<abstract>This specification defines a Jingle transport method that results in sending media data using raw datagram associations via the User Datagram Protocol (UDP). This transport method is negotiated via the Interactive Connectivity Establishment (ICE) methodology, which provides robust NAT traversal for media traffic.</abstract>
<remark><p>Specified id attribute and added it to the examples; updated namespaces to reflect changes to other Jingle specifications; completed editorial review.</p></remark>
<remark><p>For consistency with XEP-0166, removed profile attribute, changed content-replace to transport-replace, and changed content-accept to transport-accept.</p></remark>
<remark><p>Allowed batching of multiple candidates in a single transport-info action for optional interworking with the SDP offer-answer model, and added urn:ietf:rfc:3264 service discovery feature to advertise such support; updated security considerations regarding sharing of IP addresses.</p></remark>
<remark><p>Removed content-replace action from acceptance flow, since in ICE that information is sent via STUN, not in the signalling channel.</p></remark>
<remark><p>Moved pwd and ufrag attributes from candidate element to transport element since they describe session-level or media-level information.</p></remark>
<remark><p>Modified flow for ICE completion to require content-modify from initiator to responder, thus mapping to sending of revised offer in SIP; added rem-addr and rem-port attributes to map to a=remote-candidates information in SDP; changed raddr and rport attributes to rel-addr and rel-port to prevent confusion with rem-addr and rem-port attributes.</p></remark>
<remark><p>Separately defined ice-tcp and ice-udp transport methods to enable clearer definition of transport methods and reuse by application types; specified Jingle conformance, including definition of ice-udp as datagram and ice-tcp as streaming.</p></remark>
<remark><p>Updated to track ICE-12; corrected service discovery process; completed editorial review; removed mention of DTMF, which is for audio only.</p></remark>
<p>&xep0166; defines a framework for negotiating and managing out-of-band data sessions over XMPP. In order to provide a flexible framework, the base Jingle specification defines neither data transport methods nor application formats, leaving that up to separate specifications.</p>
<p>The current document defines a transport method for establishing and managing data exchanges between XMPP entities over the User Datagram Protocol (see &rfc0768;), using the ICE methodology developed within the IETF and specified in &ice; (hereafter referred to as &icecore;). Use of the <strong>ice-udp</strong> method results in a datagram transport suitable for media applications where some packet loss is tolerable (e.g., audio and video).</p>
<p>Note: &icecore; has been approved for publication as an RFC but has not yet been published as an RFC. While every effort has been made to keep this document synchronized with &icecore;, the interested reader is referred to &icecore; for a detailed description of the ICE methodology.</p>
<li>In Jingle, lists of "preferred" candidates are typically sent in the Jingle session-initiate and session-accept messages, in a way that is consistent with the SDP offer / answer model described in &rfc3264; and the process described in &icecore;. However, it is also possible to send candidates in separate transport-info messages; this enables a part to send higher-priority candidates earlier in the negotiation and lower-priority candidates later in the negotiation, or to continue sending candidates after session setup to adjust to changing network conditions.</li>
<li>Syntax from the Session Description Protocol (see &rfc4566;) is mapped to an XML syntax suitable for sending over the XMPP signalling channel.</li>
<li>Make it possible to establish and manage out-of-band connections between two XMPP entities, even if they are behind Network Address Translators (NATs) or firewalls.</li>
<li>Make it relatively easy to implement support in standard Jabber/XMPP clients.</li>
<li>Where communication with non-XMPP entities is needed, push as much complexity as possible onto server-side gateways between the XMPP network and the non-XMPP network.</li>
<p>In accordance with Section 10 of <cite>XEP-0166</cite>, this document specifies the following information related to the Jingle ice-udp transport method:</p>
<li><p>The semantics of the &TRANSPORT; element are defined in the <linkurl='#protocol-negotiate'>ICE Negotiation</link> section of this document.</p></li>
<li><p>Successful negotiation of the ice-udp method results in use of a datagram transport that is suitable for applications where some packet loss is tolerable, such as audio and video.</p></li>
<li><p>If multiple components are to be communicated by the application type that uses the transport, the transport shall support those components and assign identifiers for them as described in the specification that defines the application type.</p></li>
<p>The overall protocol flow for negotiation of the Jingle ICE-UDP Transport Method is as follows (note: many of these events happen simultaneously, not in sequence).</p>
<p>Note: The examples in this document follow the scenario described in Section 17 of &icecore;, except that we substitute the Shakespearean characters "Romeo" and "Juliet" for the generic entities "L" and "R".</p>
<p>In order for the initiator in a Jingle exchange to start the negotiation, it sends a Jingle "session-initiate" stanza that includes at least one content type, as described in <cite>XEP-0166</cite>. If the initiator wishes to negotiate the ice-udp transport method for an application format, it MUST include a &TRANSPORT; child element qualified by the 'urn:xmpp:jingle:transports:ice-udp:1' namespace &VNOTE;. This element SHOULD in turn contain one &CANDIDATE; element for each of the higher-priority transport candidates as determined in accordance with the ICE methodology, but MAY instead be empty (with each candidate to be sent as the payload of a transport-info message).</p>
<p>The &TRANSPORT; element's 'pwd' and 'ufrag' attributes MUST be included in the session-initiate request, in subsequent content-add and transport-replace actions, and when offering candidates via the transport-info action. The attributes MAY be included in a session-accept action. The values are separately generated for both the initiator and the responder, in accordance with &icecore; and as shown in the examples. The attributes are defined as follows.</p>
<p>The attributes of the <candidate/> element are described in the following table:</p>
<tablecaption='Candidate Attributes'>
<tr>
<th>Name</th>
<th>Description</th>
<th>SDP Syntax</th>
<th>Example</th>
</tr>
<tr>
<td>component</td>
<td>A Component ID as defined in &icecore;.</td>
<td>Component ID value in a=candidate line</td>
<td>1</td>
</tr>
<tr>
<td>foundation</td>
<td>A Foundation as defined in &icecore;.</td>
<td>Foundation value in a=candidate line</td>
<td>1</td>
</tr>
<tr>
<td>generation</td>
<td>An index, starting at 0, that enables the parties to keep track of updates to the candidate throughout the life of the session.</td>
<td>N/A</td>
<td>0</td>
</tr>
<tr>
<td>id</td>
<td>A unique identifier for the candidate.</td>
<td>N/A</td>
<td>el0747fg11</td>
</tr>
<tr>
<td>ip</td>
<td>The Internet Protocol (IP) address for the candidate transport mechanism; this may be either an IPv4 address or an IPv6 address.</td>
<td>IP Address value in a=candidate line</td>
<td>192.0.2.3</td>
</tr>
<tr>
<td>network</td>
<td>An index, starting at 0, referencing which network this candidate is on for a given peer (used for diagnostic purposes if the calling hardware has more than one Network Interface Card).</td>
<td>N/A</td>
<td>0</td>
</tr>
<tr>
<td>port</td>
<td>The port at the candidate IP address.</td>
<td>Port value in a=candidate line</td>
<td>45664</td>
</tr>
<tr>
<td>priority</td>
<td>A Priority as defined in &icecore;
<note>In accordance with the rules specified in Section 4.1.1 of &icecore;, the priority values shown in the examples within this document have been calculated as follows. The "type preference" for host candidates is stipulated to be "126" and for server reflexive candidates "100". The "local preference" for network 0 is stipulated to be "4096", for network 1 "2048", and for network 2 "1024".</note>
</td>
<td>Priority value in a=candidate line</td>
<td>2130706431</td>
</tr>
<tr>
<td>protocol</td>
<td>The protocol to be used. The only value defined by this specification is "udp".</td>
<td>Transport protocol field in a=candidate line</td>
<td>udp</td>
</tr>
<tr>
<td>rel-addr</td>
<td>A related address as defined in &icecore;.</td>
<td>raddr value in a=candidate line</td>
<td>10.0.1.1</td>
</tr>
<tr>
<td>rel-port</td>
<td>A related port as defined in &icecore;.</td>
<td>rport value in a=candidate line</td>
<td>8998</td>
</tr>
<tr>
<td>rem-addr</td>
<td>A IP address for a remote address as defined in &icecore;.</td>
<td>connection-address value in a=remote-candidates line</td>
<td>192.0.2.1</td>
</tr>
<tr>
<td>rem-port</td>
<td>The port for a remote address as defined in &icecore;.</td>
<td>port value in a=remote-candidates line</td>
<td>3478</td>
</tr>
<tr>
<td>type</td>
<td>A Candidate Type as defined in &icecore;. The allowable values are "host" for host candidates, "prflx" for peer reflexive candidates, "relay" for relayed candidates, and "srflx" for server reflexive candidates.</td>
<p>As described in <cite>XEP-0166</cite>, to acknowledge receipt of the session initiation request, the responder immediately returns an IQ-result.</p>
<p>Depending on the application type, a user agent controlled by a human user might need to wait for the user to affirm a desire to proceed with the session before continuing. When the user agent has received such affirmation (or if the user agent can automatically proceed for any reason, e.g. because no human intervention is expected or because a human user has configured the user agent to automatically accept sessions with a given entity), it returns a Jingle session-accept message. This message SHOULD also contain a &TRANSPORT; element that in turn contain one &CANDIDATE; element for each of the responder's higher-priority transport candidates, just as for the session-initiate message, but MAY instead be empty (with each candidate to be sent as the payload of a transport-info message).</p>
<p>Note: See the <linkurl='#security'>Security Considerations</link> section of this document regarding the exposure of IP addresses on behalf by the responder's client.</p>
<examplecaption="Responder accepts the session request"><![CDATA[
<p>The initiator and responder negotiate connectivity over ICE by exchanging XML-formatted transport candidates for the channel. This negotiation proceeds immediately in order to maximize the possibility that connectivity can be established (and therefore media can be exchanged) as quickly as possible. In order to expedite session establishment, the initiator SHOULD include transport candidates in its session-initiate message but MAY also send additional transport candidates as soon as it learns of them, even before receiving acknowledgement of the session-initiate message (i.e., the initiator MUST consider the session to be live as soon as it sends the session-initiate message). <note>Given in-order delivery as mandated by &xmppcore;, the responder will receive such transport-info messages after receiving the session-initiate message; if not, it is appropriate for the responder to return <unknown-session/> errors since according to its state machine the session does not exist.</note></p>
<p>The first step in negotiating connectivity is for each party to send transport candidates to the other party. <note>The fact that both parties send candidates means that Jingle requires each party to be a full implementation of ICE, not a lite implementation as specified in &icecore;.</note> These candidates SHOULD be gathered by following the procedure specified in Section 4.1.1 of &icecore; (typically by communicating with a standalone STUN server in order to discover the client's public IP address and port) and prioritized by following the procedure specified in Section 4.1.2 of &icecore;.</p>
<p>Each candidate or set of candidates shall be sent as <candidate/> children of a &TRANSPORT; element qualified by the 'urn:xmpp:jingle:transports:ice-udp:1' namespace. The &TRANSPORT; element is sent via a Jingle action of session-initiate, session-accept, or transport-info.</p>
<p>Either party MAY include multiple <candidate/> elements in one &TRANSPORT; element, especially in the session-initiate and session-accept messages sent at the beginning of the session negotiation. Including multiple candidates in the session-initiate and session-accept messages can help to ensure interoperability with entities that implement the SDP offer/answer model described in <cite>RFC 3264</cite>; in particular, an entity SHOULD include multiple candidates in its session-initiate or session-accept message if the other party advertises support for the "urn:ietf:rfc:3264" service discovery feature as described in the <linkurl='#support-sdp'>SDP Offer / Answer Support</link> section of this document. However, including one candidate per subsequent transport-info action typically results in a faster negotiation because the candidates most likely to succeed are sent first (in the session-info and session-accept messages) and it is not necessary to gather all candidates before beginning to send any candidates; furthermore, because certain candidates can be more "expensive" in terms of bandwidth or processing power, either party might not want to advertise the existence of such candidates unless it is necessary to do so after other candidates have failed.</p>
<p>If the party that receives a candidate in a Jingle message can successfully process a given candidate or set of candidates, it returns an IQ-result (if not, for example because the candidate data is improperly formatted, it returns an IQ-error). At this point, the receiving entity is only indicating receipt of the candidate or set of candidates, not telling the other party that the candidate will be used.</p>
<p>The initiator can keep sending candidates (without stopping to receive an acknowledgement of receipt from the responder for each candidate) until it has exhausted its supply of possible or desirable transport candidates; for each candidate or set of candidates, the responder acknowledges receipt. The responder can also keep sending potential candidates, which the initiator will acknowledge.</p>
<p>As the initiator and responder receive candidates, they probe the various transport candidates for connectivity. In performing these connectivity checks, each party SHOULD follow the procedure specified in Section 7 of &icecore;. The following business rules apply:</p>
<li>In accordance with &icecore;, the STUN Binding Requests MUST include the PRIORITY attribute (computed according to Section 7.1.1.1. of &icecore;).</li>
<li>For the purposes of the Jingle ICE-UDP Transport Method, both parties are full ICE implementations and therefore the controlling role MUST be assumed by the initiator and the controlled role MUST be assumed by the responder.</li>
<li>The STUN Binding Requests generated by the initiator MAY include the USE-CANDIDATE attribute to indicate that the initiator wishes to cease checks for this component.</li>
<li>The STUN Binding Requests generated by the initiator MUST include the ICE-CONTROLLING attribute.</li>
<li>The STUN Binding Requests generated by the responder MUST include the ICE-CONTROLLED attribute.</li>
<li>The parties MUST use STUN short term credentials to authenticate requests and perform message integrity checks. As in &icecore;, the username in the STUN Binding Request is of the form "ufrag-of-sender:ufrag-of-peer" and the password is the value of the 'pwd' attribute provided by the peer. <note>Thus when Romeo sends a STUN Binding Request to Juliet the credentials will be STUN username "8hhy:9uB6" and password "YH75Fviy6338Vbrhrlp8Yh" whereas when Juliet sends a STUN Binding Request to Romeo the credentials will be STUN username "9uB6:8hhy" and password "asd88fgpdd777uzjYhagZg".</note></li>
<p>When it receives a STUN Binding Request, each party MUST return a STUN Binding Response, which indicates either an error case or the success case. As described in Section 7.1.2.2 of &icecore;, a connectivity check succeeds if <em>all</em> of the following are true:</p>
<ol>
<li>The STUN transaction generated a success response.</li>
<li>The source IP address and port of the response equals the destination IP address and port to which the Binding Request was sent.</li>
<li>The destination IP address and port of the response match the source IP address and port from which the Binding Request was sent.</li>
<p>For the candidates exchanged in the previous section, the connectivity checks would be as follows (this diagram mirrors the example in &icecore;).</p>
<p>Note: Here the initiator (controlling agent) is using "aggressive nomination" as described in Section 8.1.1.2 of &icecore; and therefore includes the USE-CANDIDATE attribute in the STUN Binding Requests it sends.</p>
<p>If, based on STUN connectivity checks, the parties determine that they will be able to exchange media between a given pair of local candidates and remote candidates (i.e., the pair is "nominated" and ICE processing is "completed"), they can then begin using that candidate pair to exchange media. There is no need for the parties to communicate the chosen candidate pair in the signalling channel.</p>
<p>In the unlikely event that one of the parties determines that it cannot establish connectivity even after sending and checking lower-priority candidates, it SHOULD terminate the session as described in <cite>XEP-0166</cite>.</p>
<p>The creator of a content type MAY modify an existing, in-use candidate at any time during the session, for example to change the IP address or port. This is done by sending a transport-replace action with the changed candidate information, where the value of the 'generation' attribute is incremented to specify that the candidate information is a modification to an existing candidate.</p>
<p>If the transport-replace is acceptable, the recipient then sends a transport-accept action (if not, the recipient sends a transport-reject action).</p>
<p>Even after media has begun to flow, either party MAY continue to send additional candidates to the other party (e.g., because the user agent has become aware of a new media proxy or network interface card). As above, such candidates are shared by sending a transport-info action.</p>
<examplecaption="Initiator sends a subsequent candidate"><![CDATA[
<p>The parties SHOULD check the newly-offered candidate for connectivity, as described previously. If the parties determine that media can flow over the candidate, the initiating party MAY send a transport-replace action to the responder in order to use the new candidate.</p>
<section1topic='Fallback to Raw UDP'anchor='fallback'>
<p>It can happen that the responder does not support ICE, in which case it can be necessary to fall back to use of the &xep0177;. One typical scenario is communication between an ICE-aware Jingle endpoint and a non-ICE-aware SIP endpoint through a Jingle-to-SIP gateway, as follows:</p>
<ol>
<li>The Jingle endpoint sends a session-initiate request to the SIP endpoint, specifying a transport method of ICE-UDP.</li>
<li>Based on capabilities information, the gateway knows that the SIP endpoint does not support ICE, so it enables the endpoints to use its media relay. It does this by:
<ul>
<li>Sending a content-add request to the Jingle endpoint on behalf of the SIP endpoint, specifying a transport method of Raw UDP and a candidate whose IP address and port are hosted at the gateway.</li>
<li>Sending a content-remove request to the Jingle endpoint on behalf of the SIP endpoint, specifying a transport method of ICE-UDP.</li>
<li>Sending SIP INVITE to the SIP endpoint on behalf of the Jingle endpoint, speciying an IP address and port at the gateway.</li>
<p>Immediately the gateway sends a transport-replace action to Romeo, specifying a transport of Raw UDP with a candidate whose IP address and port identify a media relay at the gateway.</p>
<examplecaption="Gateway sends transport-replace on behalf of responder"><![CDATA[
<p>If an entity supports the Jingle ice-udp transport, it MUST return a feature of "urn:xmpp:jingle:transports:ice-udp:1" &VNOTE; in response to &xep0030; information requests.</p>
<p>In order for an application to determine whether an entity supports this protocol, where possible it SHOULD use the dynamic, presence-based profile of service discovery defined in &xep0115;. However, if an application has not received entity capabilities information from an entity, it SHOULD use explicit service discovery instead.</p>
<p>If an entity supports the SDP offer / answer model described in <cite>RFC 3264</cite> and therefore prefers to receive multiple candidates in a single "transport-info" action, it MUST advertise support for the "urn:ietf:rfc:3264" service discovery feature. Typically this feature will be advertised only by gateways between Jingle and SIP.</p>
<examplecaption="Service discovery information request"><![CDATA[
<p>In order to speed the negotiation process so that media can flow as quickly as possible, the initiator SHOULD gather and prioritize candidates in advance, or as soon as the principal begins the process of initiating a session.</p>
<p>This specification applies exclusively to Jabber/XMPP clients and places no additional requirements on Jabber/XMPP servers. However, service administrators might wish to deploy a STUN server in order to ease the client-to-client negotiation process, and a TURN server for media relaying (see &turn;).</p>
<p>By definition, the exchange of transport candidates results in exposure of the sender's IP addresses, which comprise a form of personally identifying information. A Jingle client MUST enable a user to control which entities will be allowed to receive such information. If a human user explicitly accepts a session request, then the client SHOULD consider that action to imply approval of IP address sharing. However, waiting for a human user to explicitly accept the session request can result in delays during session setup, since it is more efficient to immediately begin sharing transport candidates. Therefore, it is RECOMMENDED for the client to immediately send transport candidates to a contact (without waiting for explicit user approval of the session request) in the following cases:</p>
<li>The user has permanently and formally authorized the contact to view the user's presence information via a presence subscription as reflected in an XMPP roster item (see &xmppim;).</li>
<li>The user has temporarily and dynamically shared presence with the contact via "directed presence" as described in <cite>RFC 3921</cite>.</li>
<li>The user has explicitly added the contact to a "whitelist" of entities who are allowed to access the user's personally-identifying information.</li>
<section2topic='Encryption of Media'anchor='security-media'>
<p>In order to secure the data stream that is negotiated via the Jingle ICE transport, implementations SHOULD use encryption methods appropriate to the transport method and media being exchanged (for details regarding RTP exchanges, refer to &xep0167;).</p>
<p>Upon advancement of this specification from a status of Experimental to a status of Draft, the ®ISTRAR; shall add the foregoing namespaces to the registry located at &NAMESPACES;, as described in Section 4 of &xep0053;.</p>
<p>If an entity supports the SDP offer / answer model described in <cite>RFC 3264</cite> and therefore prefers to receive one "transport-info" action with multiple candidates, it MUST advertise support for the "urn:ietf:rfc:3264" feature.</p>