An XMPP Sub-protocol for WebSocket

Applications using XMPP (see and ) on the Web currently make use of BOSH (see and ), an XMPP binding to HTTP. BOSH is based on the HTTP long polling technique, and it suffers from high transport overhead compared to XMPP's native binding to TCP. It would be much better in most circumstances to avoid tunneling XMPP over HTTP long polled connections and instead use the XMPP protocol directly. However, the APIs and sandbox that browsers have provided do not allow this. The WebSocket protocol now exists to solve these kinds of problems. The WebSocket protocol is a bi-directional protocol that provides a simple message-based framing layer over raw sockets and allows for more robust and efficient communication in web applications. This document defines a binding of the XMPP protocol over the WebSocket protocol transport. It makes using XMPP within web applications simpler and more efficient.

The xmpp sub-protocol is used to transport XMPP over a WebSocket connection. The client and server agree to this protocol during the WebSocket handshake (see Section 1.3 of ). During the WebSocket handshake, the client MUST include the |Sec-WebSocket-Protocol| header in its handshake, and the value |xmpp| must be included in the list of protocols. The reply from the server MUST also contain |xmpp| in its own |Sec-WebSocket-Protocol| header in order for an XMPP sub-protocol connection to be established. Once the handshake is complete, WebSocket messages sent or received will conform to the protocol defined in the rest of this document.

Data frame messages in the XMPP sub-protocol MUST be of the text type and contain UTF-8 encoded data. The close control frame's contents are specified in . Control frames other than close are not restricted. Unless noted in text, the word "message" will mean a WebSocket message containing a text data frame.

The first message sent after the handshake is complete MUST be an XMPP opening stream tag as defined in XMPP or an XML text declaration (see Section 4.3.1 of ) followed by an XMPP opening stream tag. The stream tag MUST NOT be closed (i.e. the closing </stream:stream> tag should not appear in the message) as it is the start of the client's outgoing XML. The '<' character of the tag or text declaration MUST be the first character of the text payload. The server MUST respond with a message containing an error (see ), its own opening stream tag, or an XML text declaration followed by an opening stream tag. Except in the case of certain stream errors (see ), the opening stream tag, <stream:stream>, MUST appear in a message by itself.

Stream level errors in XMPP are terminal. Should such an error occur, the server MUST send the stream error as a complete element in a message to the client. If the error occurs during the opening of a stream, the stream error message MUST start with an opening stream tag (see Section 4.7.1 of ) and end with a closing stream tag. After the stream error and closing stream tag have been sent, the server MUST close the connection as in .

Either the server or the client may close the connection at any time. Before closing the connection, the closing party MUST close the XMPP stream if it has been established. To initiate the close, the closing party MUST send a normal WebSocket close message with an empty body. The connection is considered closed when a matching close message is received (see Section 1.4 of ). Except in the case of certain stream errors (see ), the closing stream tag, </stream:stream>, MUST appear in a message by itself.

Each XMPP stanza MUST be sent in its own message. A stanza MUST NOT be split over multiple messages. All first level children of the <stream:stream> element MUST be treated the same as stanzas (e.g. <stream:features> and <stream:error>).

After successful SASL authentication, an XMPP stream must be restarted. In these cases, as soon as the message is sent (or received) containing the success indication, both the server and client streams are implicitly closed, and new streams must be opened. The client MUST open a new stream as in and MUST NOT send a closing stream tag.

XMPP servers send whitespace pings as keepalives between stanzas, and XMPP clients can do the same thing. These extra whitespace characters are not significant in the protocol. Servers and clients SHOULD use WebSocket ping messages instead for this purpose. The XMPP Ping extension allows entities to send and respond to ping requests. A client sending a WebSocket ping is equivalent to pinging the WebSocket server, which may also be the XMPP server. When the XMPP server is not also the WebSocket server, a WebSocket ping may be useful to check the health of the intermediary server.

TLS cannot be used in The XMPP sub-protocol because the sub-protocol does not allow for raw binary data to be sent. A server MUST NOT advertise TLS as a stream feature (see Section 4.6 of ). A client MUST ignore the TLS feature if it is advertised over WebSocket.