The standard client-server architecture for XMPP communication provides a stable infrastructure for real-time communication. However, there are certain situations in which it is desirable to bypass the standard client-server architecture, including:

• Two endpoints cannot access an XMPP server
• Two endpoints want to enforce end-to-end encryption
• Two endpoints want to send a high volume of XMPP traffic but the intermediate servers enforce rate limits

The first situation is addressed by &xep0174;. However, if the endpoints already have client-to-server connections but wish to bypass those connections or leverage those streams for a higher-level application such as end-to-end encryption, it is desirable for the two endpoints to negotiate an end-to-end XML stream. This specification defines methods for doing so, where the application format is an XML stream and the transport method is any direct or mediated streaming transport, such as &xep0261; (mediated), &xep0260; (direct or mediated), or a future ice-tcp Jingle transport (direct or mediated) based on &rfc6544;.

This section provides a friendly introduction to Jingle XML streams.

First, the party that wishes to initiate the stream determines the responder's capabilities (via &xep0030; or &xep0115;). Here we assume that the responder supports a service discovery feature of 'urn:xmpp:jingle:apps:xmlstream:0' &VNOTE; corresponding to the Jingle XML stream functionality defined herein, as well as the 'urn:xmpp:jingle:transports:ibb:0' feature.

The initiator then sends a Jingle session-initiation request to the responder. The content-type of the request specifies three things:

1. An application type of "urn:xmpp:jingle:apps:xmlstream:0".

2. Options for the streaming transport method, such as In-Band Bytestreams ("IBB") as defined in XEP-0261 or SOCKS5 Bytestreams ("S5B") as defined in XEP-0260.

Note: It is STRONGLY RECOMMENDED to encrypt all end-to-end XML streams as described in Jingle-XTLS (currently located at <http://xmpp.org/extensions/inbox/jingle-xtls.html>). Those security flows are NOT described here.

In this example, the initiator is <romeo@montague.lit>, the responder is <juliet@capulet.lit>, and the initiation request specifies a transport method of "jingle-ibb" (i.e., XEP-0261).

The flow is as follows.

First the initiator sends a Jingle session-initiate.

The responder immediately acknowledges receipt of the Jingle session-initiate.

If the responding user accepts the session then her client sends a session-accept.

The initiator acknowledges receipt.

The clients can then begin to exchange XMPP data over the in-band bytestream. Because the transport is an in-band bytestream, the XMPP data is prepared as described in &xep0047; (i.e., Base64-encoded).

First the initiator sends an initial stream header to the responder.

Note: In accordance with &xmppim;, the initial stream header SHOULD include the 'to' and 'from' attributes, which SHOULD specify the full JIDs of the clients. If the initiator supports stream features and the other stream-related aspects of XMPP 1.0 as specified in RFC 3920, then it SHOULD include the version='1.0' flag as shown in the previous example.

The responder then sends a response stream header back to the initiator (because this stream header is sent in the other direction, the IBB 'seq' attribute has a value of zero, not 1).

Once the streams are established, either entity then can send XMPP message, presence, and IQ stanzas, with or without 'to' and 'from' addresses.

The responder could then send a reply.

To end the XML stream, either party sends a closing </stream:stream> element.

However, even after the application-level XML stream is terminated, the negotiated Jingle transport (here in-band bytestream) continues and could be re-used. To completely terminate the Jingle session, the terminating party would then also send a Jingle session-terminate message.

The other party then acknowledges the Jingle session-terminate.

No interaction with &IANA; is required as a result of this document.