WebRTC Gateways
Google
harald@alvestrand.no
Nokia Networks
uwe.rauschenbach@nokia.com
RTCWeb Working Group
This document specifies conformance requirements for a class of
WebRTC-compatible endpoints called "WebRTC gateways",
which interconnect between WebRTC endpoints and devices that are not
WebRTC endpoints.
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 RFC 2119.
The WebRTC model described in is focused on direct browser to
browser communication as its primary use case. Nevertheless, it is
clearly interesting to have WebRTC endpoints
connect to other types of devices, including but not limited to SIP
phones, legacy phones, CLUE-based teleconferencing systems, XMPP-based
conferencing systems, and entirely proprietary devices or systems.
WebRTC gateways are a specific type of WebRTC-compatible endpoints
which enable the
exchange of media streams between WebRTC endpoints on one side, and the
other types of devices mentioned above on the other side.
This document describes the requirements that need to be placed on
such gateways, both the requirements on WebRTC endpoints that can
be relaxed and the additional requirements that need to be applied.
A WebRTC gateway is a WebRTC-compatible endpoint, and will thus not
be conformant with all requirements for a WebRTC endpoint
(it does not do everything a WebRTC endpoint does),
but is able to interoperate with WebRTC endpoints.
A gateway will be limited in the functionality it can offer by the
system or class of devices it is gatewaying to. For instance, a
gateway into the telephone
system will not be able to relay data or video, no matter how much it
is required. Therefore, a number of functions that are mandatory to
support in WebRTC endpoints are not mandatory on gateways; the
requirement on the gateway is that it is able to negotiate those
features away correctly.
The WebRTC model is that signalling is outside the scope of the
specification. This document does not change that.
Nevertheless, any practical gateway needs to deal with signalling.
For that, this document assumes that the overall system consists of
an application running in the WebRTC browser,
possibly one or more signalling relays that mediate signalling and
thereby enable communication between the application and the gateway,
and the actual gateway that is responsible for handling the media
flows.
The application, the signalling relays (if any) and the gateway
together need to be able to:
adhere to the offer/answer semantics
deal with the description of configuration coming from the
browser; this is specified in SDP format in the WebRTC browser
API
generate the information that is needed by the browser to set
up the session, and express that information in the form of SDP.
The shorthand notation "The gateway MUST/SHOULD/MAY support <SDP
function xxx>" used below means
that an application running in the Web browser, the signalling
relays, and the gateway together MUST/SHOULD/MAY
support this functionality; it is not a requirement that this happens
at the media gateway itself.
WebRTC gateways are intended to communicate with WebRTC endpoints.
WebRTC gateways are no User Agents. They are therefore expected to
conform to the requirements for WebRTC non-browsers in [I-D.ietf-rtcweb-overview],
with the exceptions defined in this section.
Since a gateway is expected to be deployed where it can be reached
with a static IP address (as seen from the client), a WebRTC gateway
does not need to support full ICE; it therefore MAY implement ICE-Lite
only.
ICE-Lite implementations do not send consent checks, so a gateway MAY
choose not to send consent checks too, but MUST respond to consent
checks it receives.
A gateway is expected to not need to hide its location, so it does
not need to support functionality for operating only via a TURN server;
instead it MAY choose to produce Host ICE candidates only.
If a gateway serves as a media relay into another RTP domain, it MAY
choose to support only features available in that network. This means
that it MAY not (need to) support Bundle and any of the RTP/RTCP
extensions related to it, RTCP-Mux, or Trickle Ice. However, the gateway
MUST support DTLS-SRTP, since this is required for interworking with
WebRTC endpoints.
If a gateway serves as a media relay into a network or to devices not
implementing the WebRTC Datachannel, it MAY choose to not support the
Datachannel.
This document makes no request of IANA.
Note to RFC Editor: this section may be removed on publication as an
RFC.
A WebRTC gateway may operate in two security modes: Security-context
termination and security-context relaying.
Relaying is only possible where signed and encrypted content can be
passed through unchanged, and where keys can be exchanged directly
between the endpoints.
When the gateway terminates the security context, it means that the
WebRTC user has to place trust in the gateway to perform all
verification of identity and protection of content in the realm on the
other side of the gateway; there is no way the end-user can detect a
man-in-the-middle attack, an identity spoofing attack or a recording
done at the gateway. For many scenarios, this is not going to be seen as
a problem, but needs to be considered when one decides to use a
gatewayed service.
Several comments from Christer Holmberg were included.
Changes from draft-alvestrand-rtcweb-gateways-00
Aligned terminology with draft-rtcweb-overview-12
Rewrote text on signaling to improve clarity
Editorial nits
Changes from draft-alvestrand-rtcweb-gateways-01
Aligned terminology with draft-rtcweb-overview-13 ("non-browser")
Nits