draft-ietf-rtcweb-ip-handling.xml

<?xml version="1.0" encoding="us-ascii"?>
<!DOCTYPE rfc SYSTEM "rfc2629.dtd">
<?xml-stylesheet type="text/xsl" href="rfc2629.xslt" ?>
<?rfc toc="yes" ?>
<?rfc symrefs="yes" ?>
<?rfc iprnotified="no" ?>
<?rfc strict="yes" ?>
<?rfc compact="yes" ?>
<?rfc sortrefs="yes" ?>
<?rfc colonspace="yes" ?>
<?rfc rfcedstyle="no" ?>
<?rfc tocdepth="4"?>
<rfc category="std" docName="draft-ietf-rtcweb-ip-handling-08"
ipr="trust200902">
  <front>
    <title abbrev="WebRTC IP Handling">WebRTC IP Address Handling
    Requirements</title>
    <author fullname="Justin Uberti" initials="J." surname="Uberti">
      <organization>Google</organization>
      <address>
        <postal>
          <street>747 6th St S</street>
          <city>Kirkland</city>
          <region>WA</region>
          <code>98033</code>
          <country>USA</country>
        </postal>
        <email>justin@uberti.name</email>
      </address>
    </author>
    <author fullname="Guo-wei Shieh" initials="G." surname="Shieh">
      <organization>Facebook</organization>
      <address>
        <postal>
          <street>1101 Dexter Ave</street>
          <city>Seattle</city>
          <region>WA</region>
          <code>98109</code>
          <country>USA</country>
        </postal>
        <email>guoweis@facebook.com</email>
      </address>
    </author>
    <date day="3" month="June" year="2018" />
    <area>RAI</area>
    <abstract>

      <t>This document provides information and requirements for how IP
      addresses should be handled by WebRTC implementations.</t>
    </abstract>
  </front>
  <middle>
    <section title="Introduction">

      <t>One of WebRTC's key features is its support of peer-to-peer
      connections. However, when establishing such a connection,
      which involves connection attempts from various IP addresses,
      WebRTC may allow a web application to learn additional information about
      the user compared to an application that only uses the
      Hypertext Transfer Protocol (HTTP)
      <xref target="RFC7230" />. This may be problematic in certain cases.
      This document summarizes the concerns, and makes
      recommendations on how WebRTC implementations should best handle the
      tradeoff between privacy and media performance.</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>
    </section>
    <section title="Problem Statement">

      <t>In order to establish a peer-to-peer connection, WebRTC implementations
      use Interactive Connectivity Establishment (ICE)
      <xref target="RFC5245" />, which attempts to discover
      multiple IP addresses using techniques such as
      Session Traversal Utilities for NAT (STUN) <xref target="RFC5389" /> and
      Traversal Using Relays around NAT (TURN) <xref target="RFC5766" />,
      and then checks the connectivity of each
      local-address-remote-address pair in order to select the best one.
      The addresses that are collected usually consist of
      an endpoint's private physical/virtual addresses and its public Internet
      addresses.</t>

      <t>These addresses are exposed upwards to the web application, so that
      they can be communicated to the remote endpoint for its checks.
      This allows the
      application to learn more about the local network configuration than it
      would from a typical HTTP scenario, in which the web server would only
      see a single public Internet address, i.e., the address from which the
      HTTP request was sent.</t>

      <t>The information revealed falls into three categories:
      <list style="numbers">

        <t>If the client is multihomed, additional public IP addresses for the
        client can be learned. In particular,
        if the client tries to hide its physical location through a Virtual
        Private Network (VPN), and the VPN and local OS support routing over
        multiple interfaces (a "split-tunnel" VPN), WebRTC will discover
        not only the public address for the VPN, but also the ISP public address
        over which the VPN is running.</t>

        <t>If the client is behind a Network Address Translator (NAT), the
        client's private IP addresses, often
        <xref target="RFC1918" /> addresses, can be learned.</t>

        <t>If the client is behind a proxy (a client-configured "classical
        application proxy", as defined in
        <xref target="RFC1919" />, Section 3), but direct access to the
        Internet is permitted, WebRTC's STUN
        checks will bypass the proxy and reveal the
        public IP address of the client. This concern
        also applies to the "enterprise TURN server" scenario described in
        <xref target="RFC7478" />, Section 2.3.5.1, if, as above, direct
        Internet access is permitted. However, when the term "proxy"
        is used in this document, it is always in reference to an
        <xref target="RFC1919" /> proxy server.
        </t>
      </list></t>

      <t>Of these three concerns, #1 is the most significant, because for
      some users, the purpose of using a VPN is for anonymity. However,
      different VPN users will have different needs, and some VPN users (e.g.,
      corporate VPN users) may in fact prefer WebRTC to send media traffic
      directly, i.e., not through the VPN.</t>

      <t>#2 is considered to be a less significant concern, given that the
      local address values often contain minimal information (e.g.,
      192.168.0.2), or have built-in privacy protection (e.g., the
      <xref target="RFC4941" /> IPv6 addresses recommended
      by <xref target="I-D.ietf-rtcweb-transports"/>).</t>

      <t>#3 is the least common concern, as proxy administrators can already
      control this behavior through organizational firewall policy, and
      generally, forcing WebRTC traffic through a proxy server will have
      negative effects on both the proxy and on media quality.</t>

      <t>Note also that these concerns predate WebRTC; Adobe Flash Player has
      provided similar functionality since the introduction of RTMFP
      <xref target="RFC7016" /> in 2008.</t>
    </section>
    <section title="Goals">

      <t>WebRTC's support of secure peer-to-peer connections
      facilitates deployment of decentralized systems, which can have privacy
      benefits. As a result, we want to avoid blunt solutions that disable
      WebRTC or make it significantly harder to use. This document
      takes a more nuanced approach, with the following goals:
      <list style="symbols">

        <t>Provide a framework for understanding the problem so that controls
        might be provided to make different tradeoffs regarding performance
        and privacy concerns with WebRTC.</t>
        <t>Using that framework, define settings that enable peer-to-peer
        communications, each with a different balance between performance and
        privacy.</t>
        <t>Finally, provide recommendations for default settings that provide
        reasonable performance without also exposing addressing information
        in a way that might violate user expectations.</t>
      </list></t>
    </section>
    <section title="Detailed Design">
      <section title="Principles">
        <t>The key principles for our framework are stated below:
        <list style="numbers">
          <t>By default, WebRTC traffic should follow typical IP routing, i.e.,
          WebRTC should use the same interface used for HTTP traffic, and only
          the system's 'typical' public addresses (or those of an enterprise
          TURN server, if present) should be visible to the application.
          However, in the interest of optimal media quality,
          it should be possible to enable WebRTC to make use of all
          network interfaces to determine the ideal route.
          </t>

          <t>By default, WebRTC should be able to negotiate direct peer-to-peer
          connections between endpoints (i.e., without traversing a
          NAT or relay server), by providing a minimal set of local IP
          addresses to the application for use in the ICE process.
          This ensures that applications that need true
          peer-to-peer routing for bandwidth or latency reasons can operate
          successfully.
          However, it should be possible to suppress these addresses (with the
          resultant impact on direct connections) if desired.
          </t>

          <t>By default, WebRTC traffic should not be sent through
          proxy servers, due to the media quality problems
          associated with sending WebRTC traffic over TCP, which is almost
          always used when communicating with such proxies, as well as
          proxy performance issues that may result from proxying WebRTC's
          long-lived, high-bandwidth connections. However, it should be possible
          to force WebRTC to send its traffic through a configured proxy if
          desired.
          </t>
        </list></t>
      </section>

      <section title="Modes and Recommendations">
        <t>Based on these ideas, we define four specific modes of WebRTC
        behavior, reflecting different media quality/privacy tradeoffs:
        <list style="format Mode %d:">

          <t>Enumerate all addresses: WebRTC MUST use all network interfaces
          to attempt communication with STUN
          servers, TURN servers, or peers. This will converge on the best media
          path, and is ideal when media performance is the highest priority, but
          it discloses the most information.</t>

          <t>Default route + associated local addresses: WebRTC MUST follow the
          kernel routing table rules, which will typically cause media packets to
          take the same route as the application's HTTP traffic. If an
          application TURN server is present, the preferred route MUST
          be through this TURN server. Once an interface has been chosen,
          the private IPv4 and IPv6 addresses associated with this
          interface MUST be discovered and provided to the
          application. This ensures that direct connections can still be
          established in this mode.</t>

          <t>Default route only: This is the the same as Mode 2, except that the
          associated private addresses MUST NOT be provided; the only IP
          addresses gathered are those discovered via mechanisms like STUN and
          TURN (on the default route). This may cause traffic
          to hairpin through a NAT, fall back to an application TURN server, or
          fail altogether, with resulting quality implications.</t>

          <t>Force proxy: This is the same as Mode 3, but when the application's
          HTTP traffic is sent through a proxy, WebRTC media
          traffic MUST also be proxied.
          If the proxy does not support UDP (as is the case for all HTTP and
          most SOCKS <xref target="RFC1928" /> proxies), or the WebRTC
          implementation does
          not support UDP proxying, the use of UDP will be disabled, and TCP will
          be used to send and receive media through the proxy. Use of TCP will
          result in reduced media quality, in addition to any performance
          considerations associated with sending all WebRTC media through the
          proxy server.</t>
        </list></t>

        <t>Mode 1 MUST only be used when user consent has been provided.
        The details of this
        consent are left to the implementation; one potential mechanism is to tie
        this consent to getUserMedia consent. Alternatively, implementations can
        provide a specific mechanism to obtain user consent.</t>

        <t>In cases where user consent has not been obtained, Mode 2 SHOULD be
        used.</t>

        <t>These defaults provide a reasonable tradeoff that permits trusted
        WebRTC applications to achieve optimal network performance, but gives
        applications without consent (e.g., 1-way streaming or
        data channel applications) only the minimum information needed to
        achieve direct connections, as defined in Mode 2. However,
        implementations MAY choose stricter modes if desired, e.g., if a user
        indicates they want all WebRTC traffic to follow the default route.
        </t>

        <t>Note that the suggested defaults can still be used even for
        organizations that want all
        external WebRTC traffic to traverse a proxy or enterprise TURN server,
        simply by setting an organizational firewall policy that allows
        WebRTC traffic to only leave through the proxy or TURN server.
        This provides a way to ensure the proxy or TURN server is used for any
        external traffic, but still allows direct connections (and, in the
        proxy case, avoids the performance issues associated with forcing
        media through said proxy) for intra-organization traffic.</t>
      </section>
    </section>
    <section title="Implementation Guidance">
      <t>This section provides guidance to WebRTC implementations on how to
        implement the policies described above.</t>
      <section title="Ensuring Normal Routing">
        <t>When trying to follow typical IP routing, the simplest approach is to
          bind the sockets used for peer-to-peer connections to the wildcard
          addresses (0.0.0.0 for IPv4, :: for IPv6), which allows the OS to
          route WebRTC traffic the same way as it would HTTP traffic. STUN and
          TURN will work as usual, and host candidates can still be determined
          as mentioned below.
        </t>
      </section>
      <section title="Determining Host Candidates">
        <t>When binding to a wildcard address, some extra work is needed to
          determine a suitable host candidate, which we define as the
          source address that would be used for any packets sent to the
          web application host (assuming that UDP and TCP get the same
          routing). Use of the web application host as a destination ensures the
          right source address is selected, regardless of where the application
          resides (e.g., on an intranet).
        </t>
        <t>First, the appropriate remote IPv4/IPv6 address is obtained by
        resolving the host component of the web application
        URI <xref target="RFC3986"/>. If the client is behind a proxy and cannot
        resolve these IPs via DNS, the address of the proxy can be used instead.
        Or, if the web application was loaded from a file://
        URI <xref target="RFC8089"/>, rather than over the network, the
        implementation can fall back to a well-known DNS name or IP address.
        </t>
        <t>Once a suitable remote IP has been determined, the implementation
        can create a UDP socket, bind it to the appropriate wildcard address,
        and tell it to connect to the remote IP. Generally, this results in
        the socket being assigned a local address based on the kernel routing
        table, without sending any packets over the network.</t>
        <t>Finally, the socket can be queried using getsockname() or the
        equivalent to determine the appropriate host candidate.</t>
      </section>
    </section>
    <section title="Application Guidance">

      <t>The recommendations mentioned in this document may cause certain
      WebRTC applications to malfunction. In order to be robust in all
      scenarios, the following guidelines are provided for applications:
      <list style="symbols">

        <t>Applications SHOULD deploy a TURN server with support for both UDP
        and TCP connections to the server. This ensures that connectivity can
        still be established, even when Mode 3 or 4 are in use, assuming the
        TURN server can be reached.</t>

        <t>Applications SHOULD detect when they don't have access to the full
        set of ICE candidates by checking for the presence of host candidates.
        If no host candidates are present, Mode 3 or 4 above is in use; this
        knowledge can be useful for diagnostic purposes.</t>
      </list></t>
    </section>
    <section title="Security Considerations">

      <t>This document is entirely devoted to security considerations.</t>
    </section>
    <section title="IANA Considerations">

      <t>This document requires no actions from IANA.</t>
    </section>
    <section title="Acknowledgements">

      <t>Several people provided input into this document, including Bernard
      Aboba, Harald Alvestrand, Ted Hardie, Matthew Kaufmann, Eric Rescorla,
      Adam Roach, and Martin Thomson.</t>
    </section>
  </middle>
  <back>
    <references title="Normative References">
      <?rfc include='reference.RFC.2119.xml'?>
      </references>
    <references title="Informative References">
      <?rfc include='reference.RFC.1918.xml'?>
      <?rfc include='reference.RFC.1919.xml'?>
      <?rfc include='reference.RFC.1928.xml'?>
      <?rfc include='reference.RFC.3986.xml'?>
      <?rfc include='reference.RFC.4941.xml'?>
      <?rfc include='reference.RFC.5245.xml'?>
      <?rfc include='reference.RFC.5389.xml'?>
      <?rfc include='reference.RFC.5766.xml'?>
      <?rfc include='reference.RFC.7016.xml'?>
      <?rfc include='reference.RFC.7230.xml'?>
      <?rfc include='reference.RFC.7478.xml'?>
      <?rfc include='reference.RFC.8089.xml'?>
      <?rfc include='reference.I-D.ietf-rtcweb-transports'?>
      </references>
    <section title="Change log">
      <t>Changes in draft -08:
      <list style="symbols">

        <t>Discuss how enterprise TURN servers should be handled.</t>
      </list></t>

      <t>Changes in draft -07:
      <list style="symbols">

        <t>Clarify consent guidance.</t>
      </list></t>

      <t>Changes in draft -06:
      <list style="symbols">

        <t>Clarify recommendations.</t>

        <t>Split implementation guidance into two sections.</t>
      </list></t>

      <t>Changes in draft -05:
      <list style="symbols">

        <t>Separated framework definition from implementation techniques.</t>

        <t>Removed RETURN references.</t>

        <t>Use origin when determining local IPs, rather than a well-known IP.</t>
      </list></t>

      <t>Changes in draft -04:
      <list style="symbols">

        <t>Rewording and cleanup in abstract, intro, and problem statement.</t>

        <t>Added 2119 boilerplate.</t>

        <t>Fixed weird reference spacing.</t>

        <t>Expanded acronyms on first use.</t>

        <t>Removed 8.8.8.8 mention.</t>

        <t>Removed mention of future browser considerations.</t>
      </list></t>

      <t>Changes in draft -03:
      <list style="symbols">

        <t>Clarified when to use which modes.</t>

        <t>Added 2119 qualifiers to make normative statements.</t>

        <t>Defined 'proxy'.</t>

        <t>Mentioned split tunnels in problem statement.</t>
      </list></t>

      <t>Changes in draft -02:
      <list style="symbols">

        <t>Recommendations -&gt; Requirements</t>

        <t>Updated text regarding consent.</t>
      </list></t>

      <t>Changes in draft -01:
      <list style="symbols">

        <t>Incorporated feedback from Adam Roach; changes to discussion of
        cam/mic permission, as well as use of proxies, and various editorial
        changes.</t>

        <t>Added several more references.</t>
      </list></t>

      <t>Changes in draft -00:
      <list style="symbols">

        <t>Published as WG draft.</t>
      </list></t>
    </section>
  </back>
</rfc>