wdiff draft-ietf-rtcweb-security-00.txt draft-ietf-rtcweb-security-01.txt

RTC-Web E. Rescorla
Internet-Draft RTFM, Inc.
Intended status: Standards Track September 21, October 30, 2011
Expires: March 24, May 2, 2012

Security Considerations for RTC-Web
draft-ietf-rtcweb-security-00
draft-ietf-rtcweb-security-01

Abstract

The Real-Time Communications on the Web (RTC-Web) working group is
tasked with standardizing protocols for real-time communications
between Web browsers. The major use cases for RTC-Web 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 soft phones) RTC-Web communications are directly controlled by
some Web server, which poses new security challenges. For instance,
a Web browser might expose a JavaScript API which allows a server to
place a video call. Unrestricted access to such an API would allow
any site which a user visited to "bug" a user's computer, capturing
any activity which passed in front of their camera. This document
defines the RTC-Web threat model and defines an architecture which
provides security within that threat model.

Legal

THIS DOCUMENT AND THE INFORMATION CONTAINED THEREIN ARE PROVIDED ON
AN "AS IS" BASIS AND THE CONTRIBUTOR, THE ORGANIZATION HE/SHE
REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE
IETF TRUST, AND THE INTERNET ENGINEERING TASK FORCE, DISCLAIM ALL
WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY
WARRANTY THAT THE USE OF THE INFORMATION THEREIN WILL NOT INFRINGE
ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS
FOR A PARTICULAR PURPOSE.

Status of this Memo

This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at http://datatracker.ietf.org/drafts/current/.

Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."

This Internet-Draft will expire on March 24, May 2, 2012.

This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://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 . . . . . . . . . . . . . . . . . . . . . . . . . 4 5
2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 5 6
3. The Browser Threat Model . . . . . . . . . . . . . . . . . . . 5 6
3.1. Access to Local Resources . . . . . . . . . . . . . . . . 6 7
3.2. Same Origin Policy . . . . . . . . . . . . . . . . . . . . 6 7
3.3. Bypassing SOP: CORS, WebSockets, and consent to
communicate . . . . . . . . . . . . . . . . . . . . . . . 7 8
4. Security for RTC-Web Applications . . . . . . . . . . . . . . 7 8
4.1. Access to Local Devices . . . . . . . . . . . . . . . . . 7 8
4.1.1. Calling Scenarios and User Expectations . . . . . . . 8 9
4.1.1.1. Dedicated Calling Services . . . . . . . . . . . . 8 9
4.1.1.2. Calling the Site You're On . . . . . . . . . . . . 8 9
4.1.1.3. Calling to an Ad Target . . . . . . . . . . . . . 9 10
4.1.2. Origin-Based Security . . . . . . . . . . . . . . . . 9 10
4.1.3. Security Properties of the Calling Page . . . . . . . 11 12
4.2. Communications Consent Verification . . . . . . . . . . . 12 13
4.2.1. ICE . . . . . . . . . . . . . . . . . . . . . . . . . 12 13
4.2.2. Masking . . . . . . . . . . . . . . . . . . . . . . . 12 14
4.2.3. Backward Compatibility . . . . . . . . . . . . . . . . 13 14
4.2.4. IP Location Privacy . . . . . . . . . . . . . . . . . 15
4.3. Communications Security . . . . . . . . . . . . . . . . . 14 15
4.3.1. Protecting Against Retrospective Compromise . . . . . 15 16
4.3.2. Protecting Against During-Call Attack . . . . . . . . 15 17
4.3.2.1. Key Continuity . . . . . . . . . . . . . . . . . . 16 17
4.3.2.2. Short Authentication Strings . . . . . . . . . . . 16 18
4.3.2.3. Recommendations . . . . . . . . . . . . . . . . . 17 19
5. Security Considerations . . . . . . . . . . . . . . . . . . . 17 19
6. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 18 19
7. References . . . . . . . . . . . . . . . . . . . . . . . . . . 18 19
7.1. Normative References . . . . . . . . . . . . . . . . . . . 18 19
7.2. Informative References . . . . . . . . . . . . . . . . . . 18
Author's Address 20
Appendix A. A Proposed Security Architecture [No Consensus on
This] . . . . . . . . . . . . . . . . . . . . . . . . 22
A.1. Trust Hierarchy . 20

1. Introduction

The Real-Time . . . . . . . . . . . . . . . . . . . . 22
A.1.1. Authenticated Entities . . . . . . . . . . . . . . . . 22
A.1.2. Unauthenticated Entities . . . . . . . . . . . . . . . 23
A.2. Overview . . . . . . . . . . . . . . . . . . . . . . . . . 23
A.2.1. Initial Signaling . . . . . . . . . . . . . . . . . . 24
A.2.2. Media Consent Verification . . . . . . . . . . . . . . 26
A.2.3. DTLS Handshake . . . . . . . . . . . . . . . . . . . . 26
A.2.4. Communications on the Web (RTC-Web) working group is
tasked with standardizing protocols for real-time communications
between Web browsers. The major use cases for RTC-Web 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) RTC-Web Consent Freshness . . . . . . . . . 27
A.3. Detailed Technical Description . . . . . . . . . . . . . . 27
A.3.1. Origin and Web Security Issues . . . . . . . . . . . . 27
A.3.2. Device Permissions Model . . . . . . . . . . . . . . . 28
A.3.3. Communications Consent . . . . . . . . . . . . . . . . 29
A.3.4. IP Location Privacy . . . . . . . . . . . . . . . . . 29
A.3.5. Communications Security . . . . . . . . . . . . . . . 30
A.3.6. Web-Based Peer Authentication . . . . . . . . . . . . 31
A.3.6.1. Generic Concepts . . . . . . . . . . . . . . . . . 31
A.3.6.2. BrowserID . . . . . . . . . . . . . . . . . . . . 32
A.3.6.3. OAuth . . . . . . . . . . . . . . . . . . . . . . 35
A.3.6.4. Generic Identity Support . . . . . . . . . . . . . 36
Author's Address . . . . . . . . . . . . . . . . . . . . . . . . . 36

1. Introduction

Figure 1: A simple RTC-Web system

In the system shown in Figure 1, Alice and Bob both have RTC-Web
enabled browsers and they visit some Web server which operates a
calling service. Each of their browsers exposes standardized
JavaScript calling APIs which are used by the Web server to set up a
call between Alice and Bob. 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 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 attacks. Any successful system will need to be
resistant to this and other attacks.

2. Terminology

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 [RFC2119].

3. The Browser Threat Model

The security requirements for RTC-Web 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:

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 RTC-Web 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 JS 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, and
NETWORK ATTACKERS, who are able to control your network. Network
attackers correspond to the [RFC3552] "Internet Threat Model". In
general, it is desirable to build a system which is secure against
both kinds of attackers, but realistically many sites do not run
HTTPS [RFC2818] and so our ability to defend against network
attackers is necessarily somewhat limited. Most of the rest of this
section is devoted to web attackers, with the assumption that
protection against network attackers is provided by running HTTPS.

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 screw yourself", as
extensive research shows that users are prone to consent under such
circumstances.]

Similarly, while Flash SWFs 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
XMLHttpRequest() API those requests are not allowed to be made to any
server, but rather solely to the same ORIGIN from whence the script
came.[I-D.abarth-origin] (although CORS [CORS] and WebSockets
[I-D.ietf-hybi-thewebsocketprotocol] provides a 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 his
credentials) and the server (e.g., from DoS attack).

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 the 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 Cross-Origin Resource Sharing (CORS) spec [CORS] is a
response to this demand. In CORS, when a script from origin A
executes what would otherwise be a forbidden 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 [I-D.ietf-hybi-thewebsocketprotocol] 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.
In particular, 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 RTC-Web 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.

For the rest of this discussion we assume that the user is somehow
going to grant consent to some entity (e.g., a social networking
site) to initiate a call on his behalf. This consent may be limited
to a single call or may be a general consent. In order for the user
to make an intelligent decision about whether to allow a call (and
hence his camera and microphone input to be routed somewhere), he
must understand either who is requesting access, where the media is
going, or both. So, for instance, one might imagine that at the time
access to camera and microphone is requested, the user is shown a
dialog that says "site X has requested access to camera and
microphone, yes or no" (though note that this type of in-flow
interface violates one of the guidelines in Section 3). The user's
decision will of course be based on his opinion of Site X. However,
as discussed below, this is a complicated concept.

4.1.1. 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.

4.1.1.1. 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 that 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. In general, this is difficult
as most of the user interface is presented by the calling site.
However, communications security mechanisms can be used to give some
assurance, as described in Section 4.3.2.

4.1.1.2. 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 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.

4.1.1.3. Calling to an Ad Target

In both of the previous cases, the user has a direct relationship
(though perhaps a transient one) with the target of the call.
Moreover, in both cases he is actually visiting the site of the
person he is being asked to trust. However, this is not always so.
Consider the case where a user is a visiting a content site which
hosts an advertisement with an invitation to call for more
information. When the user clicks the ad, they are connected with
the advertiser or their agent.

The relationships here are far more complicated: the site the user
is actually visiting has no direct relationship with the advertiser;
they are just hosting ads from an ad network. The user has no
relationship with the ad network, but desires one with the
advertiser, at least for long enough to learn about their products.
At minimum, then, whatever consent dialog is shown needs to allow the
user to have some idea of the organization that they are actually
calling.

However, because the user also has some relationship with the hosting
site, it is also arguable that the hosting site should be allowed to
express an opinion (e.g., to be able to allow or forbid a call) since
a bad experience with an advertiser reflect negatively on the hosting
site [this idea was suggested by Adam Barth]. However, this
obviously presents a privacy challenge, as sites which host
advertisements often learn very little about whether individual users
clicked through to the ads, or even which ads were presented.

4.1.2. Origin-Based Security

As discussed in Section 3.2, the basic unit of Web sandboxing is the
origin, and so it is natural to scope consent to origin.
Specifically, a script from origin A MUST only be allowed to initiate
communications (and hence to access 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 origin, this creates a difficult
mismatch.

Arguably, 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 origin, then at any future
visit to that site (including one induced via 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 case, 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. Unfortunately, Callee-oriented consent does not work well
because a malicious site can claim that the user is calling any user
of his 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
advertising case described in Section 4.1.1.3, 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 origin
here, but to the media origin (and-or destination) rather than to the
Web origin.

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. However, this is also not that convenient an
interface, since managing individual user permissions can be painful.

While this is primarily a question not for IETF, it should be clear
that there is no really good answer. In general, if you cannot trust
the site which you have authorized for calling not to bug you then
your security situation is not really ideal. It is RECOMMENDED that
browsers have explicit (and obvious) indicators that they are in a
call in order to mitigate this risk.

4.1.3. 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 apparently
http://calling-service.example.com/ to inject JS to initiate a
call to himself.

Note that this attack does not depend on the media being insecure.
Because the call is to the attacker, it is also encrypted to him.
Moreover, it need not be executed immediately; the attacker can
"infect" the origin semi-permanently (e.g., with a web worker or a
popunder) 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 sites, if the site embeds
active content (e.g., JavaScript) that is shown below.

Figure 1: A simple RTC-Web system

In the system shown or from
an untrusted site, because that JavaScript is executed in Figure 1, Alice and Bob both have the
security context of the page [finer-grained]. Thus, it is also
dangerous to allow RTC-Web
enabled browsers and they visit some Web server functionality from HTTPS origins that
embed mixed content. Note: this issue is not restricted to PAGES
which operates contain mixed content. If a
calling service. Each page from a given origin ever
loads mixed content then it is possible for a network attacker to
infect the browser's notion of their browsers exposes standardized
JavaScript calling APIs which are used by that origin semi-permanently.

[[ OPEN ISSUE: What recommendation should IETF make about (a)
whether RTCWeb long-term consent should be available over HTTP pages
and (b) How to handle origins where the consent is to an HTTPS URL
but the page contains active mixed content? ]]

4.2. Communications Consent Verification

As discussed in Section 3.3, allowing web applications unrestricted
network access via the browser introduces the Web server risk of using the
browser as an attack platform against machines which would not
otherwise be accessible to set up a
call between Alice and Bob. While this system is the malicious site, for instance because
they are topologically
similar to restricted (e.g., behind a conventional SIP-based system (with the Web server
acting firewall or NAT).
In order to prevent this form of attack as the signaling service and browsers acting well as softphones),
control has moved cross-protocol
attacks it is important to the central Web server; the browser simply
provides API points require that are used by the calling service. As with
any Web application, the Web server can move logic between target of traffic
explicitly consent to receiving the server
and JavaScript traffic in question. Until that
consent has been verified for a given endpoint, traffic other than
the browser, but regardless consent handshake MUST NOT be sent to that endpoint.

4.2.1. ICE

Verifying receiver consent requires some sort of where explicit handshake,
but conveniently we already need one in order to do NAT hole-
punching. ICE [RFC5245] includes a handshake designed to verify that
the code receiving element wishes to receive traffic from the sender. It
is
executing, it important to remember here that the site initiating ICE is ultimately under control of
presumed malicious; in order for the server.

It should handshake to be immediately apparent that this type of system poses new
security challenges beyond those secure the
receiving element MUST demonstrate receipt/knowledge of a conventional VoIP system. In
particular, it needs some value
not available to the site (thus preventing the site from forging
responses). In order to contend achieve this objective with malicious calling services. For
example, if ICE, the calling service can cause STUN
transaction IDs must be generated by the browser and MUST NOT be made
available to make the initiating script, even via a call
at any time diagnostic interface.
Verifying receiver consent also requires verifying the receiver wants
to any callee of its choice, then receive traffic from a particular sender, and at this facility can be
used to bug time; for
example a user's computer without their knowledge, malicious site may simply by
placing a call attempt ICE to some recording service. More subtly, if the
exposed APIs allow known servers that
are using ICE for other sessions. ICE provides this verification as
well, by using the server STUN credentials as a form of per-session shared
secret. Those credentials are known to instruct the browser Web application, but
would need to send
arbitrary content, then they can also be known and used to bypass firewalls or mount
denial of service attacks. Any successful system will need by the STUN-receiving element to
be
resistant to this and other attacks.

2. Terminology

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are useful.

There also needs to be interpreted as described in RFC 2119 [RFC2119].

3. The Browser Threat Model

The security requirements some mechanism for RTC-Web follow directly from the
requirement browser to verify that
the browser's job is to protect target of the user. traffic continues to wish to receive it.
Obviously, some ICE-based mechanism will work here, but it has been
observed that because ICE keepalives are indications, they will not
work here, so some other mechanism is needed.

4.2.2. Masking

Once consent is verified, there still is some concern about
misinterpretation attacks as described by Huang et
al. [huang-w2sp] summarize the core browser security guarantee as:

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

It al.[huang-w2sp].
As long as communication is important limited to realize that this includes sites hosting arbitrary
malicious scripts. The motivation for UDP, then this requirement risk is simple:
it
probably limited, thus masking is trivial not required for attackers to divert users UDP. I.e., once
communications consent has been verified, it is most likely safe to sites of their choice.
For instance, an attacker can purchase display advertisements which
direct
allow the user (either automatically or via user clicking) implementation to send arbitrary UDP traffic to their
site, at which point the browser will execute the attacker's scripts.
Thus, it is important chosen
destination, provided that it be safe to view arbitrarily malicious
pages. Of course, browsers inevitably have bugs which cause them the STUN keepalives continue to
fall short of succeed.
In particular, this goal, but any new RTC-Web functionality must be
designed with is true for the intent data channel if DTLS is used
because DTLS (with the anti-chosen plaintext mechanisms required by
TLS 1.1) does not allow the attacker to meet this standard. The remainder generate predictable
ciphertext. However, with TCP the risk of
this section provides transparent proxies
becomes much more background on severe. If TCP is to be used, then WebSockets
style masking MUST be employed.

4.2.3. Backward Compatibility

A requirement to use ICE limits compatibility with legacy non-ICE
clients. It seems unsafe to completely remove the existing Web security
model.

In this model, then, requirement for
some check. All proposed checks have the common feature that the
browser acts as a TRUSTED COMPUTING BASE
(TCB) both from sends some message to the user's perspective candidate traffic recipient and
refuses to some extent from 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
server's. While HTML and JS provided Web application cannot forge them,
generally by having the server can cause 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
browser to execute a variety of actions, those scripts operate in legacy endpoint has a
sandbox that isolates them both from
public address) include:

o STUN checks without using ICE (i.e., the user's computer and from
each other, non-RTC-web endpoint sets
up a STUN responder.)
o Use or RTCP as detailed below.

Conventionally, we refer to either WEB ATTACKERS, who are able an implicit reachability check.

In the RTCP approach, the RTC-Web endpoint is allowed to
induce you send a
limited number of RTP packets prior to visit their sites but receiving consent. This
allows a short window of attack. In addition, some legacy endpoints
do not control the network, and
NETWORK ATTACKERS, who are able support RTCP, so this is a much more expensive solution for
such endpoints, for which it would likely be easier to control your network. Network
attackers correspond implement ICE.
For these two reasons, an RTCP-based approach does not seem to
address the [RFC3552] "Internet Threat Model". security issue satisfactorily.

In
general, it is desirable to build a system which the STUN approach, the RTC-Web endpoint is secure against
both kinds of attackers, but realistically many sites do not run
HTTPS [RFC2818] and so our ability able to defend against network
attackers verify that the
recipient is necessarily somewhat limited. Most running some kind of STUN endpoint but unless the rest of this
section STUN
responder is devoted to web attackers, integrated with the assumption ICE username/password establishment
system, the RTC-Web endpoint cannot verify that
protection against network attackers is provided by running HTTPS.

3.1. Access to Local Resources

While the browser has access recipient
consents to local resources such as keying
material, files, 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 camera and systems are tightly integrated (i.e., the microphone, it strictly limits or
forbids web servers from accessing those same resources. For
instance, while it STUN endpoint
responds with responses authenticated with ICE credentials) then this
issue does not exist. However, such a design is possible very close to produce an HTML form which will
allow file upload,
ICE-Lite implementation (indeed, arguably is one). An intermediate
approach would be to have a script cannot do so STUN extension that indicated that one
was responding to RTC-Web checks but not computing integrity checks
based on the ICE credentials. This would allow the use of standalone
STUN servers without user consent and in
fact cannot even suggest a specific file (e.g., /etc/passwd); the
user must explicitly select risk of confusing them with legacy STUN
servers. If a non-ICE legacy solution is needed, then this is
probably the file and best choice.

Once initial consent is verified, we also need to its upload.
[Note: verify continuing
consent, in many cases browsers are explicitly designed order to avoid
dialogs with attacks where two people briefly share an
IP (e.g., behind a NAT in an Internet cafe) and the semantics of "click here attacker arranges
for a large, unstoppable, traffic flow to screw yourself", as
extensive research shows that users the network and then
leaves. The appropriate technologies here are prone fairly similar to consent under such
circumstances.]

Similarly, while Flash SWFs can access
those for initial consent, though are perhaps weaker since the camera and microphone,
they explicitly require
threats is less severe.

[[ OPEN ISSUE: Exactly what should be the requirements here?
Proposals include ICE all the time or ICE but with allowing one of
these non-ICE things for legacy. ]]

4.2.4. IP Location Privacy

Note that as soon as the user consent callee sends their ICE candidates, the
callee 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 that access. avoid tracking, implementations may wish to
suppress the start of ICE negotiation until the callee has answered.
In addition, some resources simply cannot be accessed either side may wish to hide their location entirely by
forcing all traffic through a TURN server.

4.3. Communications Security

Finally, we consider a problem familiar from the browser
at all. 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, there DTLS [RFC4347] and DTLS-SRTP [RFC5763]) is
well understood. However, we must examine this technology to the
RTC-Web context, where the threat model is no real way somewhat different.

In general, it is important to run specific
executables directly from understand that unlike a script (though conventional
SIP proxy, 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 calling service (i.e., the
XMLHttpRequest() API those requests are Web server) controls not allowed to be made to any
server, but rather solely to
only the same ORIGIN from whence channel between the script
came.[I-D.abarth-origin] (although CORS [CORS] and WebSockets
[I-D.ietf-hybi-thewebsocketprotocol] provides a escape hatch from
this restriction, as described below. This SAME ORIGIN POLICY (SOP)
prevents server A from mounting attacks communicating endpoints but also the
application running on server B via the user's
browser, which protects both browser. While in principle it is
possible for the user (e.g., from misuse browser to cut the calling service out of his
credentials) and the server (e.g., from DoS attack).

More generally, SOP forces scripts from each site to run loop
and directly present trusted information (and perhaps get consent),
practice in their
own, isolated, sandboxes. While there are techniques modern browsers is to allow them avoid this whenever possible. "In-
flow" modal dialogs which require the user to interact, those interactions generally must be mutually consensual
(by each site) and consent to specific
actions are limited particularly disfavored as human factors research
indicates that unless they are made extremely invasive, users simply
agree to certain channels. For instance,
multiple pages/browser panes from them without actually consciously giving consent.
[abarth-rtcweb]. Thus, nearly all the same origin can read each
other's JS variables, UI will necessarily be
rendered by the browser but pages from under control of the different origins--or even
iframes from different origins on calling service.
This likely includes the same page--cannot.

3.3. Bypassing SOP: CORS, WebSockets, 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 is non-malicious during a call but
subsequently is compromised and consent wishes to communicate

While SOP serves attack an important security function, it also makes older call.

During-call attack by calling service.
The calling service is compromised during the call it
inconvenient wishes to write certain classes
attack.

Providing security against the former type of applications. In
particular, mash-ups, attack is practical
using the techniques discussed in which Section 4.3.1. However, it is
extremely difficult to prevent a script from origin A uses resources trusted but malicious calling
service from origin B, can only be achieved via actively attacking a certain amount of hackery.
The W3C Cross-Origin Resource Sharing (CORS) spec [CORS] is user's calls, either by mounting a
response to
MITM attack or by diverting them entirely. (Note that this demand. In CORS, when attack
applies equally to a script from origin A
executes what would otherwise network attacker if communications to the
calling service are not secured.) We discuss some potential
approaches and why they are likely to be impractical in
Section 4.3.2.

4.3.1. Protecting Against Retrospective Compromise

In a forbidden cross-origin request, retrospective attack, the browser instead contacts calling service was uncompromised
during the target server to determine whether
it is willing call, but that an attacker subsequently wants to allow cross-origin requests from A. If it is so
willing, the browser then allows recover
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 [I-D.ietf-hybi-thewebsocketprotocol] allows cross-origin
establishment content of transparent channels. Once a WebSockets connection the call. We assume that the attacker has been established from a script access to a site,
the script can exchange
any traffic it likes without being required to frame it protected media stream as a series well as having full control 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
calling service.

If the calling service has
shown that designing a correct consent verification system is
difficult. In particular, Huang et al. [huang-w2sp] have shown
vulnerabilities in access to the existing Java and Flash consent verification
techniques and traffic keying material (as
in a simplified version SDES [RFC4568]), then retrospective attack is trivial. This form
of attack is particularly serious in the WebSockets handshake.
In particular, Web context because it is important to be wary of CROSS-PROTOCOL attacks
standard practice in which Web services to run extensive logging and
monitoring. Thus, it is highly likely that if the attacking script generates traffic which key is acceptable
to some non-Web protocol state machine. In order to resist this form
part of attack, WebSockets incorporates a masking technique intended to
randomize the bits on the wire, thus making any HTTP request it more difficult will be logged somewhere and thus subject
to
generate traffic which resembles a given protocol.

4. Security subsequent compromise. It is this consideration that makes an
automatic, public key-based key exchange mechanism imperative for
RTC-Web Applications

4.1. Access to Local Devices

As discussed in Section 1, allowing arbitrary sites (this is a good idea for any communications security system)
and this mechanism SHOULD provide perfect forward secrecy (PFS). The
signaling channel/calling service can be used to initiate calls
violates authenticate this
mechanism.

In addition, the core Web security guarantee; without some access
restrictions on local devices, any malicious site could simply bug a
user. At minimum, then, it system MUST NOT be possible for arbitrary sites provide any APIs to initiate calls extract either
long-term keying material or to arbitrary location without user consent. This
immediately raises directly access any stored traffic
keys. Otherwise, an attacker who subsequently compromised the question, however, of what should
calling service might be able to use those APIs to recover the scope
of user consent.

For
traffic keys and thus compromise the rest of this discussion we assume 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 user calling service is somehow
going to grant consent to some entity (e.g., acting as a social networking
site) calling bridge and
capturing all the traffic. While in theory it is possible to initiate a call on his behalf. This consent may be limited
construct techniques which protect against this form of attack, in
practice these techniques all require far too much user intervention
to a single call or may be a general consent. In order for practical, given the user interface constraints described in
[abarth-rtcweb].

4.3.2.1. Key Continuity

One natural approach is to make an intelligent decision about whether to allow use "key continuity". While a call (and
hence his camera and microphone input malicious
calling service can present any identity it chooses to be routed somewhere), he
must understand either who is request access, where the media is
going, or both. So, for instance, one might imagine user, it
cannot produce a private key that at the time
access maps to camera and microphone a given public key. Thus,
it is requested, possible for the user is shown a
dialog that says "site X has requested access browser to camera and
microphone, yes or no" (though note that this type of in-flow
interface violates one of the guidelines in Section Section 3). The
user's decision will of course be based on his opinion of Site X.
However, as discussed below, this is a complicated concept.

4.1.1. Calling Scenarios given user's public key and User Expectations

While
generate an alarm whenever that user's key changes. SSH [RFC4251]
uses a large number similar technique. (Note that the need to avoid explicit user
consent on every call precludes the browser requiring an immediate
manual check of possible calling scenarios are possible, the
scenarios discussed in peer's key).

Unfortunately, this section illustrate many sort of key continuity mechanism is far less
useful in the
difficulties RTC-Web context. First, much of identifying the relevant scope virtue of consent.

4.1.1.1. Dedicated Calling Services

The first scenario we consider RTC-Web
(and any Web application) is that it is not bound to particular piece
of client software. Thus, it will be not only possible but routine
for a dedicated calling service. In
this case, the user has a relationship with a calling site and
repeatedly makes calls to use multiple browsers on it. It different computers which will
of course have different keying material (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 likely 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 having simply annoying.

Moreover, it is trivial to give permission for each call bypass even this kind of mechanism.
Recall that unlike the user will want to give case of SSH, the
calling service long-term access to browser never directly gets
the camera and microphone. This peer's identity from the user. Rather, it is a natural fit for a long-term consent mechanism (e.g., installing
an app store "application" to indicate permission for provided by the
calling
service.) A variant service. Even enabling a mechanism of this type would
require an API to allow the dedicated calling service to tell the browser "this
is a gaming
site (e.g., a poker site) which hosts a dedicated call to user X". All the calling service needs to
allow players do to avoid
triggering a key continuity warning is to tell the browser that "this
is a call each other.

With any kind of service to user Y" where Y is close to X. Even if the user may use actually
checks the same service other side's name (which all available evidence indicates
is unlikely), this would require (a) the browser to
talk trusted UI to many different people, there is a question about whether
provide the name and (b) the user can know who they are talking to. In general, this 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 difficult
as most of
derived from the user interface key agreement protocol. This SAS is presented designed to be
read over the voice channel and if confirmed by both sides precludes
MITM attack. The intention is that the calling site.
However, communications security mechanisms can be SAS is used once and then key
continuity (though a different mechanism from that discussed above)
is used thereafter.

Unfortunately, the SAS does not offer a practical solution to give some
assurance, as described the
problem of a compromised calling service. "Voice conversion"
systems, which modify voice from one speaker to make it sound like
another, are an active area of research. These systems are already
good enough to fool both automatic recognition systems
[farus-conversion] and humans [kain-conversion] in many cases, and
are of course likely to improve in Section 4.3.2.

4.1.1.2. Calling the Site You're On

Another simple scenario is calling the site you're actually visiting.
The paradigmatic case here is future, especially in an
environment where the "click here to talk user just wants to a
representative" windows that appear get on many shopping sites. In this
case, with the user's expectation phone call.
Thus, even if SAS is that they are calling the site
they're actually visiting. However, effective today, 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 likely not to be able to activate my microphone whenever they please. Thus, this
suggests the need so for a second consent mechanism where I only grant
consent
much longer. Moreover, it is possible for an attacker who controls
the duration of a given call. As described in
Section 3.1, great care must be taken in the design of this interface browser to avoid allow the users just clicking through.

4.1.1.3. Calling SAS to an Ad Target

In both of the previous cases, succeed and then simulate call
failure and reconnect, trusting that the user will not notice that
the "no SAS" indicator has a direct relationship
(though perhaps a transient one) with been set (which seems likely).

Even were SAS secure if used, it seems exceedingly unlikely that
users will actually use it. As discussed above, the target of 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 call.
Moreover, in both cases he SAS has not yet been checked.
However, it it is actually visiting well-known that when faced with optional mechanisms
such as fingerprints, users simply do not check them [whitten-johnny]
Thus, it is highly unlikely that users will ever perform the site of SAS
exchange.

Once uses have checked the
person he SAS once, key continuity is being asked required to trust.
avoid them needing to check it on every call. However, this is not always so.
Consider the case where a user
problematic for reasons indicated in Section 4.3.2.1. In principle
it is of course possible to render a visiting a content site which
hosts an advertisement with an invitation different UI element to call for more
information. When the user clicks the ad, they are connected with
the advertiser or their agent.

The relationships here indicate
that calls are far more complicated: the site the user
is actually visiting has no direct relationship with using an unauthenticated set of keying material
(recall that the advertiser;
they are attacker can just hosting ads from an ad network. The user has no
relationship with present a slightly different name
so that the ad network, but desires one with attack shows the
advertiser, at least for long enough same UI as a call to learn about their products.
At minimum, then, whatever consent dialog is shown needs a new device or to allow
someone you haven't called before) but as a practical matter, users
simply ignore such indicators even in the
user to have some idea rather more dire case of the organization that they
mixed content warnings.

4.3.2.3. Recommendations

[[ OPEN ISSUE: What are actually
calling.

However, because the user also has some relationship with the hosting
site, it is also arguable that best UI recommendations to make?
Proposal: take the hosting site text from [I-D.kaufman-rtcweb-security-ui]
Section 2]]

[[ OPEN ISSUE: Exactly what combination of media security primitives
should be allowed to
express an opinion (e.g., to be able specified and/or mandatory to implement? In particular,
should we allow DTLS-SRTP only, or forbid a call) since
a bad experience with an advertiser reflect negatively on the hosting
site [this idea was suggested by Adam Barth]. However, this
obviously presents a privacy challenge, as sites which host
advertisements often learn very little both DTLS-SRTP and SDES. Should
we allow RTP for backward compatibility? ]]

5. Security Considerations

This entire document is about whether individual users
clicked through security.

6. Acknowledgements

Bernard Aboba, Harald Alvestrand, Cullen Jennings, Hadriel Kaplan (S
4.2.1), Matthew Kaufman, Magnus Westerland.

7. References

7.1. Normative References

[RFC2119] Bradner, S., "Key words for use in RFCs to the ads, or even which ads were presented.

4.1.2. Origin-Based Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.

7.2. Informative References

[CORS] van Kesteren, A., "Cross-Origin Resource Sharing".

[I-D.abarth-origin]
Barth, A., "The Web Origin Concept",
draft-abarth-origin-09 (work in progress), November 2010.

[I-D.ietf-hybi-thewebsocketprotocol]
Fette, I. and A. Melnikov, "The WebSocket protocol",
draft-ietf-hybi-thewebsocketprotocol-17 (work in
progress), September 2011.

[I-D.kaufman-rtcweb-security-ui]
Kaufman, M., "Client Security User Interface Requirements
for RTCWEB", draft-kaufman-rtcweb-security-ui-00 (work in
progress), June 2011.

[RFC2818] Rescorla, E., "HTTP Over TLS", RFC 2818, May 2000.

[RFC3261] Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston,
A., Peterson, J., Sparks, R., Handley, M., and E.
Schooler, "SIP: Session Initiation Protocol", RFC 3261,
June 2002.

[RFC3552] Rescorla, E. and B. Korver, "Guidelines for Writing RFC
Text on Security

As discussed in Section 3.2, the basic unit of Web sandboxing is the
origin, Considerations", BCP 72, RFC 3552,
July 2003.

[RFC3760] Gustafson, D., Just, M., and so it is natural to scope consent to origin.
Specifically, a script from origin A MUST only be allowed to initiate
communications (and hence to access camera M. Nystrom, "Securely
Available Credentials (SACRED) - Credential Server
Framework", RFC 3760, April 2004.

[RFC4251] Ylonen, T. and microphone) if the
user has specifically authorized access C. Lonvick, "The Secure Shell (SSH)
Protocol Architecture", RFC 4251, January 2006.

[RFC4347] Rescorla, E. and N. Modadugu, "Datagram Transport Layer
Security", RFC 4347, April 2006.

[RFC4568] Andreasen, F., Baugher, M., and D. Wing, "Session
Description Protocol (SDP) Security Descriptions for that origin. It is Media
Streams", RFC 4568, July 2006.

[RFC5245] Rosenberg, J., "Interactive Connectivity Establishment
(ICE): A Protocol for Network Address Translator (NAT)
Traversal for Offer/Answer Protocols", RFC 5245,
April 2010.

[RFC5479] Wing, D., Fries, S., Tschofenig, H., and F. Audet,
"Requirements and Analysis of
course technically possible to have coarser-scoped permissions, but
because the Web model is scoped to origin, this creates Media Security Management
Protocols", RFC 5479, April 2009.

[RFC5763] Fischl, J., Tschofenig, H., and E. Rescorla, "Framework
for Establishing a difficult
mismatch.

Arguably, origin is not fine-grained enough. Consider Secure Real-time Transport Protocol
(SRTP) Security Context Using Datagram Transport Layer
Security (DTLS)", RFC 5763, May 2010.

[RFC6189] Zimmermann, P., Johnston, A., and J. Callas, "ZRTP: Media
Path Key Agreement for Unicast Secure RTP", RFC 6189,
April 2011.

[abarth-rtcweb]
Barth, A., "Prompting the situation
where Alice visits a site and authorizes it to make a single call.
If consent user is expressed solely in terms security failure", RTC-
Web Workshop.

[cranor-wolf]
Sunshine, J., Egelman, S., Almuhimedi, H., Atri, N., and
L. cranor, "Crying Wolf: An Empirical Study of SSL Warning
Effectiveness", Proceedings of origin, then at any future
visit to that site (including one induced via mash-up or ad network),
the site can bug Alice's computer, use the computer 18th USENIX Security
Symposium, 2009.

[farus-conversion]
Farrus, M., Erro, D., and J. Hernando, "Speaker
Recognition Robustness to place bogus
calls, etc. While in principle Alice could grant Voice Conversion".

[finer-grained]
Barth, A. and then revoke the
privilege, in practice privileges accumulate; if we are concerned
about this attack, something else is needed. There are a number C. Jackson, "Beware of
potential countermeasures Finer-Grained
Origins", W2SP, 2008.

[huang-w2sp]
Huang, L-S., Chen, E., Barth, A., Rescorla, E., and C.
Jackson, "Talking to this sort of issue.

Individual Consent
Ask the user for permission Yourself for each call.

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

Cryptographic Consent
Only allow calls to a given set Fun and Profit", W2SP,
2011.

[kain-conversion]
Kain, A. and M. Macon, "Design and Evaluation of peer keying material or to a
cryptographically established identity.

Unfortunately, none Voice
Conversion Algorithm based on Spectral Envelope Mapping
and Residual Prediction", Proceedings of ICASSP, May
2001.

[whitten-johnny]
Whitten, A. and J. Tygar, "Why Johnny Can't Encrypt: A
Usability Evaluation of PGP 5.0", Proceedings of these approaches is satisfactory for all
cases. As discussed above, individual consent puts the user's
approval 8th
USENIX Security Symposium, 1999.

Appendix A. A Proposed Security Architecture [No Consensus on This]

This section contains a proposed security architecture, based on the
considerations discussed in the UI flow for every call. Not only does main body of this quickly
become annoying but it can train memo. This section
is currently the user to simply click "OK", at
which point opinion of the consent becomes useless. Thus, while it may be
necssary to author and does not have individual consent in consensus
though some case, (many?) elements of this proposal do seem to have general
consensus.

A.1. Trust Hierarchy

The basic assumption of this proposal is not that network resources exist
in a
suitable solution for (for instance) hierarchy of trust, rooted in the browser, which serves as the
user's TRUSTED COMPUTING BASE (TCB). Any security property which the calling service case. Where
necessary, in-flow
user interfaces wishes to have enforced must be carefully designed to
avoid ultimately guaranteed by the risk of
browser (or transitively by some property the browser verifies).
Conversely, if the browser is compromised, then no security
guarantees are possible. Note that there are cases (e.g., Internet
kiosks) where the user blindly clicking through.

The can't really trust the browser that much. In
these cases, the level of security provided is limited by how much
they trust the browser.

Optimally, we would not rely on trust in any entities other than the
browser. However, this is unfortunately not possible if we wish to
have a functional system. Other network elements fall into two options
categories: those which can be authenticated by the browser and thus
are designed to restrict calls partly trusted--though to the minimum extent necessary--and those
which cannot be authenticated and thus are untrusted. This is a given
target. Unfortunately, Callee-oriented consent
natural extension of the end-to-end principle.

A.1.1. Authenticated Entities

There are two major classes of authenticated entities in the system:

o Calling services: Web sites whose origin we can verify (optimally
via HTTPS).
o Other users: RTC-Web peers whose origin we can verify
cryptographically (optimally via DTLS-SRTP).

Note that merely being authenticated does not work well make these entities
trusted. For instance, just because a malicious site we can claim verify that the user is calling any user
of his choice. One fix for this
https://www.evil.org/ is to tie calls to a
cryptographically established identity. While owned by Dr. Evil does not suitable for all
cases, this approach may be useful for some. If mean that we consider the
advertising case described in Section 4.1.1.3, it's not particularly
convenient to require the advertiser can
trust Dr. Evil to instantiate access our camera an iframe on microphone. However, it gives
the
hosting site just to get permission; a more convenient approach is user an opportunity to
cryptographically tie the advertiser's certificate determine whether he wishes to the
communication directly. We're still tieing permissions trust Dr.
Evil or not; after all, if he desires to origin
here, but contact Dr. Evil, it's safe
to the media origin (and-or destination) rather than temporarily give him access to the
Web origin.

Another case where media-level cryptographic identity makes sense is
when a user really does not trust camera and microphone for the calling site. For instance, I
might be worried that
purpose of 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. However, this call. The point here is also not that convenient an
interface, since managing individual user permissions we must first identify
other elements before we can be painful.

While this is primarily a question not for IETF, it should be clear determine whether to trust them.

It's also worth noting that there are settings where authentication
is no really good answer. In general, if you cannot trust non-cryptographic, such as other machines behind a firewall.
Naturally, the site which you level of trust one can have authorized for calling in identities verified in
this way depends on how strong the topology enforcement is.

A.1.2. Unauthenticated Entities

Other than the above entities, we are not generally able to bug you then
your security situation is identify
other network elements, thus we cannot trust them. This does not really ideal. It is RECOMMENDED
mean that
browsers it is not possible to have explicit (and obvious) indicators any interaction with them, but
it means that we must assume that they will behave maliciously and
design a system which is secure even if they do so.

A.2. Overview

This section describes a typical RTCWeb session and shows how the
various security elements interact and what guarantees are provided
to the user. The example in this section is a
call "best case" scenario
in order to mitigate this risk.

4.1.3. Security Properties which we provide the maximal amount of user authentication and
media privacy with the Calling Page

Origin-based minimal level of trust in the calling service.
Simpler versions with lower levels of security is intended are also possible and
are noted in the text where applicable. It's also important to
recognize the tension between security against web attackers.
However, (or performance) and privacy.
The example shown here is aimed towards settings where we must also consider are more
concerned about secure calling than about privacy, but as we shall
see, there are settings where one might wish to make different
tradeoffs--this architecture is still compatible with those settings.

For the case purposes of network attackers.
Consider this example, we assume the case where I topology shown in the
figure below. This topology is derived from the topology shown in
Figure 1, but separates Alice and Bob's identities from the process
of signaling. Specifically, Alice and Bob have granted permission to relationships with
some Identity Provider (IDP) that supports a calling
service by an origin protocol such OpenID or
BrowserID) that has the HTTP scheme, e.g.,
http://calling-service.example.com. If I ever use my computer can be used to attest to their identity. This
separation isn't particularly important in "closed world" cases where
Alice and Bob are users on an
unsecured network (e.g., a hotspot or if my own home wireless the same social 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 have
identities based on that network. However, there are important
settings where that this site is unaffiliated with not the calling service.
2. The attacker modifies my HTTP connection case, such as federation (calls from
one network to inject an IFRAME (or another) and calling on untrusted sites, such as where
two users who have a redirect) to http://calling-service.example.com
3. The attacker forges the response apparently
http://calling-service.example.com/ to inject JS to initiate relationship via a
call given social network want to himself.

Note that this attack does not depend
call each other on the media being insecure.
Because the another, untrusted, site, such as a poker site.

+----------------+
| |
| Signaling |
| Server |
| |
+----------------+
^ ^
/ \
HTTPS / \ HTTPS
/ \
/ \
v v
JS API JS API
+-----------+ +-----------+
| | Media | |
Alice | Browser |<---------->| Browser | Bob
| | (DTLS-SRTP)| |
+-----------+ +-----------+
^ ^--+ +--^ ^
| | | |
v | | v
+-----------+ | | +-----------+
| |<--------+ | |
| IDP | | | IDP |
| | +------->| |
+-----------+ +-----------+

Figure 2: A call is to with IDP-based identity

A.2.1. Initial Signaling

Alice and Bob are both users of a common calling service; they both
have approved the attacker, it is also encrypted calling service to make calls (we defer the
discussion of device access permissions till later). They are both
connected to him.
Moreover, it need not be executed immediately; the attacker can
"infect" calling service via HTTPS and so know the origin semi-permanently (e.g.,
with a web worker or a
popunder) and thus be able to bug me long after I some level of confidence. They also have left the
infected network. accounts with some
identity provider. This risk sort of identity service is created by allowing calls at all from
a page fetched over HTTP.

Even if calls are only possible from HTTPS sites, if becoming
increasingly common in the site embeds
active content (e.g., JavaScript) that Web environment in technologies such
(BrowserID, Federated Google Login, Facebook Connect, OAuth, OpenID,
WebFinger), and is fetched over HTTP often provided as a side effect service of your
ordinary accounts with some service. In this example, we show Alice
and Bob using a separate identity service, though they may actually
be using the same identity service as calling service or have no
identity service at all.

Alice is logged onto the calling service and decides to call Bob. She
can see from
an untrusted site, because the calling service that JavaScript he is executed online and the calling
service presents a JS UI in the
security context form of the page [finer-grained]. Thus, it is also
dangerous a button next to allow RTC-Web functionality from HTTPS origins Bob's name
which says "Call". Alice clicks the button, which initiates a JS
callback that
embed mixed content. Note: this issue is instantiates a PeerConnection object. This does not restricted to PAGES
which contain mixed content. If
require a page security check: JS from a given any origin ever
loads mixed content then it is possible for a network attacker allowed to
infect get this
far.

Once the browser's notion of that origin semi-permanently.

[[ OPEN ISSUE: What recommendation should IETF make about (a)
whether RTCWeb long-term consent should be available over HTTP pages PeerConnection is created, the calling service JS needs to
set up some media. Because this is an audio/video call, it creates
two MediaStreams, one connected to an audio input and (b) How one connected
to handle origins where a video input. At this point the consent first security check is
required: untrusted origins are not allowed to an HTTPS URL
but access the page contains active mixed content? ]]

4.2. Communications Consent Verification

As discussed camera and
microphone. In this case, because Alice is a long-term user of the
calling service, she has made a permissions grant (i.e., a setting in Section 3.3, allowing web applications unrestricted
access to
the via browser) to allow the calling service to access her camera and
microphone any time it wants. The browser network introduces checks this setting when
the risk of using camera and microphone requests are made and thus allows them.

In the current W3C API, once some streams have been added, Alice's
browser as an attack platform against machines which would not
otherwise + JS generates a signaling message The format of this data is
currently undefined. It may be accessible to the malicious site, for instance because
they are topologically restricted (e.g., behind a firewall complete message as defined by ROAP
[REF] or NAT). may be assembled piecemeal by the JS. In order to prevent this form of attack as well as cross-protocol
attacks either case, it is important
will contain:

o Media channel information
o ICE candidates
o A fingerprint attribute binding the message to require that Alice's public key
[RFC5763]

Prior to sending out the target of traffic
explicitly consent signaling message, the PeerConnection code
contacts the identity service and obtains an assertion binding
Alice's identity to receiving her fingerprint. The exact details depend on the traffic
identity service (though as discussed in question. Until that
consent has been verified for a given endpoint, traffic other than
the consent handshake MUST NOT Appendix A.3.6.4 I believe
PeerConnection can be sent agnostic to that endpoint.

4.2.1. ICE

Verifying receiver consent requires some sort of explicit handshake, them), but conveniently we already need one in order for now it's easiest to do NAT hole-
punching. ICE [RFC5245] includes
think of as a handshake designed BrowserID assertion.

This message is sent to verify that the receiving element wishes to receive traffic from signaling server, e.g., by XMLHttpRequest
[REF] or by WebSockets [I-D.ietf-hybi-thewebsocketprotocol]. The
signaling server processes the sender. It message from Alice's browser,
determines that this is important a call to remember here that Bob and sends a signaling message
to Bob's browser (again, the site initiating ICE format is
presumed malicious; in order for currently undefined). The JS
on Bob's browser processes it, and alerts Bob to the handshake incoming call
and to be secure Alice's identity. In this case, Alice has provided an
identity assertion and so Bob's browser contacts Alice's identity
provider (again, this is done in a generic way so the
receiving element MUST demonstrate receipt/knowledge browser has no
specific knowledge of some value
not available to the site (thus preventing it from forging
responses). In order IDP) to achieve this objective with ICE, verity the STUN
transaction IDs must be generated by assertion. This allows
the browser and MUST NOT be made
available to the initiating script, even via display a diagnostic interface.

4.2.2. Masking

Once consent trusted element indicating that a call is verified, there still
coming in from Alice. If Alice is in Bob's address book, then this
interface might also include her real name, a picture, etc. The
calling site will also provide some concern about
misinterpretation attacks as described by Huang et al.[huang-w2sp].
As long as communication is limited user interface element (e.g., a
button) to UDP, then allow Bob to answer the call, though this risk is
probably limited, thus masking most likely
not part of the trusted UI.

If Bob agrees [I am ignoring early media for now], a PeerConnection
is instantiated with the message from Alice's side. Then, a similar
process occurs as on Alice's browser: Bob's browser verifies that
the calling service is not required for UDP. I.e., once
communications consent has been verified, it approved, the media streams are created, and a
return signaling message containing media information, ICE
candidates, and a fingerprint is most likely safe sent back to
allow Alice via the implementation to send arbitrary UDP traffic to signaling
service. If Bob has a relationship with an IDP, the chosen
destination, provided message will
also come with an identity assertion.

At this point, Alice and Bob each know that the STUN keepalives continue other party wants to succeed.
However,
have a secure call with TCP them. Based purely on the risk of transparent proxies becomes much more
severe. If TCP interface provided
by the signaling server, they know that the signaling server claims
that the call is from Alice to be used, then WebSockets style masking MUST be
employed.

4.2.3. Backward Compatibility

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 Bob. Because the common feature far end sent an
identity assertion along with their message, they know that this is
verifiable from the
browser sends some message IDP as well. Of course, the call works perfectly
well if either Alice or Bob doesn't have a relationship with an IDP;
they just get a lower level of assurance. Moreover, Alice might wish
to make an anonymous call through an anonymous calling site, in which
case she would of course just not provide any identity assertion and
the candidate traffic recipient calling site would mask her identity from Bob.

A.2.2. Media Consent Verification

As described in Section 4.2. This proposal specifies that that be
performed via ICE. Thus, Alice and
refuses Bob perform ICE checks with each
other. At the completion of these checks, they are ready to send other traffic until that message has been replied to.
The message/reply pair must be generated in such a way
non-ICE data.

At this point, Alice knows that an
attacker (a) Bob (assuming he is verified via
his IDP) or someone else who controls the Web application cannot forge them,
generally by having the message contain some secret value signaling service is claiming is Bob
is willing to exchange traffic with her and (b) that must
be incorporated (e.g., echoed, hashed into, etc.). Non-ICE
candidates for this role (in cases where either Bob is at
the legacy endpoint IP address which she has a
public address) include:

o STUN checks without using verified via ICE (i.e., the non-RTC-web endpoint sets
up a STUN responder.)
o Use or RTCP as there is an implicit reachability check.

In the RTCP approach, the RTC-Web endpoint attacker
who is allowed on-path 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 that IP address detouring the traffic. Note that
it is a much more expensive solution for
such endpoints, not possible for which it would likely be easier to implement ICE.
For these two reasons, an RTCP-based approach does attacker who is on-path but not seem attached to
address
the signaling service to spoof these checks because they do not have
the ICE credentials. Bob's security issue satisfactorily.

In guarantees with respect to Alice
are the STUN approach, converse of this.

A.2.3. DTLS Handshake

Once the RTC-Web endpoint ICE checks have completed [more specifically, once some ICE
checks have completed], Alice and Bob can set up a secure channel.
This is able to verify that performed via DTLS [RFC4347] (for the data channel) and DTLS-
SRTP [RFC5763] for the media channel. Specifically, Alice and Bob
perform a DTLS handshake on every channel which has been established
by ICE. The total number of channels depends on the
recipient is running some kind amount of STUN endpoint but unless
muxing; in the STUN
responder is integrated with most likely case we are using both RTP/RTCP mux and
muxing multiple media streams on the ICE username/password establishment
system, same channel, in which case
there is only one DTLS handshake. Once the RTC-Web endpoint cannot verify that DTLS handshake has
completed, the recipient
consents keys are extracted and used to key SRTP for the media
channels.

At this particular call. This may be an issue if existing
STUN servers are operated at addresses point, Alice and Bob know that they share a set of secure
data and/or media channels with keys which are not able known to handle
bandwidth-based attacks. Thus, this approach does not seem
satisfactory either. any
third-party attacker. If the systems are tightly integrated (i.e., the STUN endpoint
responds with responses Alice and Bob authenticated with ICE credentials) via their IDPs,
then this
issue does they also know that the signaling service is not exist. However, such attacking them.
Even if they do not use an IDP, as long as they have minimal trust in
the signaling service not to perform a design man-in-the-middle attack, they
know that their communications are secure against the signaling
service as well.

A.2.4. Communications and Consent Freshness

From a security perspective, everything from here on in is very close a little
anticlimactic: Alice and Bob exchange data protected by the keys
negotiated by DTLS. Because of the security guarantees discussed in
the previous sections, they know that the communications are
encrypted and authenticated.

The one remaining security property we need to an
ICE-Lite implementation (indeed, arguably establish is one). An intermediate
approach would be "consent
freshness", i.e., allowing Alice to have a STUN extension that indicated verify that one
was responding Bob is still prepared
to RTC-Web checks but not computing integrity checks
based on the receive her communications. ICE credentials. This would allow the use of standalone
STUN servers without the risk of confusing them with legacy specifies periodic STUN
servers. If a non-ICE legacy solution is needed, then this
keepalizes but only if media is
probably not flowing. Because the best choice.

[TODO: Write something about consent freshness
issue is more difficult here, we require RTCWeb implementations to
periodically send keepalives. If a keepalive fails and RTCP].

[[ OPEN ISSUE: Exactly what should be the requirements here?
Proposals include no new ICE all
channels can be established, then the time or ICE but with allowing one session is terminated.

A.3. Detailed Technical Description

A.3.1. Origin and Web Security Issues

The basic unit of
these non-ICE things permissions for legacy. ]]

4.3. Communications Security

Finally, we consider a problem familiar from RTC-Web is the SIP world:
communications security. For obvious reasons, it MUST be possible
for origin
[I-D.abarth-origin]. Because the communicating parties security of the origin depends on
being able to establish a channel which is secure
against both message recovery and message modification. (See
[RFC5479] for more details.) This service must authenticate content from that origin, the origin can
only be provided for both securely established if data is transferred over HTTPS.
Thus, clients MUST treat HTTP and voice/video. Ideally the same security mechanisms would be
used for both types of content. Technology for providing this
service (for instance, DTLS [RFC4347] HTTPS origins as different
permissions domains and DTLS-SRTP [RFC5763]) is
well understood. However, we must examine this technology SHOULD NOT permit access to the any RTC-Web context, where the threat model is somewhat different.

In general,
functionality from scripts fetched over non-secure (HTTP) origins.
If an HTTPS origin contains mixed active content (regardless of
whether 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 present on the user's browser. While in principle it is
possible for the browser specific page attempting to cut the calling service out of access RTC-
Web functionality), any access MUST be treated as if it came from the loop
HTTP origin. For instance, if a https://www.example.com/example.html
loads https://www.example.com/example.js and directly present trusted information (and perhaps get consent),
practice in modern browsers is
http://www.example.org/jquery.js, any attempt by example.js to avoid this whenever possible. "In-
flow" modal dialogs which require the access
RTCWeb functionality MUST be treated as if it came from
http://www.example.com/. Note that many browsers already track mixed
content and either forbid it by default or display a warning.

A.3.2. Device Permissions Model

Implementations MUST obtain explicit user to consent prior to specific
actions are particularly disfavored as human factors research
indicates that unless they are made extremely invasive, users simply
agree providing
access 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 camera and/or microphone. Implementations MUST at
minimum support the peer's following two permissions models:

o Requests for one-time camera/microphone access.
o Requests for permanent access.

In addition, they SHOULD support requests for access to a single
communicating peer. E.g., "Call customerservice@ford.com". Browsers
servicing such requests SHOULD clearly indicate that identity information, which, after
all, is only meaningful in to the context of some calling service.

This limitation does not mean that preventing attack by
user when asking for permission.

API Requirement: The API MUST provide a mechanism for the calling
service is completely hopeless. However, we need requesting
JS to distinguish
between two classes indicate which of attack:

Retrospective compromise these forms of calling service.
The calling service is permissions it is non-malicious during
requesting. This allows the client to know what sort of user
interface experience to provide. In particular, browsers might
display a call non-invasive door hanger ("some features of this site
may not work..." when asking for long-term permissions) but
subsequently a more
invasive UI ("here is compromised and wishes to attack an older call.

During-call attack by calling service. your own video") for single-call
permissions. The calling service is compromised during API MAY grant weaker permissions than the call JS
asked for if the user chooses to authorize only those permissions,
but if it wishes intends to
attack.

Providing security against grant stronger ones SHOULD display the former type of attack is practical
using
appropriate UI for those permissions.

API Requirement: The API MUST provide a mechanism for the techniques discussed in Section 4.3.1. However, it is
extremely difficult requesting
JS to prevent a trusted but malicious calling
service from actively attacking a user's calls, either by mounting a
MITM attack relinquish the ability to see or by diverting them entirely. (Note that modify the media (e.g., via
MediaStream.record()). Combined with secure authentication of the
communicating peer, this attack
applies equally to allows a network attacker if communications user to be sure that the calling service are
site is not secured.) We discuss some potential
approaches accessing or modifying their conversion.

UI Requirement: The UI MUST clearly indicate when the user's camera
and why they microphone are likely to be impractical in
Section 4.3.2.

4.3.1. Protecting Against Retrospective Compromise

In a retrospective attack, the calling service was uncompromised
during use. This indication MUST NOT be
suppressable by the JS and MUST clearly indicate how to terminate
a call, but that an attacker subsequently wants and provide a UI means to recover immediately stop camera/
microphone input without the JS being able to prevent it.

UI Requirement: If the content UI indication of camera/microphone use are
displayed in the call. We assume browser such that minimizing the attacker has access to
the protected media stream as well as having full control of browser window
would hide the
calling service.

If indication, or the calling service has access to JS creating an overlapping
window would hide the traffic keying material (as
in SDES [RFC4568]), indication, 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 browser SHOULD stop
camera and
monitoring. Thus, it is highly likely that if microphone input.

Clients MAY permit the traffic key is
part formation of data channels without 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 direct
user approval. Because sites can always tunnel data through the
server, further restrictions on the data channel do not provide any
additional security. (though see Appendix A.3.3 for
RTC-Web (this is a good idea for any communications security system)
and this mechanism related issue).

Implementations which support some form of direct user authentication
SHOULD also provide perfect forward secrecy (PFS). The
signaling channel/calling service a policy by which a user can be used authorize calls only
to specific counterparties. Specifically, the implementation SHOULD
provide the following interfaces/controls:

o Allow future calls to this verified user.
o Allow future calls to any verified user who is in my system
address book (this only works with address book integration, of
course).

Implementations SHOULD also provide a different user interface
indication when calls are in progress to users whose identities are
directly verifiable. Appendix A.3.5 provides more on this.

A.3.3. Communications Consent

Browser client implementations of RTC-Web MUST implement ICE. Server
gateway implementations which operate only at public IP addresses may
implement ICE-Lite.

Browser implementations MUST verify reachability via ICE prior to authenticate this
mechanism.

In addition, the system
sending any non-ICE packets to a given destination. Implementations
MUST NOT provide any APIs to extract either
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 ICE transaction ID to recover JavaScript. [Note: this
document takes no position on the
traffic keys split between ICE in JS and thus compromise ICE in
the traffic.

4.3.2. Protecting Against During-Call Attack

Protecting against attacks during a call browser. The above text is a more difficult
proposition. Even written the way it is for editorial
convenience and will be modified appropriately if the calling service cannot directly access
keying material (as recommended WG decides on
ICE in the previous section), it can
simply mount JS.]

Implementations MUST send keepalives no less frequently than every 30
seconds regardless of whether traffic is flowing or not. If a man-in-the-middle attack on
keepalive fails then the connection, telling
Alice implementation MUST either attempt to find a
new valid path via ICE or terminate media for that she is calling Bob and Bob ICE component.
Note that he is calling Alice, while ICE [RFC5245]; Section 10 keepalives use STUN Binding
Indications which are one-way and therefore not sufficient. We will
need to define a new mechanism for this. [OPEN ISSUE: what to do
here.]

A.3.4. IP Location Privacy

As mentioned in fact the calling service is acting as Section 4.2.4 above, a calling bridge and
capturing all side effect of the traffic. While in theory it default ICE
behavior is possible to
construct techniques that the peer learns one's IP address, which protect against this form leaks large
amounts of attack, location information, especially for mobile devices. This
has negative privacy consequences in
practice these techniques all require far too much some circumstances. The
following two API requirements are intended to mitigate this issue:

API Requirement: The API MUST provide a mechanism to suppress ICE
negotiation (though perhaps to allow candidate gathering) until
the user intervention has decided to be practical, given answer the user interface constraints described in
[abarth-rtcweb].

4.3.2.1. Key Continuity

One natural approach call [note: determining when
the call has been answered is a question for the JS.] This
enables a user to use "key continuity". While prevent a malicious peer from learning their IP address if
they elect not to answer a call.

API Requirement: The API MUST provide a mechanism for the calling service can present any identity it chooses
application to indicate that only TURN candidates are to be used.
This prevents the user, it
cannot produce peer from learning one's IP address at all.

A.3.5. Communications Security

Implementations MUST implement DTLS and DTLS-SRTP. All data channels
MUST be secured via DTLS. DTLS-SRTP MUST be offered for every media
channel and MUST be the default; i.e., if an implementation receives
an offer for DTLS-SRTP and SDES and/or plain RTP, DTLS-SRTP MUST be
selected.

[OPEN ISSUE: What should the settings be here? MUST?]
Implementations MAY support SDES and RTP for media traffic for
backward compatibility purposes.

API Requirement: The API MUST provide a private key mechanism to indicate that maps a
fresh DTLS key pair is to be generated for a given public key. Thus,
it specific call. This
is possible intended to allow for the browser unlinkability. Note that there are also
settings where it is attractive to note a given user's public use the same keying material
repeatedly, especially those with key and
generate an alarm whenever continuity-based
authentication.

API Requirement: The API MUST provide a mechanism to indicate that user's a
fresh DTLS key changes. SSH [RFC4251]
uses pair is to be generated for a similar technique. (Note that specific call. This
is intended to allow for unlinkability.

API Requirement: When DTLS-SRTP is used, the need API MUST NOT permit the
JS to avoid explicit obtain the negotiated keying material. This requirement
preserves the end-to-end security of the media.

UI Requirements: A user-oriented client MUST provide an
"inspector" interface which allows the user
consent on every call precludes to determine the browser requiring an immediate
manual check
security characteristics of the peer's key).

Unfortunately, this sort of key continuity mechanism is far less
useful media. [largely derived from
[I-D.kaufman-rtcweb-security-ui]
The following properties SHOULD be displayed "up-front" in the RTC-Web context. First, much of
browser chrome, i.e., without requiring the virtue of RTC-Web
(and any Web application) is that it is not bound user to particular piece
of ask for them:

* A client software. Thus, it will be not only possible but routine MUST provide a user interface through which a user may
determine the security characteristics for currently-displayed
audio and video stream(s)
* A client MUST provide a user to use multiple browsers on different computers interface through which will a user may
determine the security characteristics for transmissions of course have different keying material (SACRED [RFC3760]
notwithstanding.) Thus, users will frequently be alerted
their microphone audio and camera video.
* The "security characteristics" MUST include an indication as to key
mismatches which are in fact completely legitimate, with
whether or not the result
that they are trained to simply click through them. As it transmission is known
that users routinely will click through far more dire warnings
[cranor-wolf], it seems extremely unlikely cryptographically protected
and whether that any key continuity
mechanism will be effective rather than simply annoying.

Moreover, it protection is trivial to bypass even this kind of mechanism.
Recall based on a key that unlike the case was
delivered out-of-band (from a server) or was generated as a
result of SSH, a pairwise negotiation.
* If the browser never far endpoint was directly gets
the peer's identity from verified Appendix A.3.6 the user. Rather, it is provided by
"security characteristics" MUST include the
calling service. Even enabling a mechanism of this type would
require an API verified
information.
The following properties are more likely to allow require some "drill-
down" from the calling service to tell user:

* If the browser "this transmission is a call to user X". All cryptographically protected, the calling service needs to do to avoid
triggering a key continuity warning The
algorithms in use (For example: "AES-CBC" or "Null Cipher".)
* If the transmission is to tell cryptographically protected, the browser that "this
"security characteristics" MUST indicate whether PFS is a call to user Y" where Y
provided.
* If the transmission is close to X. Even if cryptographically protected via an end-
to-end mechanism the user actually
checks "security characteristics" MUST include
some mechanism to allow an out-of-band verification of the other side's name (which all available evidence indicates
peer, such as a certificate fingerprint or an SAS.

A.3.6. Web-Based Peer Authentication

A.3.6.1. Generic Concepts

In a number of cases, it is unlikely), this would require (a) desirable for the browser endpoint (i.e., the
browser) to trusted UI be able to
provide directly identity the name and (b) endpoint on the user other
side without trusting only the signaling service to not which they are
connected. For instance, users may be fooled by similar
appearing names.

4.3.2.2. Short Authentication Strings

ZRTP [RFC6189] uses making a "short call via a federated
system where they wish to get direct authentication string" (SAS) which is
derived from of the key agreement protocol. This SAS is designed to other
side. Alternately, they may be
read over the voice channel and if confirmed by both sides precludes
MITM attack. The intention is that the SAS is used once and then key
continuity (though making a different mechanism from that discussed above)
is used thereafter.

Unfortunately, the SAS does not offer call on a practical solution site which they
minimally trust (such as a poker site) but to someone who has an
identity on a site they do trust (such as a social network.)

Recently, a number of Web-based identity technologies (OAuth,
BrowserID, Facebook Connect), etc. have been developed. While the
problem of
details vary, what these technologies share is that they have a compromised calling service. "Voice conversion"
systems, Web-
based (i.e., HTTP/HTTPS identity provider) which modify voice from one speaker attests to make it sound like
another, are your
identity. For instance, if I have an active area of research. These systems are already
good enough account at example.org, I could
use the example.org identity provider to fool both automatic recognition systems
[farus-conversion] and humans [kain-conversion] in many cases, and
are of course likely prove to improve in future, especially in an
environment where the user just wants others that I was
alice@example.org. The development of these technologies allows us
to get separate calling from identity provision: I could call you on with
Poker Galaxy but identify myself as alice@example.org.

Whatever the phone call.
Thus, even if SAS underlying technology, the general principle is effective today, it that the
party which is likely not to be so for
much longer. Moreover, it being authenticated is NOT the signaling site but
rather the user (and their browser). Similarly, the relying party is possible for an attacker who controls
the browser to allow and not the SAS signaling site. This means that the
PeerConnection API MUST arrange to succeed and then simulate call
failure and reconnect, trusting talk directly to the identity
provider in a way that cannot be impersonated by the calling site.
The following sections provide two examples of this.

A.3.6.2. BrowserID

BrowserID [https://browserid.org/] is a technology which allows a
user will not notice with a verified email address to generate an assertion
(authenticated by their identity provider) attesting to their
identity (phrased as an email address). The way that
the "no SAS" indicator has been set (which seems likely).

Even were SAS secure if used, it seems exceedingly unlikely this is used in
practice is that
users will actually use it. As discussed above, the browser UI
constraints preclude requiring relying party embeds JS in their site which
talks to the BrowserID code (either hosted on a trusted intermediary
or embedded in the browser). That code generates the SAS exchange prior assertion which
is passed back to completing the call and so it must relying party for verification. The assertion
can be voluntary; at most the browser will
provide some UI indicator that verified directly or with a Web service provided by the SAS has not yet been checked.
However, it it
identity provider. It's relatively easy to extend this functionality
to authenticate RTC-Web calls, as shown below.

+----------------------+ +----------------------+
| | | |
| Alice's Browser | | Bob's Browser |
| | OFFER ------------> | |
| Calling JS Code | | Calling JS Code |
| ^ | | ^ |
| | | | | |
| v | | v |
| PeerConnection | | PeerConnection |
| | ^ | | | ^ |
| Finger| |Signed | |Signed | | |
| print | |Finger | |Finger | |"Alice"|
| | |print | |print | | |
| v | | | v | |
| +--------------+ | | +---------------+ |
| | BrowserID | | | | BrowserID | |
| | Signer | | | | Verifier | |
| +--------------+ | | +---------------+ |
| ^ | | ^ |
+-----------|----------+ +----------|-----------+
| |
| Get certificate |
v | Check
+----------------------+ | certificate
| | |
| Identity |/-------------------------------+
| Provider |
| |
+----------------------+

The way this mechanism works is well-known that when faced with optional mechanisms
such as fingerprints, users simply do not check them [whitten-johnny]
Thus, follows. On Alice's side, Alice
goes to initiate a call.

1. The calling JS instantiates a PeerConnection and tells it is highly unlikely that users will ever perform the SAS
exchange.

Once uses 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 interested in Section 4.3.2.1. In principle having it authenticated via BrowserID.
2. The PeerConnection instantiates the BrowserID signer in an
invisible IFRAME. The IFRAME is of course possible to render a different UI element to indicate
that calls are using tagged with an unauthenticated set of keying material
(recall origin that the attacker can just present a slightly different name
so
indicates that it was generated by 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. Recommendations

[[ OPEN ISSUE: What are the best UI recommendations to make?
Proposal: take the text PeerConnection (this
prevents ordinary JS from [I-D.kaufman-rtcweb-security-ui]
Section 2]]

[[ OPEN ISSUE: Exactly what combination of media security primitives
should be specified and/or mandatory to implement? In particular,
should we allow DTLS-SRTP only, or both DTLS-SRTP and SDES. Should
we allow RTP for backward compatibility? ]]

5. Security Considerations

This entire document implementing it). The BrowserID signer
is about security.

6. Acknowledgements

7. References

7.1. Normative References

[RFC2119] Bradner, S., "Key words for use in RFCs provided with Alice's fingerprint. Note that the IFRAME here
does not render any UI. It is being used solely to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.

7.2. Informative References

[CORS] van Kesteren, A., "Cross-Origin Resource Sharing".

[I-D.abarth-origin]
Barth, A., "The Web Origin Concept",
draft-abarth-origin-09 (work in progress), November 2010.

[I-D.ietf-hybi-thewebsocketprotocol]
Fette, I. and A. Melnikov, "The WebSocket protocol",
draft-ietf-hybi-thewebsocketprotocol-15 (work in
progress), September 2011.

[I-D.kaufman-rtcweb-security-ui]
Kaufman, M., "Client Security User Interface Requirements
for RTCWEB", draft-kaufman-rtcweb-security-ui-00 (work allow the
browser to load the BrowserID signer in
progress), June 2011.

[RFC2818] Rescorla, E., "HTTP Over TLS", RFC 2818, May 2000.

[RFC3261] Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston,
A., Peterson, J., Sparks, R., Handley, M., isolation, especially
from the calling site.
3. The BrowserID signer contacts Alice's identity provider,
authenticating as Alice (likely via a cookie).
4. The identity provider returns a short-term certificate attesting
to Alice's identity and E.
Schooler, "SIP: Session Initiation Protocol", RFC 3261,
June 2002.

[RFC3552] Rescorla, E. her short-term public key.

5. The Browser-ID code signs the fingerprint and B. Korver, "Guidelines returns the signed
assertion + certificate to the PeerConnection. [Note: there are
well-understood Web mechanisms for Writing RFC
Text this that I am excluding here
for simplicity.]
6. The PeerConnection returns the signed information to the calling
JS code.
7. The signed assertion gets sent over the wire to Bob's browser
(via the signaling service) as part of the call setup.

Obviously, the format of the signed assertion varies depending on Security Considerations", BCP 72, RFC 3552,
July 2003.

[RFC3760] Gustafson, D., Just, M., and M. Nystrom, "Securely
Available Credentials (SACRED) - Credential Server
Framework", RFC 3760, April 2004.

[RFC4251] Ylonen, T.
what signaling style the WG ultimately adopts. However, for
concreteness, if something like ROAP were adopted, then the entire
message might look like:

{
"messageType":"OFFER",
"callerSessionId":"13456789ABCDEF",
"seq": 1
"sdp":"
v=0\n
o=- 2890844526 2890842807 IN IP4 192.0.2.1\n
s= \n
c=IN IP4 192.0.2.1\n
t=2873397496 2873404696\n
m=audio 49170 RTP/AVP 0\n
a=fingerprint: SHA-1 \
4A:AD:B9:B1:3F:82:18:3B:54:02:12:DF:3E:5D:49:6B:19:E5:7C:AB\n",
"identity":{
"identityType":"browserid",
"assertion": {
"digest":"<hash of fingerprint and session IDs>",
"audience": "[TBD]"
"valid-until": 1308859352261,
}, // signed using user's key
"certificate": {
"email": "rescorla@gmail.com",
"public-key": "<ekrs-public-key>",
"valid-until": 1308860561861,
} // certificate is signed by gmail.com
}
}

Note that we only expect to sign the fingerprint values and C. Lonvick, "The Secure Shell (SSH)
Protocol Architecture", RFC 4251, January 2006.

[RFC4347] Rescorla, E. the
session IDs, in order to allow the JS or calling service to modify
the rest of the SDP, while protecting the identity binding. [OPEN
ISSUE: should we sign seq too?]

[TODO: NEed to talk about Audience a bit.]
On Bob's side, he receives the signed assertion as part of the call
setup message and N. Modadugu, "Datagram Transport Layer
Security", RFC 4347, April 2006.

[RFC4568] Andreasen, F., Baugher, M., a similar procedure happens to verify it.

1. The calling JS instantiates a PeerConnection and D. Wing, "Session
Description Protocol (SDP) Security Descriptions for Media
Streams", RFC 4568, July 2006.

[RFC5245] Rosenberg, J., "Interactive Connectivity Establishment
(ICE): A Protocol for Network Address Translator (NAT)
Traversal for Offer/Answer Protocols", RFC 5245,
April 2010.

[RFC5479] Wing, D., Fries, S., Tschofenig, H., provides it the
relevant signaling information, including the signed assertion.
2. The PeerConnection instantiates a BrowserID verifier in an IFRAME
and F. Audet,
"Requirements provides it the signed assertion.
3. The BrowserID verifier contacts the identity provider to verify
the certificate and Analysis then uses the key to verify the signed
fingerprint.
4. Alice's verified identity is returned to the PeerConnection (it
already has the fingerprint).
5. At this point, Bob's browser can display a trusted UI indication
that Alice is on the other end of Media Security Management
Protocols", RFC 5479, April 2009.

[RFC5763] Fischl, J., Tschofenig, H., the call.

When Bob returns his answer, he follows the converse procedure, which
provides Alice with a signed assertion of Bob's identity and E. Rescorla, "Framework keying
material.

A.3.6.3. OAuth

While OAuth is not directly designed for Establishing user-to-user authentication,
with a Secure Real-time Transport Protocol
(SRTP) Security Context Using Datagram Transport Layer
Security (DTLS)", RFC 5763, May 2010.

[RFC6189] Zimmermann, P., Johnston, A., and J. Callas, "ZRTP: Media
Path Key Agreement little lateral thinking it can be made to serve. We use the
following mapping of OAuth concepts to RTC-Web concepts:

Table 1

The idea here is that when Alice wants to authenticate to Bob (i.e.,
for Unicast Secure RTP", RFC 6189,
April 2011.

[abarth-rtcweb]
Barth, A., "Prompting Bob to be aware that she is calling). In order to do this, she
allows Bob to see a resource on the identity provider that is bound
to the call, her identity, and her public key. Then Bob retrieves
the user is security failure", RTC-
Web Workshop.

[cranor-wolf]
Sunshine, J., Egelman, S., Almuhimedi, H., Atri, N., resource from the identity provider, thus verifying the binding
between Alice and
L. cranor, "Crying Wolf: An Empirical Study of SSL Warning
Effectiveness", Proceedings the call.

Alice IDP Bob
---------------------------------------------------------
Call-Id, Fingerprint ------->
<------------------- Auth Code
Auth Code ---------------------------------------------->
<----- Get Token + Auth Code
Token --------------------->
<------------- Get call-info
Call-Id, Fingerprint ------>

This is a modified version of a common OAuth flow, but omits the 18th USENIX Security
Symposium, 2009.

[farus-conversion]
Farrus, M., Erro, D., and J. Hernando, "Speaker
Recognition Robustness
redirects required to Voice Conversion".

[finer-grained]
Barth, A. and C. Jackson, "Beware of Finer-Grained
Origins", W2SP, 2008.

[huang-w2sp]
Huang, L-S., Chen, E., Barth, A., Rescorla, E., and C.
Jackson, "Talking have the client point the resource owner to Yourself for Fun and Profit", W2SP,
2011.

[kain-conversion]
Kain, A. the
IDP, which is acting as both the resource server and M. Macon, "Design the
authorization server, since Alice already has a handle to the IDP.

Above, we have referred to "Alice", but really what we mean is the
PeerConnection. Specifically, the PeerConnection will instantiate an
IFRAME with JS from the IDP and Evaluation of will use that IFRAME to communicate
with the IDP, authenticating with Alice's identity (e.g., cookie).
Similarly, Bob's PeerConnection instantiates an IFRAME to talk to the
IDP.

A.3.6.4. Generic Identity Support

I believe it's possible to build a Voice
Conversion Algorithm based on Spectral Envelope Mapping generic interface between the
PeerConnection and Residual Prediction", Proceedings of ICASSP, May
2001.

[whitten-johnny]
Whitten, A. any identity sub-module so that the PeerConnection
just gets pointed to the IDP (which the relying party either trusts
or not) and J. Tygar, "Why Johnny Can't Encrypt: A
Usability Evaluation of PGP 5.0", Proceedings of JS from the 8th
USENIX Security Symposium, 1999. IDP provides the concrete interfaces.
However, I need to work out the details, so I'm not specifying this
yet. If it works, the previous two sections will just be examples.

Author's Address

Eric Rescorla
RTFM, Inc.
2064 Edgewood Drive
Palo Alto, CA 94303
USA

Phone: +1 650 678 2350
Email: ekr@rtfm.com