<?xml version='1.0' encoding='UTF-8'?>
<!DOCTYPE xep SYSTEM 'xep.dtd' [
  <!ENTITY % ents SYSTEM 'xep.ent'>
%ents;
]>
<?xml-stylesheet type='text/xsl' href='xep.xsl'?>
<xep>
  <header>
    <title>Jabber HTTP Polling</title>
    <abstract>This document defines an XMPP protocol extension that enables access to a Jabber server from behind firewalls which do not allow outgoing sockets on port 5222, via HTTP requests.</abstract>
    &LEGALNOTICE;
    <number>0025</number>
    <status>Obsolete</status>
    <type>Historical</type>
    <sig>Standards</sig>
    <dependencies>
      <spec>XMPP Core</spec>
    </dependencies>
    <supersedes/>
    <supersededby>
      <spec>XEP-0124</spec>
    </supersededby>
    <shortname>httppoll</shortname>
    &hildjj;
    <author>
      <firstname>Craig</firstname>
      <surname>Kaes</surname>
      <email>ckaes@jabber.com</email>
      <jid>ckaes@corp.jabber.com</jid>
    </author>
    <author>
      <firstname>David</firstname>
      <surname>Waite</surname>
      <email>mass@akuma.org</email>
      <jid>mass@akuma.org</jid>
    </author>
    <revision>
      <version>1.2</version>
      <date>2009-06-03</date>
      <initials>psa</initials>
      <remark><p>Per a vote of the XMPP Council, changed status to Obsolete.</p></remark>
    </revision>
    <revision>
      <version>1.1</version>
      <date>2006-07-26</date>
      <initials>psa</initials>
      <remark><p>Per a vote of the Jabber Council, changed status to Deprecated.</p></remark>
    </revision>
    <revision>
      <version>1.0</version>
      <date>2002-10-11</date>
      <initials>psa</initials>
      <remark><p>Per a vote of the Jabber Council, advanced status to Active.</p></remark>
    </revision>
    <revision>
      <version>0.2</version>
      <date>2002-09-23</date>
      <initials>dew</initials>
      <remark><p>Changed format to allow socket-equivalent security.</p></remark>
    </revision>
    <revision>
      <version>0.1</version>
      <date>2002-03-14</date>
      <initials>jjh</initials>
      <remark><p>Initial version.</p></remark>
    </revision>
  </header>
  <section1 topic="Introduction">
    <p><em>Note Well: This protocol specified in this document has been superseded by the protocol specified in &xep0124;.</em></p>
    <p>
      This specification documents a method to allow Jabber clients to access Jabber
      servers from behind existing firewalls. Although several similar methods
      have been proposed, this approach should work through all known firewall
      configurations which allow outbound HTTP access.
    </p>
  </section1>
  <section1 topic="Background">
    <p>
      In general, a firewall is a box that protects a network from outsiders,
      by controlling the IP connections that are allowed to pass through the
      box. Often, a firewall will also allow access outside only by proxy,
      either explicit proxy support or implicit through Network Address
      Translation (NAT).
    </p>
    <p>
      In the interest of security, many firewall administrators do not allow
      outbound connections to unknown and unused ports. Until Jabber becomes
      more widely deployed, port 5222/tcp (for Jabber client connections) will
      often be blocked.
    </p>
    <p>
      The best solution for sites that are concerned about security is to run
      their own Jabber server, either inside the firewall, or in a DMZ
      <note>
        DMZ definition at
        <link
          url="http://searchwebmanagement.techtarget.com/sDefinition/0,,sid27_gci213891,00.html">
          searchwebmanagement.com
        </link>
      </note>
      network. However, there are network configuration where an external
      Jabber server must still be used and port 5222/tcp outbound cannot be
      allowed. In these situations, different methods for connecting to a
      Jabber server are required. Several methods exist today for doing this
      traversal. Most rely on the fact that a most firewalls are configured to
      allow access through port 80/tcp. Although some less-complicated
      firewalls will allow any protocol to traverse this port, many will proxy,
      filter, and verify requests on this port as HTTP. Because of this, a
      normal Jabber connection on port 80/tcp will not suffice.
    </p>
    <p>
      In addition, many firewalls/proxy servers will also not allow or not
      honor HTTP Keep-alives (as defined in section 19.7.1.1 of &rfc2068;)
      and will consider long-lived socket connections as security issues.
      Because of this the traditional Jabber connection model, where one
      socket is one stream is one session, will not work reliably.
    </p>
    <p>
      In light of all of the ways that default firewall rules can interfere
      with Jabber connectivity, a lowest-common denominator approach was
      selected. HTTP is used to send XML as POST requests and receieve pending
      XML within the responses. Additional information is prepended in the
      request body to ensure an equivalent level of security to TCP/IP sockets.
    </p>
  </section1>
  <section1 topic="Normal data transfer">
    <p>
      The client makes HTTP requests periodically to the server. Whenever the
      client has something to send, that XML is included in the body of the
      request. When the server has something to send to the client, it must be
      contained in the body of the response.
    </p>
    <p>
      In some browser/platform combinations, sending cookies from the client is
      not possible due to design choices and limitations in the
      browser. Therefore, a work-around was needed to support clients based on
      these application platforms.
    </p>
    <p>
      All requests to the server are HTTP POST requests, with Content-Type:
      application/x-www-form-urlencoded. Responses from the server have
      Content-Type: text/xml. Both the request and response bodies are UTF-8
      encoded text, even if an HTTP header to the contrary exists. All
      responses contain a Set-Cookie header with an identifier, which is sent
      along with future requests as described below. This identifier cookie
      must have a name of 'ID'. The first request to a server always uses 0 as
      the identifier. The server must always return a 200 response code,
      sending any session errors as specially-formatted identifiers.
    </p>
    <p>The client sends requests with bodies in the following format:</p>
      <code>
          identifier ; key [ ; new_key] , [xml_body]
      </code>
      <p>If the identifier is zero, key indicates an initial key. In this case,
      new_key should not be specified, and must be ignored.</p>
      <table caption="Request Values">
        <tr>
          <th>Identifier</th>
          <th>Purpose</th>
        </tr>
        <tr>
          <td>identifier</td>
          <td>
            To uniquely identify the session server-side. This field is only
            used to identify the session, and provides no security.
          </td>
        </tr>
        <tr>
          <td>key</td>
          <td>
            To verify this request is from the originator of the session. The
            client generates a new key in the manner described below for each
            request, which the server then verifies before processing the
            request.
          </td>
        </tr>
        <tr>
          <td>new_key</td>
          <td>
            The key algorithm can exhaust valid keys in a sequence, which
            requires a new key sequence to be used in order to continue the
            session. The new key is sent along with the last used key in the
            old sequence.
          </td>
        </tr>
        <tr>
          <td>xml_body</td>
          <td>
            The body of text to send. Since a POST must be sent in order for
            the server to respond with recent messages, a client may send
            a request without an xml_body in order to just retrieve new
            incoming packets. This is not required to be a full XML document or
            XML fragment, it does not need to start or end on element boundaries.
          </td>
        </tr>
      </table>
    <p>
      The identifier is everything before the first semicolon, and must consist
      of the characters [A-Za-z0-9:-]. The identifier returned from the first
      request is the identifier for the session. Any new identifier that ends
      in ':0' indicates an error, with the entire identifier indicating the
      specific error condition. Any new identifier that does not end in ':0' is
      a server programming error, the client should discontinue the
      session. For new sessions, the client identifier is considered to be 0.
    </p>
    <section2 topic="Error conditions">
      <p>
        Any identifier that ends in ':0' indicates an error. Any previous
        identifier associated with this session is no longer valid.
      </p>
      <section3 topic="Unknown Error">
        <p>
          Server returns ID=0:0. The response body can contain a textual error
          message.
        </p>
      </section3>
      <section3 topic="Server Error">
        <p>Server returns ID=-1:0</p>
      </section3>
      <section3 topic="Bad Request">
        <p>Server returns ID=-2:0</p>
      </section3>
      <section3 topic="Key Sequence Error">
        <p>Server returns ID=-3:0</p>
      </section3>
    </section2>
    <p>
      The key is a client security feature to allow TCP/IP socket equivalent
      security. It does not protect against intermediary attacks, but does
      prevent a person who is capable of listening to the HTTP traffic from
      sending messages and receiving incoming traffic from another machine.
    </p>
    <p>The key algorithm should be familiar with those with knowledge of Jabber zero-knowledge authentication.</p>
    <code>
        K(n, seed) = Base64Encode(SHA1(K(n - 1, seed))), for n &gt; 0
        K(0, seed) = seed, which is client-determined
    </code>
    <p>Note: Base64 encoding is defined in &rfc3548;. SHA1 is defined in &rfc3174;.</p>
    <p>
      No framing is implied by a single request or reply. A single request can
      have no content sent, in which case the body contains only the identifier
      followed by a comma. A reply may have no content to send, in which case
      the body is empty. Zero or more XMPP packets may be sent in a single
      request or reply, including partial XMPP packets.
    </p>
    <p>
      The absense of a long-lived connection requires the server to consider
      client traffic as a heartbeat to keep the session alive. If a
      server-configurable period of time passes without a successful POST
      request sent by the client, the server must end the client session. Any
      client requests using the identifier associated with that now dead
      session must return an error of '0:0'.
    </p>
    <p>
      The maximum period of time to keep a client session active without an
      incoming POST request is not defined, but five minutes is the recommended
      minimum. The maximum period of time recommended for clients between
      requests is two minutes; if the client has not sent any XML out for two
      minutes, a request without an XML body should be sent. If a client is
      disconnecting from the server, a closing &lt;stream:stream&gt; must be
      sent to end the session. Failure to do this may have the client continue
      to be represented to other users as available.
    </p>
    <p>
      If the server disconnects the user do to a session timeout, the server
      MUST bounce pending IQ requests and either bounce or store offline
      incoming messages.
    </p>
  </section1>
  <section1 topic="Usage">
    <p>The following is the sequence used for client communication:</p>
      <ol>
        <li>
          The client generates some initial K(0, seed) and runs the algorithm
          above 'n' times to determine the initial key sent to the server,
          K(n, seed)
        </li>
        <li>
          The client sends the request to the server to start the stream,
          including an identifier with a value of zero and K(n, seed)
        </li>
        <li>
          The server responds with the session identifier in the headers
          (within the Set-Cookie field).
        </li>
        <li>
          For each further request done by the client, the identifier from the
          server and K(n - 1, seed) are sent along.
        </li>
        <li>
          The server verifies the incoming value by generating
          K(1, incoming_value), and verifying that value against the value sent
          along with the last client request. If the values do not match, the
          request should be ignored or logged, with an error code being
          returned of -3:0. The request must not be processed, and must not
          extend the session keepalive.
        </li>
        <li>
          The client may send a new key K(m, seed') at any point, but should
          do this for n &gt; 0 and must do this for n = 0. If K(0, seed) is
          sent without a new key, the client will not be able to continue the
          session.
        </li>
      </ol>
    <example caption="Initial request (without keys)">

<![CDATA[POST /wc12/webclient HTTP/1.1
Content-Type: application/x-www-form-urlencoded
Host: webim.jabber.com

0,<stream:stream to="jabber.com"
  xmlns="jabber:client"
  xmlns:stream="http://etherx.jabber.org/streams">]]>

    </example>
    <example caption="Initial response">

<![CDATA[Date: Fri, 15 Mar 2002 20:30:30 GMT
Server: Apache/1.3.20
Set-Cookie: ID=7776:2054; path=/webclient/; expires=-1
Content-Type: text/xml

<?xml version='1.0'?>
<stream:stream xmlns:stream='http://etherx.jabber.org/streams'
  id='3C9258BB'
  xmlns='jabber:client' from='jabber.com'>]]>

    </example>
    <example caption="Next request (without keys)">

<![CDATA[POST /wc12/webclient HTTP/1.1
Content-Type: application/x-www-form-urlencoded
Host: webim.jabber.com

7776:2054,<iq type="get" id="WEBCLIENT3">
  <query xmlns="jabber:iq:auth">
    <username>hildjj</username>
  </query>
</iq>]]>

    </example>
    <example caption='key sequence'>
      K(0, "foo") = "foo"
      K(1, "foo") = "C+7Hteo/D9vJXQ3UfzxbwnXaijM="
      K(2, "foo") = "6UU8CDmH3O4aHFmCqSORCn721+M="
      K(3, "foo") = "vFFYSOhGyaGUgLrldtMBX7x91Wc="
      K(4, "foo") = "ZaDxCilBVTHS9dJfbBo1NsC2b+8="
      K(5, "foo") = "moPFsvHytDGiJQOjp186AMXAeP0="
      K(6, "foo") = "VvxEk07IFy6hUmG/PPBlTLE2fiA="
    </example>
      <example caption="Initial request (with keys)">

<![CDATA[POST /wc12/webclient HTTP/1.1
Content-Type: application/x-www-form-urlencoded
Host: webim.jabber.com

0;VvxEk07IFy6hUmG/PPBlTLE2fiA=,<stream:stream to="jabber.com"
  xmlns="jabber:client"
  xmlns:stream="http://etherx.jabber.org/streams">]]>

      </example>
      <example caption="Next request (with keys)">

<![CDATA[POST /wc12/webclient HTTP/1.1
Content-Type: application/x-www-form-urlencoded
Host: webim.jabber.com

7776:2054;moPFsvHytDGiJQOjp186AMXAeP0=,<iq type="get" id="WEBCLIENT3">
  <query xmlns="jabber:iq:auth">
    <username>hildjj</username>
  </query>
</iq>]]>

    </example>
    <example caption='Changing key'>
<![CDATA[POST /wc12/webclient HTTP/1.1
Content-Type: application/x-www-form-urlencoded
Host: webim.jabber.com

7776:2054;C+7Hteo/D9vJXQ3UfzxbwnXaijM=;Tr697Eff02+32FZp38Xaq2+3Bv4=,<presence/>]]>
    </example>
  </section1>
  <section1 topic="Known issues">
    <ul>
      <li>This method works over HTTPS, which is good from the standpoint of functionality, but bad in the sense that a massive amount of hardware would be needed to support reasonable polling intervals for non-trivial numbers of clients.</li>
    </ul>
  </section1>
</xep>