="NaCl"
        target="http://cr.yp.to/highspeed/naclcrypto-20090310.pdf">
         <front>
            <title>Cryptography in NaCL</title>
            <author initials="D.J." surname="Bernstein"
               fullname="Daniel J. bernstein">
               <organization />
            </author>
            <date month="March" year="2013"/>
         </front>
      </reference>
      <reference anchor="RFC6954">
        <front>
          <title>
          Using the Elliptic Curve Cryptography (ECC) Brainpool Curves for the Internet Key Exchange Protocol Version 2 (IKEv2)
          </title>
          <author initials="J." surname="Merkle" fullname="J. Merkle" />
          <author initials="M." surname="Lochter" fullname="M. Lochter" />
          <date year="2013" month="July"/>
        </front>
        <seriesInfo name="RFC" value="6954"/>
        <format type="TXT" octets="20366" target="http://www.rfc-editor.org/rfc/rfc6954.txt"/>
      </reference>
      <reference anchor="RFC6989">
        <front>
          <title>
          Additional Diffie-Hellman Tests for the Internet Key Exchange Protocol Version 2 (IKEv2)
          </title>
          <author initials="Y." surname="Sheffer" fullname="Y. Sheffer"><organization/></author>
          <author initials="S." surname="Fluhrer" fullname="S. Fluhrer"><organization/></author>
          <date year="2013" month="July"/>
        </front>
        <seriesInfo name="RFC" value="6989"/>
        <format type="TXT" octets="21099" target="http://www.rfc-editor.org/rfc/rfc6989.txt"/>
      </reference>
    </references>
    <!-- ====================================================================== -->
    <section anchor="curve25519func" title="The curve25519 function">

      <section title="Formulas">

        <t>This section completes <xref target="crypto" /> by defining
          the Curve25519 function and the common public key G.
          It is meant as an alternative, self-contained specification
          for the Curve25519 function, possibly easier to follow than
          the original paper for most implementors.</t>

        <section anchor="field" title="Field Arithmetic">

          <t>Throughout this section, P denotes the integer 2^255-19 =
            0x7FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFED.
            The letters X and Z, and their numbered variants such as x1,
            z2, etc. denote integers modulo P, that is integers
            between 0 and P-1 and every operation between them is
            implictly done modulo P. For addition, subtraction and
            multiplication this means doing the operation in the usual
            way and then replacing the result with the remainder of its
            division by P. For division, "X / Z" means mutliplying (mod P) X
            by the modular inverse of Z mod P.</t>

          <t>A convenient way to define the modular inverse of Z mod P is
            as Z^(P-2) mod P, that is Z to the power of 2^255-21 mod P. It
            is also a practical way of computing it, using a
            square-and-multiply method.</t>

          <t>The four operations +, -, *, / modulo P are known as the
            field operations. Techniques for efficient implementation
            of the field operations are outside the scope of this
            document.</t>

        </section>

        <section title="Conversion to and from internal format">

          <t>For the purpose of this section, we will define a
            Curve25519 point as a pair (X, Z) were X and Z are integers
            mod P (as defined above). Though public keys were defined to
            be strings of 32 bytes, internally they are represented as
            curve points. This subsection describes the conversion
            process as two functions: PubkeyToPoint and PointToPubkey.</t>

          <figure>
            <artwork><![CDATA[
    PubkeyToPoint:
    Input: a public key b_0, ..., b_31
    Output: a Curve25519 point (X, Z)
        1. Set X = b_0 + 256 * b_1 + ... + 256^31 * b_31 mod P
        2. Set Z = 1
        3. Output (X, Z)
              ]]></artwork>
          </figure>

          <figure>
            <artwork><![CDATA[
    PointToPubkey:
    Input: a Curve25519 point (X, Z)
    Output: a public key b_0, ..., b_31
        1. Set x1 = X / Z mod P
        2. Set b_0, ... b_31 such that
            x1 = b_0 + 256 * b_1 + ... + 256^31 * b_31 mod P
        3. Output b_0, ..., b_31
              ]]></artwork>
          </figure>

        </section>

        <section title="Scalar Multiplication">

          <t>We first introduce the DoubleAndAdd function, defined as
            follows (formulas taken from <xref target="EFD" />).</t>

          <figure>
            <artwork><![CDATA[
    DoubleAndAdd:
    Input: two points (X2, Z2), (X3, Z3), and an integer mod P: X1
    Output: two points (X4, Z4), (X5, Z5)
    Constant: the integer mod P: a24 = 121666 = 0x01DB42
    Variables: A, AA, B, BB, E, C, D, DA, CB are integers mod P
        1. Do the following computations mod P:
            A  = X2 + Z2
            AA = A2
            B  = X2 - Z2
            BB = B2
            E  = AA - BB
            C  = X3 + Z3
            D  = X3 - Z3
            DA = D * A
            CB = C * B
            X5 = (DA + CB)^2
            Z5 = X1 * (DA - CB)^2
            X4 = AA * BB
            Z4 = E * (BB + a24 * E)
        2. Output (X4, Z4) and (X5, Z5)
              ]]></artwork>
          </figure>

          <t>This may be taken as the abstract definition of an
            arbitrary-looking function. However, let's mention "the true
            meaning" of this function, without justification, in order
            to help the reader make more sense of it.  It is possible to
            define operations "+" and "-" between Curve25519 points.
            Then, assuming (X2, Z2) - (X3, Z3) = (X1, 1), the
            DoubleAndAdd function returns points such that (X4, Z4) =
            (X2, Z2) + (X2, Z2) and (X5, Z5) = (X2, Z2) + (X3, Z3).</t>

          <t>Taking the "+" operation as granted, we can define
            multiplication of a Curve25519 point by a positive integer
            as N * (X, Z) = (X, Z) + ... + (X, Z), with N point
            additions. It is possible to compute this operation, known
            as scalar multiplication, using an algorithm called the
            Montgomery ladder, as follows.</t>

          <figure>
            <artwork><![CDATA[
    ScalarMult:
    Input: a Curve25519 point: (X, 1) and a 255-bits integer: N
    Output: a point (X1, Z1)
    Variable: a point (X2, Z2)
        0. View N as a sequence of bits b_254, ..., b_0,
            with b_254 the most significant bit
            and b_0 the least significant bit.
        1. Set X1 = 1 and Z1 = 0
        2. Set X2 = X and Z2 = 1
        3. For i from 254 downwards to 0, do:
            If b_i == 0, then:
                Set (X2, Z2) and (X1, Z1) to the output of
                DoubleAndAdd((X2, Z2), (X1, Z1), X)
            else:
                Set (X1, Z1) and (X2, Z2) to the output of
                DoubleAndAdd((X1, Z1), (X2, Z2), X)
        4. Output (X1, Z1)
              ]]></artwork>
          </figure>

        </section>

        <section title="Conclusion">

          <t>We are now ready to define the Curve25519 function
            itself.</t>

          <figure>
            <artwork><![CDATA[
    Curve25519:
    Input: a public key P and a secret key S
    Output: a public key Q
    Variables: two Curve25519 points (X, Z) and (X1, Z1)
        1. Set (X, Z) = PubkeyToPoint(P)
        2. Set (X1, Z1) = ScalarMult((X, Z), S)
        3. Set Q = PointToPubkey((X1, Z1))
        4. Output Q
              ]]></artwork>
          </figure>

          <t>The common public key G mentioned in the first paragraph of
            <xref target="crypto" /> is defined as G = PointToPubkey((9,
            1).</t>

        </section>

      </section>

      <section title="Test vectors">

        <t>The following test vectors are taken from <xref target="NaCl"
            />. Compared to this reference, the private key strings
          have been applied the ClampC function of the reference and
          converted to integers in order to fit the description given in
          <xref target="Curve25519" /> and the present memo.</t>

        <t>The secret key of party A is denoted by S_a, it public key by
          P_a, and similarly for party B. The shared secret is SS.</t>

        <figure>
          <artwork><![CDATA[
          S_a = 0x6A2CB91DA5FB77B12A99C0EB872F4CDF
                  4566B25172C1163C7DA518730A6D0770

          P_a = 85 20 F0 09 89 30 A7 54 74 8B 7D DC B4 3E F7 5A
                0D BF 3A 0D 26 38 1A F4 EB A4 A9 8E AA 9B 4E 6A

          S_b = 0x6BE088FF278B2F1CFDB6182629B13B6F
                  E60E80838B7FE1794B8A4A627E08AB58

          P_b = DE 9E DB 7D 7B 7D C1 B4 D3 5B 61 C2 EC E4 35 37
                3F 83 43 C8 5B 78 67 4D AD FC 7E 14 6F 88 2B 4F

           SS = 4A 5D 9D 5B A4 CE 2D E1 72 8E 3B F4 80 35 0F 25
                E0 7E 21 C9 47 D1 9E 33 76 F0 9B 3C 1E 16 17 42
            ]]></artwork>
        </figure>

      </section>

      <section title="Side-channel considerations">

        <t>Curve25519 was specifically designed so that correct, fast,
          constant-time implementations are easier to produce. In
          particular, using a Montgomery ladder as described in the
          previous section ensures that, for any valid value of the
          secret key, the same sequence of field operations are
          performed, which eliminates a major source of side-channel
          leakage.</t>

        <t>However, merely using Curve25519 with a Montgomery ladder does
          not prevent all side-channels by itself, and some point are the
          responsibility of implementors:
          <list style="numbers">
            <t>In step 3 of SclarMult, avoid branches depending on
              b_i, as well as memory access patterns depending on b_i,
              for example by using safe conditional swaps on the inputs
              and outputs of DoubleAndAdd.</t>
            <t>Avoid data-dependant branches and memory access patterns
              in the implementation of field operations.</t>
          </list>
        </t>

        <t>Techniques for implementing the field operations in constant
          time and with high performance are out of scope of this
          document. Let's mention however that, provided constant-time
          multiplication is available, division can be computed in
          constant time using exponentiation as described in <xref
            target="field" />.</t>

        <t>If using constant-time implementations of the field
          operations is not convenient, an option to reduce the
          information leaked this way is to replace step 2 of the
          SclarMult function with:</t>
        <figure>
          <artwork><![CDATA[
        2a. Pick Z uniformly randomly between 1 and P-1 included
        2b. Set X2 = X * Z and Z2 = Z
            ]]></artwork>
        </figure>
        <t>This method is known as randomizing projective coordinates.
          However, it is no guaranteed to avoid all side-channel leaks
          related to field operations.</t>

        <t>Side-channel attacks are an active reseach domain that still
          sees new significant results, so implementors of the
          Curve25519 function are advised to follow recent security
          research closely.</t>

      </section>

    </section>

  </back>
</rfc>

