The WebSocket Protocol as a Transport for the Binary Floor Control Protocol (BFCP)

The WebSocket protocol enables two-way message exchange between clients and servers on top of a persistent TCP connection, optionally secured with Transport Layer Security (TLS) . The initial protocol handshake makes use of Hypertext Transfer Protocol (HTTP) semantics, allowing the WebSocket protocol to reuse existing HTTP infrastructure. The Binary Floor Control Protocol (BFCP) is a protocol to coordinate access to shared resources in a conference. It is defined in and is used between floor participants and floor control servers, and between floor chairs (i.e., moderators) and floor control servers. Modern web browsers include a WebSocket client stack complying with the WebSocket API as specified by the W3C. It is expected that other client applications (those running in personal computers and devices such as smartphones) will also make a WebSocket client stack available. This document extends the applicability of and to enable the usage of BFCP in these scenarios. The transport over which BFCP entities exchange messages depends on how the clients obtain information to contact the floor control server (e.g. using an Session Description Protocol (SDP) offer/answer exchange per or the procedure described in RFC5018 ). defines two transports for BFCP: TCP and UDP. This specification defines a new WebSocket sub-protocol (as defined in Section 1.9 in ) for transporting BFCP messages between a WebSocket client and server. This sub-protocol provides a reliable and boundary preserving transport for BFCP when run on top of TCP. Since WebSocket provides a reliable transport, the extensions defined in for sending BFCP over unreliable transports are not applicable.

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 .

Any BFCP entity capable of opening outbound connections to WebSocket servers and communicating using the WebSocket BFCP sub-protocol as defined by this document. Any BFCP entity capable of listening for inbound connections from WebSocket clients and communicating using the WebSocket BFCP sub-protocol as defined by this document.

The WebSocket protocol is a transport layer on top of TCP (optionally secured with TLS ) in which both client and server exchange message units in both directions. The protocol defines a connection handshake, WebSocket sub-protocol and extensions negotiation, a frame format for sending application and control data, a masking mechanism, and status codes for indicating disconnection causes. The WebSocket connection handshake is based on HTTP and utilizes the HTTP GET method with an "Upgrade" request. This is sent by the client and then answered by the server (if the negotiation succeeded) with an HTTP 101 status code. Once the handshake is completed the connection upgrades from HTTP to the WebSocket protocol. This handshake procedure is designed to reuse the existing HTTP infrastructure. During the connection handshake, client and server agree on the application protocol to use on top of the WebSocket transport. Such an application protocol (also known as a "WebSocket sub-protocol") defines the format and semantics of the messages exchanged by the endpoints. This could be a custom protocol or a standardized one (as the WebSocket BFCP sub-protocol defined in this document). Once the HTTP 101 response is processed both client and server reuse the underlying TCP connection for sending WebSocket messages and control frames to each other. Unlike plain HTTP, this connection is persistent and can be used for multiple message exchanges. The WebSocket protocol defines message units to be used by applications for the exchange of data, so it provides a message boundary-preserving transport layer. These message units can contain either UTF-8 text or binary data, and can be split into multiple WebSocket text/binary transport frames as needed by the WebSocket stack. The WebSocket API for web browsers only defines callbacks to be invoked upon receipt of an entire message unit, regardless of whether it was received in a single WebSocket frame or split across multiple frames.

The term WebSocket sub-protocol refers to an application-level protocol layered on top of a WebSocket connection. This document specifies the WebSocket BFCP sub-protocol for carrying BFCP messages over a WebSocket connection.

The BFCP WebSocket Client and BFCP WebSocket Server negotiate usage of the WebSocket BFCP sub-protocol during the WebSocket handshake procedure as defined in Section 1.3 of . The Client MUST include the value "bfcp" in the Sec-WebSocket-Protocol header in its handshake request. The 101 reply from the Server MUST contain "bfcp" in its corresponding Sec-WebSocket-Protocol header. Below is an example of a WebSocket handshake in which the Client requests the WebSocket BFCP sub-protocol support from the Server:

The handshake response from the Server accepting the WebSocket BFCP sub-protocol would look as follows:

Once the negotiation has been completed, the WebSocket connection is established and can be used for the transport of BFCP messages. The WebSocket messages transmitted over this connection MUST conform to the negotiated WebSocket sub-protocol.

BFCP messages use a TLV (Type-Length-Value) binary encoding, therefore BFCP WebSocket Clients and BFCP WebSocket Servers MUST be transported in unfragmented binary WebSocket frames (FIN:1,opcode:%x2) to exchange BFCP messages. The WebSocket frame data MUST be a valid BCFP message, so the length of the payload of the WebSocket frame MUST be lower than the maximum size allowed (2^16 +12 bytes) for a BCFP message as described in . In addition, the encoding rules for reliable protocols defined in MUST be followed. While this specification assumes that BFCP encoding is only TLV binary, future documents may define other mechanisms like JSON serialization.

WebSocket provides a reliable transport and therefore the BFCP WebSocket sub-protocol defined by this document also provides reliable BFCP transport. Thus, client and server transactions using WebSocket for transport MUST follow the procedures for reliable transports as defined in and . BFCP WebSocket clients cannot receive incoming WebSocket connections initiated by any other peer. This means that a BFCP WebSocket client MUST actively initiate a connection towards a BFCP WebSocket server. Each BFCP message MUST be carried within a single WebSocket message, and a WebSocket message MUST NOT contain more than one BFCP message.

Rules to generate an 'm' line for a BFCP stream are described in , Section 3 New values are defined for the transport field: TCP/WS/BFCP and TCP/WSS/BFCP. TCP/WS/BFCP is used when BFCP runs on top of WS, which in turn runs on top of TCP. TCP/WSS/BFCP is used when BFCP runs on top of WSS, which in turn runs on top of TLS and TCP. When TCP is used as the transport, the port field is set following the rules in Section 3 and Section 8.1 of . Depending on the value of the SDP 'setup' attribute defined in , the port field contains the port to which the remote endpoint will direct BFCP messages or is irrelevant (i.e., the endpoint will initiate the connection towards the remote endpoint) and should be set to a value of 9, which is the discard port. Connection attribute and port MUST follow the rules of Some web browsers do not allow non-secure WebSocket connections to be made. So, while the recommendation to use Secure WebSockets (i.e. TCP/WSS) is for security reasons, it is also to achieve maximum compatibility among clients.

defines a new SDP attribute to indicate the connection Uniform Resource Identifier (URI) for the WebSocket Client. The SDP attribute 'ws-uri' defined in Section 3.1 of MUST be used when BFCP runs on top of WS, which in turn runs on top of TCP. The SDP attribute 'wss-uri' defined in Section 3.2 of MUST be used when BFCP runs on top of WSS, which in turn runs on top of TLS and TCP. When the 'ws-uri' or 'wss-uri' attribute is present in the media section of the SDP, the IP and port information provided in the 'c' lines SHALL be ignored and the full URI SHALL be used instead to open the WebSocket connection. The port provided in the 'm' line SHALL be ignored too, as the a=ws-uri or a=wss-uri SHALL provide port number when needed.

An endpoint (i.e., both the offerer and the answerer) MUST create an SDP media description ("m=" line) for each BFCP-over-WebSocket media stream and MUST assign either TCP/WSS/BFCP or TCP/WS/BFCP value to the "proto" field of the "m=" line depending on whether the endpoint wishes to use secure WebSocket or WebSocket. Furthermore, the server side, which could be either the offerer or answerer, MUST add an "a=ws-uri" or "a=wss-uri" attribute in the media section depending on whether it wishes to use WebSocket or secure WebSocket. This new attribute MUST follow the syntax defined in . Additionally, the SDP Offer/Answer procedures defined in Section 4 of MUST be followed for the "m=" line associated with a BFCP-over-WebSocket media stream.

The following is an example of an "m=" line for a BFCP connection. In this example, the offerer sends the SDP with the "proto" field having a value of TCP/WSS/BFCP * indicating that the offerer wishes to use secure WebSocket as a transport for the media stream.

It is possible that an endpoint (e.g., a browser) sends an offerless INVITE to the server. In such cases the server will act as SDP offerer. The server MUST assign the SDP "setup" attribute with a value of "passive". The server MUST have an "a=ws-uri" or "a=wss-uri" attribute in the media section depending on whether the server wishes to use WebSocket or secure WebSocket. This attribute MUST follow the syntax defined in Section 3. For BFCP application, the "proto" value in the "m=" line MUST be TCP/WSS/BFCP if WebSocket is over TLS, else it MUST be TCP/WS/BFCP.

Section 9 of states that BFCP clients and floor control servers SHOULD authenticate each other prior to accepting messages, and RECOMMENDS that mutual TLS/DTLS authentication be used. However, browser-based WebSocket clients have no control over the use of TLS in the WebSocket API , so it is RECOMMENDED that standard Web-based methods for client and server authentication are used, as follows. When a BFCP WebSocket client connects to a BFCP WebSocket server, it SHOULD use TCP/WSS as its transport. The WebSocket client SHOULD inspect the TLS certificate offered by the server and verify that it is valid. Since the WebSocket API does not distinguish between certificate errors and other kinds of failure to establish a connection, it is expected that browser vendors will warn end users directly of any kind of problem with the server certificate. A floor control server that receives a message over TCP/WS can request the use of TCP/WSS by generating an Error message, as described in Section 13.8 of , with an Error code with a value of 9 (use TLS). Prior to sending BFCP requests, a BFCP WebSocket client connects to a BFCP WebSocket server and performs the connection handshake. As described in the handshake procedure involves a HTTP GET method request from the client and a response from the server including an HTTP 101 status code. In order to authorize the WebSocket connection, the BFCP WebSocket server MAY inspect any cookie headers present in the HTTP GET request. For many web applications the value of such a cookie is provided by the web server once the user has authenticated themselves to the web server, which could be done by many existing mechanisms. As an alternative method, the BFCP WebSocket Server could request HTTP authentication by replying to the Client's GET method request with a HTTP 401 status code. The WebSocket protocol covers this usage in Section 4.1: If the status code received from the server is not 101, the WebSocket client stack handles the response per HTTP procedures, in particular the client might perform authentication if it receives 401 status code.

Considerations from , and RFC5018 apply. BFCP relies on lower-layer security mechanisms to provide replay and integrity protection and confidentiality. It is RECOMMENDED that the BFCP traffic transported over a WebSocket communication be protected by using a secure WebSocket connection (using TLS over TCP).

This specification requests IANA to register the WebSocket BFCP sub-protocol under the "WebSocket Subprotocol Name" Registry with the following data: bfcp WebSocket Transport for BFCP (Binary Floor Control Protocol) RFC&rfc.number; [[NOTE TO RFC EDITOR: Please change &rfc.number; to the number assigned to this specification, and remove this paragraph on publication.]]

This document defines two new values for the SDP 'proto' field under the Session Description Protocol (SDP) Parameters registry. The resulting entries are shown in below:

[[NOTE TO RFC EDITOR: Please change &rfc.number; to the number assigned to this specification, and remove this paragraph on publication.]]

The authors want to thank Robert Welbourn, from Acme Packet, who made significant contributions to the first version of this document. This work benefited from the thorough review and constructive comments of Charles Eckel, Christer Holmberg and Paul Kyzivat.