Internet Engineering Task Force (IETF)                       E. Rescorla
Request for Comments: 8826                                       Mozilla
Category: Standards Track                                   January 2021
ISSN: 2070-1721


                  Security Considerations for WebRTC

Abstract

  WebRTC is a protocol suite for use with real-time applications that
  can be deployed in browsers -- "real-time communication on the Web".
  This document defines the WebRTC threat model and analyzes the
  security threats of WebRTC in that model.

Status of This Memo

  This is an Internet Standards Track document.

  This document is a product of the Internet Engineering Task Force
  (IETF).  It represents the consensus of the IETF community.  It has
  received public review and has been approved for publication by the
  Internet Engineering Steering Group (IESG).  Further information on
  Internet Standards is available in Section 2 of RFC 7841.

  Information about the current status of this document, any errata,
  and how to provide feedback on it may be obtained at
  https://www.rfc-editor.org/info/rfc8826.

Copyright Notice

  Copyright (c) 2021 IETF Trust and the persons identified as the
  document authors.  All rights reserved.

  This document is subject to BCP 78 and the IETF Trust's Legal
  Provisions Relating to IETF Documents
  (https://trustee.ietf.org/license-info) in effect on the date of
  publication of this document.  Please review these documents
  carefully, as they describe your rights and restrictions with respect
  to this document.  Code Components extracted from this document must
  include Simplified BSD License text as described in Section 4.e of
  the Trust Legal Provisions and are provided without warranty as
  described in the Simplified BSD License.

  This document may contain material from IETF Documents or IETF
  Contributions published or made publicly available before November
  10, 2008.  The person(s) controlling the copyright in some of this
  material may not have granted the IETF Trust the right to allow
  modifications of such material outside the IETF Standards Process.
  Without obtaining an adequate license from the person(s) controlling
  the copyright in such materials, this document may not be modified
  outside the IETF Standards Process, and derivative works of it may
  not be created outside the IETF Standards Process, except to format
  it for publication as an RFC or to translate it into languages other
  than English.

Table of Contents

  1.  Introduction
  2.  Terminology
  3.  The Browser Threat Model
    3.1.  Access to Local Resources
    3.2.  Same-Origin Policy
    3.3.  Bypassing SOP: CORS, WebSockets, and Consent to Communicate
  4.  Security for WebRTC Applications
    4.1.  Access to Local Devices
      4.1.1.  Threats from Screen Sharing
      4.1.2.  Calling Scenarios and User Expectations
        4.1.2.1.  Dedicated Calling Services
        4.1.2.2.  Calling the Site You're On
      4.1.3.  Origin-Based Security
      4.1.4.  Security Properties of the Calling Page
    4.2.  Communications Consent Verification
      4.2.1.  ICE
      4.2.2.  Masking
      4.2.3.  Backward Compatibility
      4.2.4.  IP Location Privacy
    4.3.  Communications Security
      4.3.1.  Protecting Against Retrospective Compromise
      4.3.2.  Protecting Against During-Call Attack
        4.3.2.1.  Key Continuity
        4.3.2.2.  Short Authentication Strings
        4.3.2.3.  Third-Party Identity
        4.3.2.4.  Page Access to Media
      4.3.3.  Malicious Peers
    4.4.  Privacy Considerations
      4.4.1.  Correlation of Anonymous Calls
      4.4.2.  Browser Fingerprinting
  5.  Security Considerations
  6.  IANA Considerations
  7.  References
    7.1.  Normative References
    7.2.  Informative References
  Acknowledgements
  Author's Address

1.  Introduction

  The Real-Time Communications on the Web (RTCWEB) Working Group has
  standardized protocols for real-time communications between Web
  browsers, generally called "WebRTC" [RFC8825].  The major use cases
  for WebRTC technology are real-time audio and/or video calls, Web
  conferencing, and direct data transfer.  Unlike most conventional
  real-time systems (e.g., SIP-based [RFC3261] soft phones), WebRTC
  communications are directly controlled by some Web server.  A simple
  case is shown below.

                            +----------------+
                            |                |
                            |   Web Server   |
                            |                |
                            +----------------+
                                ^        ^
                               /          \
                      HTTPS   /            \   HTTPS
                        or   /              \   or
                 WebSockets /                \ WebSockets
                           v                  v
                        JS API              JS API
                  +-----------+            +-----------+
                  |           |    Media   |           |
                  |  Browser  |<---------->|  Browser  |
                  |           |            |           |
                  +-----------+            +-----------+
                      Alice                     Bob

                     Figure 1: A Simple WebRTC System

  In the system shown in Figure 1, Alice and Bob both have WebRTC-
  enabled browsers and they visit some Web server which operates a
  calling service.  Each of their browsers exposes standardized
  JavaScript (JS) calling APIs (implemented as browser built-ins) which
  are used by the Web server to set up a call between Alice and Bob.
  The Web server also serves as the signaling channel to transport
  control messages between the browsers.  While this system is
  topologically similar to a conventional SIP-based system (with the
  Web server acting as the signaling service and browsers acting as
  softphones), control has moved to the central Web server; the browser
  simply provides API points that are used by the calling service.  As
  with any Web application, the Web server can move logic between the
  server and JavaScript in the browser, but regardless of where the
  code is executing, it is ultimately under control of the server.

  It should be immediately apparent that this type of system poses new
  security challenges beyond those of a conventional Voice over IP
  (VoIP) system.  In particular, it needs to contend with malicious
  calling services.  For example, if the calling service can cause the
  browser to make a call at any time to any callee of its choice, then
  this facility can be used to bug a user's computer without their
  knowledge, simply by placing a call to some recording service.  More
  subtly, if the exposed APIs allow the server to instruct the browser
  to send arbitrary content, then they can be used to bypass firewalls
  or mount denial-of-service (DoS) attacks.  Any successful system will
  need to be resistant to this and other attacks.

  A companion document [RFC8827] describes a security architecture
  intended to address the issues raised in this document.

2.  Terminology

  The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
  "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
  "OPTIONAL" in this document are to be interpreted as described in
  BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all
  capitals, as shown here.

3.  The Browser Threat Model

  The security requirements for WebRTC follow directly from the
  requirement that the browser's job is to protect the user.  Huang et
  al. [huang-w2sp] summarize the core browser security guarantee as
  follows:

     Users can safely visit arbitrary web sites and execute scripts
     provided by those sites.

  It is important to realize that this includes sites hosting arbitrary
  malicious scripts.  The motivation for this requirement is simple: it
  is trivial for attackers to divert users to sites of their choice.
  For instance, an attacker can purchase display advertisements which
  direct the user (either automatically or via user clicking) to their
  site, at which point the browser will execute the attacker's scripts.
  Thus, it is important that it be safe to view arbitrarily malicious
  pages.  Of course, browsers inevitably have bugs which cause them to
  fall short of this goal, but any new WebRTC functionality must be
  designed with the intent to meet this standard.  The remainder of
  this section provides more background on the existing Web security
  model.

  In this model, then, the browser acts as a Trusted Computing Base
  (TCB) both from the user's perspective and to some extent from the
  server's.  While HTML and JavaScript provided by the server can cause
  the browser to execute a variety of actions, those scripts operate in
  a sandbox that isolates them both from the user's computer and from
  each other, as detailed below.

  Conventionally, we refer to either Web attackers, who are able to
  induce you to visit their sites but do not control the network, or
  network attackers, who are able to control your network.  Network
  attackers correspond to the [RFC3552] "Internet Threat Model".  Note
  that in some cases, a network attacker is also a Web attacker, since
  transport protocols that do not provide integrity protection allow
  the network to inject traffic as if they were any communications
  peer.  TLS, and HTTPS in particular, prevent against these attacks,
  but when analyzing HTTP connections, we must assume that traffic is
  going to the attacker.

3.1.  Access to Local Resources

  While the browser has access to local resources such as keying
  material, files, the camera, and the microphone, it strictly limits
  or forbids Web servers from accessing those same resources.  For
  instance, while it is possible to produce an HTML form which will
  allow file upload, a script cannot do so without user consent and in
  fact cannot even suggest a specific file (e.g., /etc/passwd); the
  user must explicitly select the file and consent to its upload.
  (Note: In many cases, browsers are explicitly designed to avoid
  dialogs with the semantics of "click here to bypass security checks",
  as extensive research [cranor-wolf] shows that users are prone to
  consent under such circumstances.)

  Similarly, while Flash programs (SWFs) [SWF] can access the camera
  and microphone, they explicitly require that the user consent to that
  access.  In addition, some resources simply cannot be accessed from
  the browser at all.  For instance, there is no real way to run
  specific executables directly from a script (though the user can of
  course be induced to download executable files and run them).

3.2.  Same-Origin Policy

  Many other resources are accessible but isolated.  For instance,
  while scripts are allowed to make HTTP requests via the fetch() API
  (see [fetch]) when requests are made to a server other than from the
  same *origin* from whence the script came [RFC6454] they are not able
  to read the responses.  Cross-Origin Resource Sharing (CORS) [fetch]
  and WebSockets [RFC6455] provide an escape hatch from this
  restriction, as described below.  This same-origin policy (SOP)
  prevents server A from mounting attacks on server B via the user's
  browser, which protects both the user (e.g., from misuse of their
  credentials) and server B (e.g., from DoS attacks).

  More generally, SOP forces scripts from each site to run in their
  own, isolated, sandboxes.  While there are techniques to allow them
  to interact, those interactions generally must be mutually consensual
  (by each site) and are limited to certain channels.  For instance,
  multiple pages/browser panes from the same origin can read each
  other's JS variables, but pages from different origins -- or even
  IFRAMEs from different origins on the same page -- cannot.

3.3.  Bypassing SOP: CORS, WebSockets, and Consent to Communicate

  While SOP serves an important security function, it also makes it
  inconvenient to write certain classes of applications.  In
  particular, mash-ups, in which a script from origin A uses resources
  from origin B, can only be achieved via a certain amount of hackery.
  The W3C CORS spec [fetch] is a response to this demand.  In CORS,
  when a script from origin A executes a potentially unsafe cross-
  origin request, the browser instead contacts the target server to
  determine whether it is willing to allow cross-origin requests from
  A.  If it is so willing, the browser then allows the request.  This
  consent verification process is designed to safely allow cross-origin
  requests.

  While CORS is designed to allow cross-origin HTTP requests,
  WebSockets [RFC6455] allows cross-origin establishment of transparent
  channels.  Once a WebSockets connection has been established from a
  script to a site, the script can exchange any traffic it likes
  without being required to frame it as a series of HTTP request/
  response transactions.  As with CORS, a WebSockets transaction starts
  with a consent verification stage to avoid allowing scripts to simply
  send arbitrary data to another origin.

  While consent verification is conceptually simple -- just do a
  handshake before you start exchanging the real data -- experience has
  shown that designing a correct consent verification system is
  difficult.  In particular, Huang et al. [huang-w2sp] have shown
  vulnerabilities in the existing Java and Flash consent verification
  techniques and in a simplified version of the WebSockets handshake.
  It is important to be wary of CROSS-PROTOCOL attacks in which the
  attacking script generates traffic which is acceptable to some non-
  Web protocol state machine.  In order to resist this form of attack,
  WebSockets incorporates a masking technique intended to randomize the
  bits on the wire, thus making it more difficult to generate traffic
  which resembles a given protocol.

4.  Security for WebRTC Applications

4.1.  Access to Local Devices

  As discussed in Section 1, allowing arbitrary sites to initiate calls
  violates the core Web security guarantee; without some access
  restrictions on local devices, any malicious site could simply bug a
  user.  At minimum, then, it MUST NOT be possible for arbitrary sites
  to initiate calls to arbitrary locations without user consent.  This
  immediately raises the question, however, of what should be the scope
  of user consent.

  In order for the user to make an intelligent decision about whether
  to allow a call (and hence their camera and microphone input to be
  routed somewhere), they must understand either who is requesting
  access, where the media is going, or both.  As detailed below, there
  are two basic conceptual models:

  1.  You are sending your media to entity A because you want to talk
      to entity A (e.g., your mother).

  2.  Entity A (e.g., a calling service) asks to access the user's
      devices with the assurance that it will transfer the media to
      entity B (e.g., your mother).

  In either case, identity is at the heart of any consent decision.
  Moreover, the identity of the party the browser is connecting to is
  all that the browser can meaningfully enforce; if you are calling A,
  A can simply forward the media to C.  Similarly, if you authorize A
  to place a call to B, A can call C instead.  In either case, all the
  browser is able to do is verify and check authorization for whoever
  is controlling where the media goes.  The target of the media can of
  course advertise a security/privacy policy, but this is not something
  that the browser can enforce.  Even so, there are a variety of
  different consent scenarios that motivate different technical consent
  mechanisms.  We discuss these mechanisms in the sections below.

  It's important to understand that consent to access local devices is
  largely orthogonal to consent to transmit various kinds of data over
  the network (see Section 4.2).  Consent for device access is largely
  a matter of protecting the user's privacy from malicious sites.  By
  contrast, consent to send network traffic is about preventing the
  user's browser from being used to attack its local network.  Thus, we
  need to ensure communications consent even if the site is not able to
  access the camera and microphone at all (hence WebSockets's consent
  mechanism) and similarly, we need to be concerned with the site
  accessing the user's camera and microphone even if the data is to be
  sent back to the site via conventional HTTP-based network mechanisms
  such as HTTP POST.

4.1.1.  Threats from Screen Sharing

  In addition to camera and microphone access, there has been demand
  for screen and/or application sharing functionality.  Unfortunately,
  the security implications of this functionality are much harder for
  users to intuitively analyze than for camera and microphone access.
  (See <https://lists.w3.org/Archives/Public/public-
  webrtc/2013Mar/0024.html> for a full analysis.)

  The most obvious threats are simply those of "oversharing".  I.e.,
  the user may believe they are sharing a window when in fact they are
  sharing an application, or may forget they are sharing their whole
  screen, icons, notifications, and all.  This is already an issue with
  existing screen sharing technologies and is made somewhat worse if a
  partially trusted site is responsible for asking for the resource to
  be shared rather than having the user propose it.

  A less obvious threat involves the impact of screen sharing on the
  Web security model.  A key part of the Same-Origin Policy is that
  HTML or JS from site A can reference content from site B and cause
  the browser to load it, but (unless explicitly permitted) cannot see
  the result.  However, if a Web application from a site is screen
  sharing the browser, then this violates that invariant, with serious
  security consequences.  For example, an attacker site might request
  screen sharing and then briefly open up a new window to the user's
  bank or webmail account, using screen sharing to read the resulting
  displayed content.  A more sophisticated attack would be to open up a
  source view window to a site and use the screen sharing result to
  view anti-cross-site request forgery tokens.

  These threats suggest that screen/application sharing might need a
  higher level of user consent than access to the camera or microphone.

4.1.2.  Calling Scenarios and User Expectations

  While a large number of possible calling scenarios are possible, the
  scenarios discussed in this section illustrate many of the
  difficulties of identifying the relevant scope of consent.

4.1.2.1.  Dedicated Calling Services

  The first scenario we consider is a dedicated calling service.  In
  this case, the user has a relationship with a calling site and
  repeatedly makes calls on it.  It is likely that rather than having
  to give permission for each call, the user will want to give the
  calling service long-term access to the camera and microphone.  This
  is a natural fit for a long-term consent mechanism (e.g., installing
  an app store "application" to indicate permission for the calling
  service).  A variant of the dedicated calling service is a gaming
  site (e.g., a poker site) which hosts a dedicated calling service to
  allow players to call each other.

  With any kind of service where the user may use the same service to
  talk to many different people, there is a question about whether the
  user can know who they are talking to.  If I grant permission to
  calling service A to make calls on my behalf, then I am implicitly
  granting it permission to bug my computer whenever it wants.  This
  suggests another consent model in which a site is authorized to make
  calls but only to certain target entities (identified via media-plane
  cryptographic mechanisms as described in Section 4.3.2 and especially
  Section 4.3.2.3).  Note that the question of consent here is related
  to but distinct from the question of peer identity: I might be
  willing to allow a calling site to in general initiate calls on my
  behalf but still have some calls via that site where I can be sure
  that the site is not listening in.

4.1.2.2.  Calling the Site You're On

  Another simple scenario is calling the site you're actually visiting.
  The paradigmatic case here is the "click here to talk to a
  representative" windows that appear on many shopping sites.  In this
  case, the user's expectation is that they are calling the site
  they're actually visiting.  However, it is unlikely that they want to
  provide a general consent to such a site; just because I want some
  information on a car doesn't mean that I want the car manufacturer to
  be able to activate my microphone whenever they please.  Thus, this
  suggests the need for a second consent mechanism where I only grant
  consent for the duration of a given call.  As described in
  Section 3.1, great care must be taken in the design of this interface
  to avoid the users just clicking through.  Note also that the user
  interface chrome, which is the representation through which the user
  interacts with the user agent itself, must clearly display elements
  showing that the call is continuing in order to avoid attacks where
  the calling site just leaves it up indefinitely but shows a Web UI
  that implies otherwise.

4.1.3.  Origin-Based Security

  Now that we have described the calling scenarios, we can start to
  reason about the security requirements.

  As discussed in Section 3.2, the basic unit of Web sandboxing is the
  origin, and so it is natural to scope consent to the origin.
  Specifically, a script from origin A MUST only be allowed to initiate
  communications (and hence to access the camera and microphone) if the
  user has specifically authorized access for that origin.  It is of
  course technically possible to have coarser-scoped permissions, but
  because the Web model is scoped to the origin, this creates a
  difficult mismatch.

  Arguably, the origin is not fine-grained enough.  Consider the
  situation where Alice visits a site and authorizes it to make a
  single call.  If consent is expressed solely in terms of the origin,
  then on any future visit to that site (including one induced via a
  mash-up or ad network), the site can bug Alice's computer, use the
  computer to place bogus calls, etc.  While in principle Alice could
  grant and then revoke the privilege, in practice privileges
  accumulate; if we are concerned about this attack, something else is
  needed.  There are a number of potential countermeasures to this sort
  of issue.

  Individual Consent
     Ask the user for permission for each call.

  Callee-oriented Consent
     Only allow calls to a given user.

  Cryptographic Consent
     Only allow calls to a given set of peer keying material or to a
     cryptographically established identity.

  Unfortunately, none of these approaches is satisfactory for all
  cases.  As discussed above, individual consent puts the user's
  approval in the UI flow for every call.  Not only does this quickly
  become annoying but it can train the user to simply click "OK", at
  which point the consent becomes useless.  Thus, while it may be
  necessary to have individual consent in some cases, this is not a
  suitable solution for (for instance) the calling service case.  Where
  necessary, in-flow user interfaces must be carefully designed to
  avoid the risk of the user blindly clicking through.

  The other two options are designed to restrict calls to a given
  target.  Callee-oriented consent provided by the calling site would
  not work well because a malicious site can claim that the user is
  calling any user of their choice.  One fix for this is to tie calls
  to a cryptographically established identity.  While not suitable for
  all cases, this approach may be useful for some.  If we consider the
  case of advertising, it's not particularly convenient to require the
  advertiser to instantiate an IFRAME on the hosting site just to get
  permission; a more convenient approach is to cryptographically tie
  the advertiser's certificate to the communication directly.  We're
  still tying permissions to the origin here, but to the media origin
  (and/or destination) rather than to the Web origin.  [RFC8827]
  describes mechanisms which facilitate this sort of consent.

  Another case where media-level cryptographic identity makes sense is
  when a user really does not trust the calling site.  For instance, I
  might be worried that the calling service will attempt to bug my
  computer, but I also want to be able to conveniently call my friends.
  If consent is tied to particular communications endpoints, then my
  risk is limited.  Naturally, it is somewhat challenging to design UI
  primitives which express this sort of policy.  The problem becomes
  even more challenging in multi-user calling cases.

4.1.4.  Security Properties of the Calling Page

  Origin-based security is intended to secure against Web attackers.
  However, we must also consider the case of network attackers.
  Consider the case where I have granted permission to a calling
  service by an origin that has the HTTP scheme, e.g., <http://calling-
  service.example.com>.  If I ever use my computer on an unsecured
  network (e.g., a hotspot or if my own home wireless network is
  insecure), and browse any HTTP site, then an attacker can bug my
  computer.  The attack proceeds like this:

  1.  I connect to <http://anything.example.org/>.  Note that this site
      is unaffiliated with the calling service.

  2.  The attacker modifies my HTTP connection to inject an IFRAME (or
      a redirect) to <http://calling-service.example.com>.

  3.  The attacker forges the response from <http://calling-
      service.example.com/> to inject JS to initiate a call to
      themselves.

  Note that this attack does not depend on the media being insecure.
  Because the call is to the attacker, it is also encrypted to them.
  Moreover, it need not be executed immediately; the attacker can
  "infect" the origin semi-permanently (e.g., with a Web worker or a
  popped-up window that is hidden under the main window) and thus be
  able to bug me long after I have left the infected network.  This
  risk is created by allowing calls at all from a page fetched over
  HTTP.

  Even if calls are only possible from HTTPS [RFC2818] sites, if those
  sites include active content (e.g., JavaScript) from an untrusted
  site, that JavaScript is executed in the security context of the page
  [finer-grained].  This could lead to compromise of a call even if the
  parent page is safe.  Note: This issue is not restricted to *pages*
  which contain untrusted content.  If any page from a given origin
  ever loads JavaScript from an attacker, then it is possible for that
  attacker to infect the browser's notion of that origin semi-
  permanently.

4.2.  Communications Consent Verification

  As discussed in Section 3.3, allowing Web applications unrestricted
  network access via the browser introduces the risk of using the
  browser as an attack platform against machines which would not
  otherwise be accessible to the malicious site, for instance, because
  they are topologically restricted (e.g., behind a firewall or NAT).
  In order to prevent this form of attack as well as cross-protocol
  attacks, it is important to require that the target of traffic
  explicitly consent to receiving the traffic in question.  Until that
  consent has been verified for a given endpoint, traffic other than
  the consent handshake MUST NOT be sent to that endpoint.

  Note that consent verification is not sufficient to prevent overuse
  of network resources.  Because WebRTC allows for a Web site to create
  data flows between two browser instances without user consent, it is
  possible for a malicious site to chew up a significant amount of a
  user's bandwidth without incurring significant costs to themselves by
  setting up such a channel to another user.  However, as a practical
  matter there are a large number of Web sites which can act as data
  sources, so an attacker can at least use downlink bandwidth with
  existing Web APIs.  However, this potential DoS vector reinforces the
  need for adequate congestion control for WebRTC protocols to ensure
  that they play fair with other demands on the user's bandwidth.

4.2.1.  ICE

  Verifying receiver consent requires some sort of explicit handshake,
  but conveniently we already need one in order to do NAT hole-
  punching.  Interactive Connectivity Establishment (ICE) [RFC8445]
  includes a handshake designed to verify that the receiving element
  wishes to receive traffic from the sender.  It is important to
  remember here that the site initiating ICE is presumed malicious; in
  order for the handshake to be secure, the receiving element MUST
  demonstrate receipt/knowledge of some value not available to the site
  (thus preventing the site from forging responses).  In order to
  achieve this objective with ICE, the Session Traversal Utilities for
  NAT (STUN) transaction IDs must be generated by the browser and MUST
  NOT be made available to the initiating script, even via a diagnostic
  interface.  Verifying receiver consent also requires verifying the
  receiver wants to receive traffic from a particular sender, and at
  this time; for example, a malicious site may simply attempt ICE to
  known servers that are using ICE for other sessions.  ICE provides
  this verification as well, by using the STUN credentials as a form of
  per-session shared secret.  Those credentials are known to the Web
  application, but would need to also be known and used by the STUN-
  receiving element to be useful.

  There also needs to be some mechanism for the browser to verify that
  the target of the traffic continues to wish to receive it.  Because
  ICE keepalives are indications, they will not work here.  [RFC7675]
  describes the mechanism for providing consent freshness.

4.2.2.  Masking

  Once consent is verified, there still is some concern about
  misinterpretation attacks as described by Huang et al. [huang-w2sp].
  This does not seem like it is of serious concern with DTLS because
  the ICE handshake enforces receiver consent and there is little
  evidence of passive DTLS proxies of the type studied by Huang.
  However, because RTCWEB can run over TCP there is some concern that
  attackers might control the ciphertext by controlling the plaintext
  input to SCTP.  This risk is only partially mitigated by the fact
  that the SCTP stack controls the framing of the packets.

  Note that in principle an attacker could exert some control over
  Secure Real-time Transport Protocol (SRTP) packets by using a
  combination of the WebAudio API and extremely tight timing control.
  The primary risk here seems to be carriage of SRTP over Traversal
  Using Relays around NAT (TURN) TCP.  However, as SRTP packets have an
  extremely characteristic packet header it seems unlikely that any but
  the most aggressive intermediaries would be confused into thinking
  that another application-layer protocol was in use.

4.2.3.  Backward Compatibility

     |  Note: The RTCWEB WG ultimately decided to require ICE.  This
     |  section provides context for that decision.

  A requirement to use ICE limits compatibility with legacy non-ICE
  clients.  It seems unsafe to completely remove the requirement for
  some check.  All proposed checks have the common feature that the
  browser sends some message to the candidate traffic recipient and
  refuses to send other traffic until that message has been replied to.
  The message/reply pair must be generated in such a way that an
  attacker who controls the Web application cannot forge them,
  generally by having the message contain some secret value that must
  be incorporated (e.g., echoed, hashed into, etc.).  Non-ICE
  candidates for this role (in cases where the legacy endpoint has a
  public address) include:

  *  STUN checks without using ICE (i.e., the non-RTC-web endpoint sets
     up a STUN responder).

  *  Use of the RTP Control Protocol (RTCP) as an implicit reachability
     check.

  In the RTCP approach, the WebRTC endpoint is allowed to send a
  limited number of RTP packets prior to receiving consent.  This
  allows a short window of attack.  In addition, some legacy endpoints
  do not support RTCP, so this is a much more expensive solution for
  such endpoints, for which it would likely be easier to implement ICE.
  For these two reasons, an RTCP-based approach does not seem to
  address the security issue satisfactorily.

  In the STUN approach, the WebRTC endpoint is able to verify that the
  recipient is running some kind of STUN endpoint but unless the STUN
  responder is integrated with the ICE username/password establishment
  system, the WebRTC endpoint cannot verify that the recipient consents
  to this particular call.  This may be an issue if existing STUN
  servers are operated at addresses that are not able to handle
  bandwidth-based attacks.  Thus, this approach does not seem
  satisfactory either.

  If the systems are tightly integrated (i.e., the STUN endpoint
  responds with responses authenticated with ICE credentials), then
  this issue does not exist.  However, such a design is very close to
  an ICE-Lite implementation (indeed, arguably is one).  An
  intermediate approach would be to have a STUN extension that
  indicated that one was responding to WebRTC checks but not computing
  integrity checks based on the ICE credentials.  This would allow the
  use of standalone STUN servers without the risk of confusing them
  with legacy STUN servers.  If a non-ICE legacy solution is needed,
  then this is probably the best choice.

  Once initial consent is verified, we also need to verify continuing
  consent, in order to avoid attacks where two people briefly share an
  IP (e.g., behind a NAT in an Internet cafe) and the attacker arranges
  for a large, unstoppable, traffic flow to the network and then
  leaves.  The appropriate technologies here are fairly similar to
  those for initial consent, though are perhaps weaker since the
  threats are less severe.

4.2.4.  IP Location Privacy

  Note that as soon as the callee sends their ICE candidates, the
  caller learns the callee's IP addresses.  The callee's server-
  reflexive address reveals a lot of information about the callee's
  location.  In order to avoid tracking, implementations may wish to
  suppress the start of ICE negotiation until the callee has answered.
  In addition, either side may wish to hide their location from the
  other side entirely by forcing all traffic through a TURN server.

  In ordinary operation, the site learns the browser's IP address,
  though it may be hidden via mechanisms like Tor
  <https://www.torproject.org> or a VPN.  However, because sites can
  cause the browser to provide IP addresses, this provides a mechanism
  for sites to learn about the user's network environment even if the
  user is behind a VPN that masks their IP address.  Implementations
  may wish to provide settings which suppress all non-VPN candidates if
  the user is on certain kinds of VPN, especially privacy-oriented
  systems such as Tor.  See [RFC8828] for additional information.

4.3.  Communications Security

  Finally, we consider a problem familiar from the SIP world:
  communications security.  For obvious reasons, it MUST be possible
  for the communicating parties to establish a channel which is secure
  against both message recovery and message modification.  (See
  [RFC5479] for more details.)  This service must be provided for both
  data and voice/video.  Ideally the same security mechanisms would be
  used for both types of content.  Technology for providing this
  service (for instance, SRTP [RFC3711], DTLS [RFC6347], and DTLS-SRTP
  [RFC5763]) is well understood.  However, we must examine this
  technology in the WebRTC context, where the threat model is somewhat
  different.

  In general, it is important to understand that unlike a conventional
  SIP proxy, the calling service (i.e., the Web server) controls not
  only the channel between the communicating endpoints but also the
  application running on the user's browser.  While in principle it is
  possible for the browser to cut the calling service out of the loop
  and directly present trusted information (and perhaps get consent),
  practice in modern browsers is to avoid this whenever possible.
  "In-flow" modal dialogs which require the user to consent to specific
  actions are particularly disfavored as human factors research
  indicates that unless they are made extremely invasive, users simply
  agree to them without actually consciously giving consent
  [abarth-rtcweb].  Thus, nearly all the UI will necessarily be
  rendered by the browser but under control of the calling service.
  This likely includes the peer's identity information, which, after
  all, is only meaningful in the context of some calling service.

  This limitation does not mean that preventing attack by the calling
  service is completely hopeless.  However, we need to distinguish
  between two classes of attack:

  Retrospective compromise of calling service:
     The calling service is non-malicious during a call but
     subsequently is compromised and wishes to attack an older call
     (often called a "passive attack").

  During-call attack by calling service:
     The calling service is compromised during the call it wishes to
     attack (often called an "active attack").

  Providing security against the former type of attack is practical
  using the techniques discussed in Section 4.3.1.  However, it is
  extremely difficult to prevent a trusted but malicious calling
  service from actively attacking a user's calls, either by mounting a
  Man-in-the-Middle (MITM) attack or by diverting them entirely.  (Note
  that this attack applies equally to a network attacker if
  communications to the calling service are not secured.)  We discuss
  some potential approaches in Section 4.3.2.

4.3.1.  Protecting Against Retrospective Compromise

  In a retrospective attack, the calling service was uncompromised
  during the call, but an attacker subsequently wants to recover the
  content of the call.  We assume that the attacker has access to the
  protected media stream as well as full control of the calling
  service.

  If the calling service has access to the traffic keying material (as
  in Security Descriptions (SDES) [RFC4568]), then retrospective attack
  is trivial.  This form of attack is particularly serious in the Web
  context because it is standard practice in Web services to run
  extensive logging and monitoring.  Thus, it is highly likely that if
  the traffic key is part of any HTTP request it will be logged
  somewhere and thus subject to subsequent compromise.  It is this
  consideration that makes an automatic, public key-based key exchange
  mechanism imperative for WebRTC (this is a good idea for any
  communications security system), and this mechanism SHOULD provide
  Forward Secrecy (FS).  The signaling channel/calling service can be
  used to authenticate this mechanism.

  In addition, if end-to-end keying is used, the system MUST NOT
  provide any APIs to either extract long-term keying material or to
  directly access any stored traffic keys.  Otherwise, an attacker who
  subsequently compromised the calling service might be able to use
  those APIs to recover the traffic keys and thus compromise the
  traffic.

4.3.2.  Protecting Against During-Call Attack

  Protecting against attacks during a call is a more difficult
  proposition.  Even if the calling service cannot directly access
  keying material (as recommended in the previous section), it can
  simply mount a man-in-the-middle attack on the connection, telling
  Alice that she is calling Bob and Bob that he is calling Alice, while
  in fact the calling service is acting as a calling bridge and
  capturing all the traffic.  Protecting against this form of attack
  requires positive authentication of the remote endpoint such as
  explicit out-of-band key verification (e.g., by a fingerprint) or a
  third-party identity service as described in [RFC8827].

4.3.2.1.  Key Continuity

  One natural approach is to use "key continuity".  While a malicious
  calling service can present any identity it chooses to the user, it
  cannot produce a private key that maps to a given public key.  Thus,
  it is possible for the browser to note a given user's public key and
  generate an alarm whenever that user's key changes.  The Secure Shell
  (SSH) protocol [RFC4251] uses a similar technique.  (Note that the
  need to avoid explicit user consent on every call precludes the
  browser requiring an immediate manual check of the peer's key.)

  Unfortunately, this sort of key continuity mechanism is far less
  useful in the WebRTC context.  First, much of the virtue of WebRTC
  (and any Web application) is that it is not bound to a particular
  piece of client software.  Thus, it will be not only possible but
  routine for a user to use multiple browsers on different computers
  that will of course have different keying material (Securely
  Available Credentials (SACRED) [RFC3760] notwithstanding).  Thus,
  users will frequently be alerted to key mismatches which are in fact
  completely legitimate, with the result that they are trained to
  simply click through them.  As it is known that users routinely will
  click through far more dire warnings [cranor-wolf], it seems
  extremely unlikely that any key continuity mechanism will be
  effective rather than simply annoying.

  Moreover, it is trivial to bypass even this kind of mechanism.
  Recall that unlike the case of SSH, the browser never directly gets
  the peer's identity from the user.  Rather, it is provided by the
  calling service.  Even enabling a mechanism of this type would
  require an API to allow the calling service to tell the browser "this
  is a call to user X."  All the calling service needs to do to avoid
  triggering a key continuity warning is to tell the browser that "this
  is a call to user Y" where Y is confusable with X.  Even if the user
  actually checks the other side's name (which all available evidence
  indicates is unlikely), this would require (a) the browser to use the
  trusted UI to provide the name and (b) the user to not be fooled by
  similar appearing names.

4.3.2.2.  Short Authentication Strings

  ZRTP [RFC6189] uses a "Short Authentication String" (SAS) which is
  derived from the key agreement protocol.  This SAS is designed to be
  compared by the users (e.g., read aloud over the voice channel or
  transmitted via an out-of-band channel) and if confirmed by both
  sides precludes MITM attack.  The intention is that the SAS is used
  once and then key continuity (though with a different mechanism from
  that discussed above) is used thereafter.

  Unfortunately, the SAS does not offer a practical solution to the
  problem of a compromised calling service.  "Voice cloning" systems,
  which mimic the voice of a given speaker are an active area of
  research [deepfakes-ftc] and are already being used in real-world
  attacks [deepfakes-fraud].  These attacks are likely to improve in
  future, especially in an environment where the user just wants to get
  on with the phone call.  Thus, even if the SAS is effective today, it
  is likely not to be so for much longer.

  Additionally, it is unclear that users will actually use an SAS.  As
  discussed above, the browser UI constraints preclude requiring the
  SAS exchange prior to completing the call and so it must be
  voluntary; at most the browser will provide some UI indicator that
  the SAS has not yet been checked.  However, it is well known that
  when faced with optional security mechanisms, many users simply
  ignore them [whitten-johnny].

  Once users have checked the SAS once, key continuity is required to
  avoid them needing to check it on every call.  However, this is
  problematic for reasons indicated in Section 4.3.2.1.  In principle
  it is of course possible to render a different UI element to indicate
  that calls are using an unauthenticated set of keying material
  (recall that the attacker can just present a slightly different name
  so that the attack shows the same UI as a call to a new device or to
  someone you haven't called before), but as a practical matter, users
  simply ignore such indicators even in the rather more dire case of
  mixed content warnings.

4.3.2.3.  Third-Party Identity

  The conventional approach to providing communications identity has of
  course been to have some third-party identity system (e.g., PKI) to
  authenticate the endpoints.  Such mechanisms have proven to be too
  cumbersome for use by typical users (and nearly too cumbersome for
  administrators).  However, a new generation of Web-based identity
  providers (BrowserID, Federated Google Login, Facebook Connect, OAuth
  [RFC6749], OpenID [OpenID], WebFinger [RFC7033]) has been developed
  and use Web technologies to provide lightweight (from the user's
  perspective) third-party authenticated transactions.  It is possible
  to use systems of this type to authenticate WebRTC calls, linking
  them to existing user notions of identity (e.g., Facebook
  adjacencies).  Specifically, the third-party identity system is used
  to bind the user's identity to cryptographic keying material which is
  then used to authenticate the calling endpoints.  Calls which are
  authenticated in this fashion are naturally resistant even to active
  MITM attack by the calling site.

  Note that there is one special case in which PKI-style certificates
  do provide a practical solution: calls from end users to large sites.
  For instance, if you are making a call to Amazon.com, then Amazon can
  easily get a certificate to authenticate their media traffic, just as
  they get one to authenticate their Web traffic.  This does not
  provide additional security value in cases in which the calling site
  and the media peer are one and the same, but might be useful in cases
  in which third parties (e.g., ad networks or retailers) arrange for
  calls but do not participate in them.

4.3.2.4.  Page Access to Media

  Identifying the identity of the far media endpoint is a necessary but
  not sufficient condition for providing media security.  In WebRTC,
  media flows are rendered into HTML5 MediaStreams which can be
  manipulated by the calling site.  Obviously, if the site can modify
  or view the media, then the user is not getting the level of
  assurance they would expect from being able to authenticate their
  peer.  In many cases, this is acceptable because the user values
  site-based special effects over complete security from the site.
  However, there are also cases where users wish to know that the site
  cannot interfere.  In order to facilitate that, it will be necessary
  to provide features whereby the site can verifiably give up access to
  the media streams.  This verification must be possible both from the
  local side and the remote side.  I.e., users must be able to verify
  that the person called has engaged a secure media mode (see
  Section 4.3.3).  In order to achieve this, it will be necessary to
  cryptographically bind an indication of the local media access policy
  into the cryptographic authentication procedures detailed in the
  previous sections.

  It should be noted that the use of this secure media mode is left to
  the discretion of the site.  When such a mode is engaged, the browser
  will need to provide indicia to the user that the associated media
  has been authenticated as coming from the identified user.  This
  allows WebRTC services that wish to claim end-to-end security to do
  so in a way that can be easily verified by the user.  This model
  requires that the remote party's browser be included in the TCB, as
  described in Section 3.

4.3.3.  Malicious Peers

  One class of attack that we do not generally try to prevent is
  malicious peers.  For instance, no matter what confidentiality
  measures you employ the person you are talking to might record the
  call and publish it on the Internet.  Similarly, we do not attempt to
  prevent them from using voice or video processing technology for
  hiding or changing their appearance.  While technologies (Digital
  Rights Management (DRM), etc.) do exist to attempt to address these
  issues, they are generally not compatible with open systems and
  WebRTC does not address them.

  Similarly, we make no attempt to prevent prank calling or other
  unwanted calls.  In general, this is in the scope of the calling
  site, though because WebRTC does offer some forms of strong
  authentication, that may be useful as part of a defense against such
  attacks.

4.4.  Privacy Considerations

4.4.1.  Correlation of Anonymous Calls

  While persistent endpoint identifiers can be a useful security
  feature (see Section 4.3.2.1), they can also represent a privacy
  threat in settings where the user wishes to be anonymous.  WebRTC
  provides a number of possible persistent identifiers such as DTLS
  certificates (if they are reused between connections) and RTCP CNAMEs
  (if generated according to [RFC6222] rather than the privacy-
  preserving mode of [RFC7022]).  In order to prevent this type of
  correlation, browsers need to provide mechanisms to reset these
  identifiers (e.g., with the same lifetime as cookies).  Moreover, the
  API should provide mechanisms to allow sites intended for anonymous
  calling to force the minting of fresh identifiers.  In addition, IP
  addresses can be a source of call linkage [RFC8828].

4.4.2.  Browser Fingerprinting

  Any new set of API features adds a risk of browser fingerprinting,
  and WebRTC is no exception.  Specifically, sites can use the presence
  or absence of specific devices as a browser fingerprint.  In general,
  the API needs to be balanced between functionality and the
  incremental fingerprint risk.  See [Fingerprinting].

5.  Security Considerations

  This entire document is about security.

6.  IANA Considerations

  This document has no IANA actions.

7.  References

7.1.  Normative References

  [RFC2119]  Bradner, S., "Key words for use in RFCs to Indicate
             Requirement Levels", BCP 14, RFC 2119,
             DOI 10.17487/RFC2119, March 1997,
             <https://www.rfc-editor.org/info/rfc2119>.

  [RFC8174]  Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC
             2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174,
             May 2017, <https://www.rfc-editor.org/info/rfc8174>.

7.2.  Informative References

  [abarth-rtcweb]
             Barth, A., "Prompting the user is security failure", RTC-
             Web Workshop, September 2010, <http://rtc-
             web.alvestrand.com/home/papers/barth-security-
             prompt.pdf?attredirects=0>.

  [cranor-wolf]
             Sunshine, J., Egelman, S., Almuhimedi, H., Atri, N., and
             L. Cranor, "Crying Wolf: An Empirical Study of SSL Warning
             Effectiveness", Proceedings of the 18th USENIX Security
             Symposium, August 2009,
             <https://www.usenix.org/legacy/event/sec09/tech/
             full_papers/sunshine.pdf>.

  [deepfakes-fraud]
             Statt, N., "Thieves are now using AI deepfakes to trick
             companies into sending them money", September 2019,
             <https://www.theverge.com/2019/9/5/20851248/deepfakes-ai-
             fake-audio-phone-calls-thieves-trick-companies-stealing-
             money>.

  [deepfakes-ftc]
             Lyons, K., "FTC says the tech behind audio deepfakes is
             getting better", January 2020,
             <https://www.theverge.com/2020/1/29/21080553/ftc-
             deepfakes-audio-cloning-joe-rogan-phone-scams>.

  [fetch]    van Kesteren, A., "Fetch",
             <https://fetch.spec.whatwg.org/>.

  [finer-grained]
             Jackson, C. and A. Barth, "Beware of Finer-Grained
             Origins", Web 2.0 Security and Privacy (W2SP 2008), July
             2008.

  [Fingerprinting]
             Doty, N., Ed., "Mitigating Browser Fingerprinting in Web
             Specifications", March 2019,
             <https://www.w3.org/TR/fingerprinting-guidance/>.

  [huang-w2sp]
             Huang, L-S., Chen, E.Y., Barth, A., Rescorla, E., and C.
             Jackson, "Talking to Yourself for Fun and Profit", Web 2.0
             Security and Privacy (W2SP 2011), May 2011.

  [OpenID]   Sakimura, N., Bradley, J., Jones, M., de Medeiros, B., and
             C. Mortimore, "OpenID Connect Core 1.0", November 2014,
             <https://openid.net/specs/openid-connect-core-1_0.html>.

  [RFC2818]  Rescorla, E., "HTTP Over TLS", RFC 2818,
             DOI 10.17487/RFC2818, May 2000,
             <https://www.rfc-editor.org/info/rfc2818>.

  [RFC3261]  Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston,
             A., Peterson, J., Sparks, R., Handley, M., and E.
             Schooler, "SIP: Session Initiation Protocol", RFC 3261,
             DOI 10.17487/RFC3261, June 2002,
             <https://www.rfc-editor.org/info/rfc3261>.

  [RFC3552]  Rescorla, E. and B. Korver, "Guidelines for Writing RFC
             Text on Security Considerations", BCP 72, RFC 3552,
             DOI 10.17487/RFC3552, July 2003,
             <https://www.rfc-editor.org/info/rfc3552>.

  [RFC3711]  Baugher, M., McGrew, D., Naslund, M., Carrara, E., and K.
             Norrman, "The Secure Real-time Transport Protocol (SRTP)",
             RFC 3711, DOI 10.17487/RFC3711, March 2004,
             <https://www.rfc-editor.org/info/rfc3711>.

  [RFC3760]  Gustafson, D., Just, M., and M. Nystrom, "Securely
             Available Credentials (SACRED) - Credential Server
             Framework", RFC 3760, DOI 10.17487/RFC3760, April 2004,
             <https://www.rfc-editor.org/info/rfc3760>.

  [RFC4251]  Ylonen, T. and C. Lonvick, Ed., "The Secure Shell (SSH)
             Protocol Architecture", RFC 4251, DOI 10.17487/RFC4251,
             January 2006, <https://www.rfc-editor.org/info/rfc4251>.

  [RFC4568]  Andreasen, F., Baugher, M., and D. Wing, "Session
             Description Protocol (SDP) Security Descriptions for Media
             Streams", RFC 4568, DOI 10.17487/RFC4568, July 2006,
             <https://www.rfc-editor.org/info/rfc4568>.

  [RFC5479]  Wing, D., Ed., Fries, S., Tschofenig, H., and F. Audet,
             "Requirements and Analysis of Media Security Management
             Protocols", RFC 5479, DOI 10.17487/RFC5479, April 2009,
             <https://www.rfc-editor.org/info/rfc5479>.

  [RFC5763]  Fischl, J., Tschofenig, H., and E. Rescorla, "Framework
             for Establishing a Secure Real-time Transport Protocol
             (SRTP) Security Context Using Datagram Transport Layer
             Security (DTLS)", RFC 5763, DOI 10.17487/RFC5763, May
             2010, <https://www.rfc-editor.org/info/rfc5763>.

  [RFC6189]  Zimmermann, P., Johnston, A., Ed., and J. Callas, "ZRTP:
             Media Path Key Agreement for Unicast Secure RTP",
             RFC 6189, DOI 10.17487/RFC6189, April 2011,
             <https://www.rfc-editor.org/info/rfc6189>.

  [RFC6222]  Begen, A., Perkins, C., and D. Wing, "Guidelines for
             Choosing RTP Control Protocol (RTCP) Canonical Names
             (CNAMEs)", RFC 6222, DOI 10.17487/RFC6222, April 2011,
             <https://www.rfc-editor.org/info/rfc6222>.

  [RFC6347]  Rescorla, E. and N. Modadugu, "Datagram Transport Layer
             Security Version 1.2", RFC 6347, DOI 10.17487/RFC6347,
             January 2012, <https://www.rfc-editor.org/info/rfc6347>.

  [RFC6454]  Barth, A., "The Web Origin Concept", RFC 6454,
             DOI 10.17487/RFC6454, December 2011,
             <https://www.rfc-editor.org/info/rfc6454>.

  [RFC6455]  Fette, I. and A. Melnikov, "The WebSocket Protocol",
             RFC 6455, DOI 10.17487/RFC6455, December 2011,
             <https://www.rfc-editor.org/info/rfc6455>.

  [RFC6749]  Hardt, D., Ed., "The OAuth 2.0 Authorization Framework",
             RFC 6749, DOI 10.17487/RFC6749, October 2012,
             <https://www.rfc-editor.org/info/rfc6749>.

  [RFC7022]  Begen, A., Perkins, C., Wing, D., and E. Rescorla,
             "Guidelines for Choosing RTP Control Protocol (RTCP)
             Canonical Names (CNAMEs)", RFC 7022, DOI 10.17487/RFC7022,
             September 2013, <https://www.rfc-editor.org/info/rfc7022>.

  [RFC7033]  Jones, P., Salgueiro, G., Jones, M., and J. Smarr,
             "WebFinger", RFC 7033, DOI 10.17487/RFC7033, September
             2013, <https://www.rfc-editor.org/info/rfc7033>.

  [RFC7675]  Perumal, M., Wing, D., Ravindranath, R., Reddy, T., and M.
             Thomson, "Session Traversal Utilities for NAT (STUN) Usage
             for Consent Freshness", RFC 7675, DOI 10.17487/RFC7675,
             October 2015, <https://www.rfc-editor.org/info/rfc7675>.

  [RFC8445]  Keranen, A., Holmberg, C., and J. Rosenberg, "Interactive
             Connectivity Establishment (ICE): A Protocol for Network
             Address Translator (NAT) Traversal", RFC 8445,
             DOI 10.17487/RFC8445, July 2018,
             <https://www.rfc-editor.org/info/rfc8445>.

  [RFC8825]  Alvestrand, H., "Overview: Real-Time Protocols for
             Browser-Based Applications", RFC 8825,
             DOI 10.17487/RFC8825, January 2021,
             <https://www.rfc-editor.org/info/rfc8825>.

  [RFC8827]  Rescorla, E., "WebRTC Security Architecture", RFC 8827,
             DOI 10.17487/RFC8827, January 2021,
             <https://www.rfc-editor.org/info/rfc8827>.

  [RFC8828]  Uberti, J. and G. Shieh, "WebRTC IP Address Handling
             Requirements", RFC 8828, DOI 10.17487/RFC8828, January
             2021, <https://www.rfc-editor.org/info/rfc8828>.

  [SWF]      "SWF File Format Specification Version 19", April 2013,
             <https://www.adobe.com/content/dam/acom/en/devnet/pdf/swf-
             file-format-spec.pdf>.

  [whitten-johnny]
             Whitten, A. and J.D. Tygar, "Why Johnny Can't Encrypt: A
             Usability Evaluation of PGP 5.0", Proceedings of the 8th
             USENIX Security Symposium, August 1999,
             <https://www.usenix.org/legacy/publications/library/
             proceedings/sec99/whitten.html>.

Acknowledgements

  Bernard Aboba, Harald Alvestrand, Dan Druta, Cullen Jennings, Alan
  Johnston, Hadriel Kaplan (Section 4.2.1), Matthew Kaufman, Martin
  Thomson, Magnus Westerlund.

Author's Address

  Eric Rescorla
  Mozilla

  Email: [email protected]

You are viewing proxied material from gopher.fnord.one. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.