[Search] [txt|xml|pdfized|bibtex] [Tracker] [Email] [Diff1] [Diff2] [Nits]
Versions: 00 01 02                                                      
Network Working Group                                          J. Hodges
Internet-Draft                                                    PayPal
Intended status: Standards Track                             Sep 8, 2011
Expires: March 11, 2012

       Web Security Framework: Problem Statement and Requirements


   Web-based malware and attacks are proliferating rapidly on the
   Internet.  New web security mechanisms are also rapidly growing in
   number, although in an incoherent fashion.  This document provides a
   brief overview of the present situation and the various seemingly
   piece-wise approaches being taken to mitigate the threats.  It then
   provides an overview of requirements as presently being expressed by
   the community in various online and face-to-face discussions.

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 11, 2012.

Copyright Notice

   Copyright (c) 2011 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents
   (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

Hodges                   Expires March 11, 2012                 [Page 1]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

   the Trust Legal Provisions and are provided without warranty as
   described in the Simplified BSD License.

Table of Contents

   1.  Introduction . . . . . . . . . . . . . . . . . . . . . . . . .  3
     1.1.  Where to Discuss This Draft  . . . . . . . . . . . . . . .  4
   2.  Document Conventions . . . . . . . . . . . . . . . . . . . . .  4
   3.  Overall Constraints  . . . . . . . . . . . . . . . . . . . . .  5
   4.  Overall Requirements . . . . . . . . . . . . . . . . . . . . .  6
   5.  Attacks and Threats to Address . . . . . . . . . . . . . . . .  7
     5.1.  Attacks  . . . . . . . . . . . . . . . . . . . . . . . . .  8
     5.2.  Threats  . . . . . . . . . . . . . . . . . . . . . . . . .  8
   6.  Use Cases  . . . . . . . . . . . . . . . . . . . . . . . . . .  9
   7.  Detailed Functional Requirements . . . . . . . . . . . . . . . 10
   8.  Extant Policies to Coalesce  . . . . . . . . . . . . . . . . . 18
   9.  Example Concrete Approaches  . . . . . . . . . . . . . . . . . 18
   10. Security Considerations  . . . . . . . . . . . . . . . . . . . 18
   11. References . . . . . . . . . . . . . . . . . . . . . . . . . . 18
   12. Informative References . . . . . . . . . . . . . . . . . . . . 22
   Author's Address . . . . . . . . . . . . . . . . . . . . . . . . . 22

Hodges                   Expires March 11, 2012                 [Page 2]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

1.  Introduction

   Over the past few years, we have seen a proliferation of AJAX-based
   web applications (AJAX being shorthand for asynchronous JavaScript
   and XML), as well as Rich Internet Applications (RIAs), based on so-
   called Web 2.0 technologies.  These applications bring both luscious
   eye-candy and convenient functionality--e.g. social networking--to
   their users, making them quite compelling.  At the same time, we are
   seeing an increase in attacks against these applications and their
   underlying technologies [1].  The latter include (but aren't limited
   to) Cross-Site-Request Forgery (CSRF) -based attacks [2], content-
   sniffing cross-site-scripting (XSS) attacks [3], attacks against
   browsers supporting anti-XSS policies [4], clickjacking attacks [5],
   malvertising attacks [6], as well as man-in-the-middle (MITM) attacks
   against "secure" (e.g.  Transport Layer Security (TLS/SSL)-based [7])
   web sites along with distribution of the tools to carry out such
   attacks (e.g. sslstrip) [8].

   During the same time period we have also witnessed the introduction
   of new web security indicators, techniques, and policy communication
   mechanisms sprinkled throughout the various layers of the Web and
   HTTP.  We have a new cookie security flag called HTTPOnly [9].  We
   have the anti-clickjacking X-Frame-Options HTTP header [10], the
   Strict-Transport-Security HTTP header [11], anti-CSRF headers (e.g.
   Origin) [12], an anti-sniffing header (X-Content-Type-Options:
   nosniff) [13], various approaches to content restrictions [14] [15]
   and notably Mozilla Content Security Policy (CSP; conveyed via a HTTP
   header) [16], the W3C's Cross-Origin Resource Sharing (CORS; also
   conveyed via a HTTP header) [17], as well as RIA security controls
   such as the crossdomain.xml file used to express a site's Adobe Flash
   security policy [18].  There's also the Application Boundaries
   Enforcer (ABE) [19], included as a part of NoScript [20], a popular
   Mozilla Firefox security extension.  Sites can express their ABE
   rule-set at a well-known web address for downloading by individual
   clients [21], similarly to Flash's crossdomain.xml.  Amidst this
   haphazard collage of new security mechanisms at least one browser
   vendor has even devised a new HTTP header that disables one of their
   newly created security features: witness the X-XSS-Protection header
   that disables the new anti-XSS features [22] in Microsoft's Internet
   Explorer 8 (IE8).

   Additionally, there are various proposals aimed at addressing other
   facets of inherent web vulnerabilities, for example: JavaScript
   postMessage-based mashup communications [23], hypertext isolation
   techniques [24], and service security policies advertised via the
   Domain Name System (DNS) [25].  Going even further, there are efforts
   to redesign web browser architectures [26], of which Google Chrome
   and IE8 are deployed examples.  An even more radical approach is

Hodges                   Expires March 11, 2012                 [Page 3]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

   exhibited in the Gazelle Web Browser [27], which features a browser
   kernel embodied in a multi-principal OS construction providing cross-
   principal protection and fair sharing of all system resources.

   Not to be overlooked is the fact that even though there is a plethora
   of "standard" browser security features--e.g. the Same Origin Policy
   (SOP), network-related restrictions, rules for third-party cookies,
   content-handling mechanisms, etc. [28]--they are not implemented
   uniformly in today's various popular browsers and RIA frameworks
   [29].  This makes life even harder for web site administrators in
   that allowances must be made in site security posture and approaches
   in consideration of which browser a user may be wielding at any
   particular time.

   Although industry and researchers collectively are aware of all the
   above issues, we observe that the responses to date have been issue-
   specific and uncoordinated.  What we are ending up with looks perhaps
   similar to Frankenstein's monster [30]--a design with noble intents
   but whose final execution is an almost-random amalgamation of parts
   that do not work well together.  It can even cause destruction on its
   own [31].

   Thus, the goal of this document is to define the requirements for a
   common framework expressing security constraints on HTTP
   interactions.  Functionally, this framework should be general enough
   that it can be used to unite the various individual solutions above,
   and specific enough that it can address vulnerabilities not addressed
   by current solutions, and guide the development of future mechanisms.

   Overall, such a framework would provide web site administrators the
   tools for managing, in a least privilege [33] manner, the overall
   security characteristics of their web site/applications when realized
   in the context of user agents.

1.1.  Where to Discuss This Draft

   Please disscuss this draft on the websec@ietf.org mailing list

2.  Document Conventions

   Note:  ..is a note to the reader.  These are points that should be
          expressly kept in mind and/or considered.

   [[XXXn: Some of the more major known issues are marked like this
   (where "n" in "XXXn" is a number). --JeffH]]

Hodges                   Expires March 11, 2012                 [Page 4]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

   [[TODOn: Things to fix (where "n" in "TODOn" is a number). --JeffH]]

   We will also be making use of the WebSec WG issue tracker, so use of
   the above two issue & TODO marks will evolve accordingly.

3.  Overall Constraints

   Regardless of the overall approaches chosen for conveying site
   security policies, we believe that to be deployed at Internet-scale,
   and to be as widely usable as possible for both novice and expert
   alike, the overall solution approach will need to address these three
   points of tension:


         There has been much debate during the discussion of some policy
         mechanisms (e.g.  CSP) as to how fine-grained such mechanisms
         should be.  The argument against fine-grained mechanisms is
         that site administrators will cause themselves pain by
         instantiating policies that do not yield the intended results.
         E.g. simply copying the expressed policies of a similar site.
         The claim is that this would occur for various reasons stemming
         from the mechanisms' complexity [34].


         Not infrequently, the complexity of underlying facilities, e.g.
         in server software, is not well-packaged and thus
         administrators are obliged to learn more about the intricacies
         of these systems than otherwise might be necessary.  This is
         sometimes used as an argument for "dumbing down" the
         capabilities of policy expression mechanisms [34].


         Research shows that when security warnings are displayed, users
         are often given too much information as well as being allowed
         to relatively easily bypass the warnings and continue with
         their potentially compromising activity [35] [36] [37] [38]
         [39].  Thus users have become trained to "click through"
         security notifications "in order to get work done", though not
         infrequently rendering themselves insecure and perhaps
         compromised [40].

   In the next section we discuss various high-level requirements
   derived with the guidance of the latter tension points.

Hodges                   Expires March 11, 2012                 [Page 5]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

4.  Overall Requirements

   1.  Policy conveyance:


             We believe that a regime based on HTTP header(s) is
             appropriate.  However we must devise a generalized,
             extensible HTTP security header(s) such that the on-going
             "bloat" of the number of disjoint HTTP security headers is
             mitigated and there is a documented framework that we can
             leverage as new approaches and/or threats emerge.

             Note:  The distinction between in-band and out-of-band
                    signaling is difficult to characterize because some
                    seemingly out-of-band mechanisms rely on the same
                    protocols (HTTP/HTTPS) and infrastructure
                    (transparent proxy servers) as the protocols they
                    ostensibly protect.

             It may be reasonable to devise a small set of headers to
             convey different classes of policies, e.g. web application
             content policies versus web application network
             capabilities policies.


             This policy communication mechanism must be secure and
             should have two facets, one for communicating securely out-
             of-band of the HTTP protocol to allow for secure client
             policy store bootstrapping. potential approaches are
             factory-installed web browser configuration, site security
             policy download a la Flash's crossdomain.xml and Maone's
             ABE for Web Authors [21], and DNS-based policy
             advertisement leveraging the security of DNS Security
             (DNSSEC) [32].

   2.  Granularity:

          In terms of granularity, vast arrays of stand-alone blog,
          wiki, hosted web account, and other "simple" web sites could
          ostensibly benefit from relatively simple, pre-determined
          policies.  However, complex sites--e.g. payment, ecommerce,
          software-as-a-service, mashup sites, etc.--often differ in
          various ways, as well as being inherently complex
          implementation-wise.  One-size-fits-all policies will
          generally not work well for them.  Thus, we believe that to be
          effective for a broad array of web site and application types,

Hodges                   Expires March 11, 2012                 [Page 6]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

          the policy expression mechanism must fundamentally facilitate
          fine-grained control.  For example, CSP offers such control.
          In order to address the less complex needs of the more simple
          classes of web sites, the policy expression mechanism could
          have a "macro"-like feature enabling "canned policy profiles".
          Or, the configuration facilities of various components of the
          web infrastructure can be enhanced to provide an appropriately
          simple veneer over the complexity.

   3.  Configurability:

          With respect to configurability, development effort should be
          applied to creating easy-to-use administrative interfaces
          addressing the simple cases, like those mentioned above, while
          providing advanced administrators the tools to craft and
          manage fine-grained multi-faceted policies.  Thus more casual
          or novice administrators can be aided in readily choosing, or
          be provided with, safe default policies while other classes of
          sites have the tools to craft the detailed policies they
          require.  Examples of such an approach are Microsoft's
          "Packaging Wizard" [41] that easily auto-generates a quite
          complicated service deployment descriptor on behalf of less
          experienced administrators, and Firefox's simple Preferences
          dialog [42] as compared to its detailed about:config
          configuration editor page [43].  In both cases, simple usage
          by inexperienced users is anticipated and provided for on one
          hand, while complex tuning of the myriad underlying
          preferences is provided for on the other.

   4.  Usability:

          Much has been learned over the last few years about what does
          and does not work with respect to security indicators in web
          browsers and web pages, as noted above, these lessons should
          be applied to the security indicators rendered by new proposed
          security mechanisms.  We believe that in cases of user agents
          venturing into insecure situations, their response should be
          to fail the connections by default without user recourse,
          rather than displaying warnings along with bypass mechanisms,
          as is current practice.  For example, the Strict Transport
          Security specification stipulates the former hard-fail

5.  Attacks and Threats to Address

   This section enumerates various attacks and threats that ought to be
   mitigated by a web security policy framework.  In terms of defining

Hodges                   Expires March 11, 2012                 [Page 7]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

   threats in contrast to attacks, Lucas supplied this:

   <"Re: More on XSS mitigation (was Re: XSS mitigation in browsers)"
   (Lucas Adamski).  http://lists.w3.org/Archives/Public/

      "...  There's a fundamental question about whether we should be
      looking at these problems from an attack vs threat standpoint.  An
      attack is XSS [or CSRF, or Response Splitting, etc.].  A threat is
      that an attacker could compromise a site via content injection to
      trick the user to disclosing confidential information (by
      injecting a plugin or CSS to steal data or fool the user into
      sending their password to the attacker's site). ..."

5.1.  Attacks

   The below attacks should be possible to mitigate via a web
   application security framework (see [44] for a definition and
   taxonomy of attacks):

   1.  cross-site-scripting (XSS) [2] [44]

   2.  Man-in-the-middle (MITM) attacks against "secure" (e.g.
       Transport Layer Security (TLS/SSL)-based [7] [8] [44]) web sites.
       For example, be able to subsume the HSTS header [11].

   3.  Cross-Site-Request Forgery (CSRF) [3] [44] (?)

   4.  Response Splitting [44]

   5.  more (ie eg from [44] ?) ?

5.2.  Threats

   Via the attacks above, an attacker can..

   1.  Obtain a victim's confidential web application credentials (e.g.
       cookie theft), and use the credentials to impersonate the victim
       and enter into transactions, often with the aim of monetizing the
       transaction results to the attacker's benefit.

   2.  Insert themselves as a Man-in-the-Middle (MITM) between victim
       and various services, thus is able to instigate, control,
       intercept, and attempt to monetize various transactions and
       interactions with web applications, to the benefit of the

Hodges                   Expires March 11, 2012                 [Page 8]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

   3.  Enumerate various user agent information stores, e.g. browser
       history, facilitating views of the otherwise confidential habits
       of the victim.  This information could possibly be used in
       various offline attacks against the victim directly, e.g.
       blackmail, denial of services, law enforcement actions, etc.

   4.  Use gathered information and credentials to construct and present
       a falsified persona of the victim (e.g. for character

   There is a risk of exfiltration of otherwise confidential victim
   information with all the threats listed above.

6.  Use Cases

   This section outlines various concrete use cases.  Where applicable,
   source email messages are cited.

   1.  I'm a web application site administrator.  My web app includes
       static user-supplied content (e.g. submitted from user agents via
       HTML FORM + HTTP POST), but either my developers don't properly
       sanitize user-supplied content in all cases or/and content
       injection vulnerabilities exist or materialize (for various

       This leaves my web app vulnerable to cross-site scripting.  I
       wish I could set overall web app-wide policies that prevent user-
       supplied content from injecting malicious content (e.g.
       JavaScript) into my web app.

   2.  I'm a web application site administrator.  My web application is
       intended, and configured, to be uniformly served over HTTPS, but
       my developers mistakenly keep including content via insecure
       channels (e.g. via HTTP-only; resulting in so-called "mixed

       I wish I could set a policy for my web app that prevents user
       agents from loading content insecurely even if my web app is
       otherwise telling them to do so.

   3.  I'm a web application site administrator.  My site has a policy
       that we can only include content from certain trusted providers
       (e.g., our CDN, Amazon S3), but my developers keep adding
       dependencies on origins I don't trust.  I wish I could set a
       policy for my site that prevents my web app from accidentally
       loading resources outside my whitelist.

Hodges                   Expires March 11, 2012                 [Page 9]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

   4.  I'm a web application site administrator.  I want to ensure that
       my web app is never framed by other web apps.

   5.  I'm a developer of a web application which will be included (i.e.
       framed) by third parties within their own web apps.  I would like
       to ensure that my web app directs user agents to only load
       resources from URIs I expect it to (possibly even down to
       specific URI paths), without affecting the containing web app or
       any other web apps it also includes.

   6.  I'm a web application site administrator.  My web app frames
       other web apps whose behavior, properties, and policies are not
       100% known or predictable.

       I need to be able to apply policies that both protect my web app
       from potential vulnerabilities or attacks introduced by the
       framed web apps, and that work to ensure that the desired
       interactions between my web app and the framed apps are securely

7.  Detailed Functional Requirements

   Many of the below functional requirements are extracted from a recent
   discussion on the [public-web-security] list.  Particular messages
   are cited inline and appropriate quotes extracted and reproduced
   here.  Inline citations are provided for definitions of various

   1.   Policy expression syntax:

        *  Declarative.

              <"declarative languages".  http://www.encyclopedia.com/

        *  Extensible.


        <"Re: XSS mitigation in browsers" (Lucas Adamski).  http://

           "On a conceptual level, I am not really a believer in the
           current proliferation of orthogonal atomic mechanisms
           intended to solve very specific problems.  Security is a

Hodges                   Expires March 11, 2012                [Page 10]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

           holistic discipline, and so I'm a big supporter of investing
           in an extensible declarative security policy mechanism that
           could evolve as the web and the threats that it faces do.
           Web developers have a hard enough time with security already
           without being expected to master a potentially large number
           of different security mechanisms, each with their own unique
           threat model, implementation and syntax.  Not to mention
           trying to figure out how they're expected to interact with
           each other... how to manage the gaps and intersections
           between the models."

        <"Re: Scope and complexity (was Re: More on XSS mitigation)"
        (Adam Barth).  http://lists.w3.org/Archives/Public/

           "I guess I wish we had an extensibility model more like HTML
           where we could grow the security protections over time.  For
           example, we can probably agree that both <canvas> and <video>
           are great additions to HTML that might not have made sense
           when folks were designing HTML 1.0.

           As long as you're not relying on the security policy as a
           first line of defense, the extensibility story for security
           policies is even better than it is with HTML tags.  With an
           HTML tag, you need a fall-back for browsers that don't
           support the tag, whereas with a security policy, you'll
           always have your first line of defense.

           Ideally, we could come up with a policy mechanism that let us
           nail XSS today and that fostered innovation in security for
           years to come.  In the short term, you could view the
           existing CSP features (e.g., clickjacking protection) as the
           first wave of innovation.  If those pieces are popular, then
           it should be easy for other folks to adopt them."

   2.   Tooling:

        *  We will need tools to (idealy) analyze a web application and
           generate a starting point security policy.

        <"Re: More on XSS mitigation" (John Wilander).  http://

           "*Developers Will Want a Policy Generator* A key issue for
           in-the-field success of CSP is how to write, generate and
           maintain the policies.  Just look at the epic failure of Java
           security policies.  The Java policy framework was designed

Hodges                   Expires March 11, 2012                [Page 11]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

           for static releases shipped on CDs, not for moving code,
           added frameworks, new framework versions etc.  The world of
           web apps is so dynamic I'm still amazed.  If anything, for
           instance messy security policies, gets in the way of daily
           releases it's a no go.  At least until there's an exploit.
           Where am I going with this?  Well, we should implement a PoC
           *policy generator* and run it on some fairly large websites
           before we nail the standard.  There will be subtleties found
           which we can address and we can bring the PoC to production
           level while the standard is being finalized and shipped in
           browsers.  Then we release the policy generator along with
           policy enforcement -- success! "

   3.   Performance:

        *  Minimizing performance impact is a first-order concern.

        <"Re: More on XSS mitigation" (John Wilander).  http://

           "*We Mustn't Spoil Performance* Web developers (and browser
           developers) are so hung up on performance that we really need
           to look at what they're up to and make sure we don't spoil
           things.  Especially load performance now that it's part of
           Google's rating."

   4.   Granularity:

        *  For example, discriminate between:

           +  "inline" script in <head> versus <body>, or not.

           +  "inline" script and "src=" loaded script.

           +  Classes of "content", e.g. scriptable content, passive
              multimedia, nested documents, etc.

        <"Proposal to move the debate forward" (Daniel Veditz).  http://

           "We oscillated several times between lumpy and granular.
           Fewer classes (simpler) is always more attractive, easier to
           explain and understand.  The danger is that future features
           then end up being added to the existing lumps, possibly
           enabling things that the site isn't aware they need to now
           filter.  It's a constant problem as we expand the

Hodges                   Expires March 11, 2012                [Page 12]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

           capabilities of browsers -- sites that used to be perfectly
           secure are suddenly hackable because all the new browsers
           added feature-X."

   5.   Notifications and reporting:

        *  Convey to the user agent an identifier (e.g. a URI) denoting
           where to send policy violation reports.  Could also specify a
           DOM event to be dedicated for this purpose.

        *  An ability to specify that a origin's policies are to be
           enforced in a "report only" mode will be useful for debugging
           policies as well as site-policy interactions.  E.g. for
           answering the question: "does my policy 'break' my site?".

        <"[Content Security Policy] Proposal to move the debate forward"
        (Brandon Sterne).  http://lists.w3.org/Archives/Public/

   3. Violation Reporting
      a. report-uri: URI to which a report will be sent upon policy
      b. SecurityViolation event: DOM event fired upon policy violations

   6.   Facilitating Separation of Duties:

        *  Specifically, allowing for Web Site operations/deployment
           personnel to apply site policy, rather then having it being
           encoded in the site implementation code by side developers/

        <"RE: Content Security Policy and iframe@sandbox" (Andrew
        Steingruebl).  http://lists.w3.org/Archives/Public/

           "... 2.  SiteC is also totally in control of all HTTP headers
           it emits.  It could just as easily indicate policy choices
           for all frames via CSP.  It could advertise a blanket policy
           (No JS, No ActiveX).  Advertising a page-specific, or frame/
           target specific policy is substantially more difficult and
           probably unwieldy.  But, depending on how SiteC is
           configured, setting a global site policy via headers offers a
           potential separation of duties that #1 does not, it allows
           website admin to specific things that each web developer

Hodges                   Expires March 11, 2012                [Page 13]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

           might not be able to. ..."

   7.   Hierarchical Policy Application:

        *  The notion that policy emitted by the application's source
           origin is able to constrain behavior and policies of
           contained origins.

        <"RE: Content Security Policy and iframe@sandbox" (Andrew
        Steingruebl).  http://lists.w3.org/Archives/Public/

           "...  I could imagine a tweak to CSP wherein CSP would
           control all contents hierarchically.  I already spoke to
           Brandon about it, but it was just a quick brainstorm.

           You could imagine revoking permissions in the frame hierarchy
           and not granting them back.  This does start to get awfully
           ugly, but just as CSP controls loading policy for itself, it
           could also control loading policy for children, ..."

   8.   Framing Policy Hierarchy, cross-origin, granularity:

           <"Re: Content Security Policy and iframe@sandbox") (Andy
           Steingruebl, Adam Barth) http://lists.w3.org/Archives/Public/

On Sat, Feb 12, 2011 at 9:01 PM, Steingruebl, Andy
                <asteingruebl@paypal-inc.com> wrote:
>> -----Original Message-----
>> From: Adam Barth [mailto:w3c@adambarth.com]
>> That all sounds very abstract. If you have some concrete examples,
>> that might be more productive to discuss. When enforcing policy
>> supplied by one origin on another origin, we need to be careful to
>> consider the case where the policy providing origin is the attacker
>> and the origin on which the policy is being enforced is the victim.
> SiteA  wants to make sure it cannot ever be framed.  It deploys
X-Frame-Options headers and framebusting JS, and maybe even CSP
> SiteB wants to make sure it never loads data from anything other than
SiteB (no non-origin loads).  It outputs CSP headers to this effect
> SiteC wants to make sure that any content it frames cannot run ActiveX

Hodges                   Expires March 11, 2012                [Page 14]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

controls, nor do a 401 authentication.  It can't really do this with
current iframe sandboxing, but pretend it could...
> SiteC wants to control the behavior of children that it frames.  It
needs to advertise this policy to a web browser.  It has two choices:
> 1. It can do it inline in the HTML it outputs with extra attributes of
the iframe it creates.  SiteC is in complete control of the HTML that
creates the iframe.  I can impose any policy via sandbox attributes.
Currently for example, it can disable JS in the frame.  If it frames
SiteA, SiteA's framebusting JS will never run, but the browser will
respect its X-Frame-Options headers.
> 2. SiteC is also totally in control of all HTTP headers it emits.  It
could just as easily indicate policy choices for all frames via CSP.  It
could advertise a blanket policy (No JS, No ActiveX).  Advertising a
page-specific, or frame/target specific policy is substantially more
difficult and probably unwieldy. But, depending on how SiteC is
configured, setting a global site policy via headers offers a potential
separation of duties that #1 does not, it allows website admin to
specific things that each web developer might not be able to.
> 3. Because all of Site A,B,C are in different origins, they don't
really have to worry about polluting other origins, but they do have to
worry about problematic behavior such as top-nav, 401-auth popups, etc.
Parents need to constrain certain behavior of things they embed,
according to certain rules of whether the child allows itself to be
> I totally get how existing iframe sandboxing that turns off JS is
problematic for sites [due to] older browsers that don't support
X-Frame-Options.  We already have a complicated interaction between
these multiple security controls.
> Can you give me an example of why my #1/#2 are actually that
different?  Whether we control behavior with headers of inline content,
each site is totally responsible for what it emits, and can already
control in some interesting ways the behavior of content it

In this example, the trade-off for Site C seems to boil down to the
granularity of the policy.  Using attributes on a frame is more
fine-grained because Site C can make these decisions on an
iframe-by-iframe basis whereas using a document-wide policy is more

Of course, there's a trade-off between different granularities.  On
the one hand, fine-grained gives the site more control over how

Hodges                   Expires March 11, 2012                [Page 15]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

different iframes behavior.  On the other hand, it's much easier to
audit and understand the effects of a coarse-grained policy.


   9.   Policy Delivery:

        *  The web application policy must be communicated by the web
           application to the user agent.  There are various approaches
           and they have tradeoffs between security, audience, and

        <"[Content Security Policy] Proposal to move the debate forward"
        (Brandon Sterne).  http://lists.w3.org/Archives/Public/

   6. Policy delivery
      a. HTTP header
      b. <meta> (or <link>) tag, to be superseded by header if present
      c. policy-uri: a URI from which the policy will be fetched; can be
         specified in either header or tag

        <"Re: [Content Security Policy] Proposal to move the debate
        forward" (gaz Heyes).  http://lists.w3.org/Archives/Public/

           a) Policy shouldn't be defined in a http header it's too
           messy and what happens when there's a mistake?

           b) As discussed on the list there is no need to have a
           separate method as it can be generated by an attacker.  If a
           policy doesn't exist then an attacker can now DOS the web
           site via meta.

           c) We have a winner, a http header specifying a link to the
           policy file is the way to go IMO, my only problem with it is
           devs implementing it.  Yes facebook would and probably
           twitter would but Dave's tea shop wouldn't pay enough money
           to hire a web dev who knew how to implement a custom http
           header yet they would know how to validate HTML.  So the
           question is are we bothered about little sites that are
           likely to have nice tea and XSS holes?  If so I suggest

Hodges                   Expires March 11, 2012                [Page 16]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

           updating the HTML W3C validator to require a security policy
           to pass validation if not I suggest a policy file delivered
           by http header.

   10.  Policy Conflict Resolution:


        <"RE: Content Security Policy and iframe@sandbox" (Andrew
        Steingruebl).  http://lists.w3.org/Archives/Public/

 > -----Original Message-----
 > From: public-web-security-request@w3.org [mailto:public-web-security-
 > request@w3.org] On Behalf Of Adam Barth
 > @sandbox and CSP are very different.  The primary difference is who
 > choses the policy.  In the case of @sandbox, the embedder chooses
 > the policy. In CSP, the provider of the resource chooses the policy.

 While this is true today, I could imagine a tweak to CSP wherein CSP
 would control all contents hierarchically.  I already spoke to Brandon
 about it, but it was just a quick brainstorm.

 You could imagine revoking permissions in the frame hierarchy and not
 granting them back.  This does start to get awfully ugly, but just as
 CSP controls loading policy for itself, it could also control loading
 policy for children, right?

 Fundamentally, since the existing security model doesn't really provide
 for strict separation of parent/child (popups, 401's, top-nav) CSP and
 iframe sandbox both try to control the behavior of resources we pull
 from other parties.

 Do we think that these are both special cases of a general security
 policy (my intuition says yes) or that they have some quite orthogonal
 types of security controls that cannot be mixed into a single policy

 One clear problem that comes to mind is that there are policies that
 come from the "child" such as X-Frame-Options that must break the
 ordinary parent/child relationship from a precedence standpoint.

Hodges                   Expires March 11, 2012                [Page 17]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

8.  Extant Policies to Coalesce

   Presently, this section lists a grab-bag of individually-expressed
   web app security policies which a more general substrate could
   ostensibly encompass (in order to, for example, reduce "header bloat"
   and bytes-on-the-wire issues), as well as reduce functional policy






      CSP frame-ancestors


9.  Example Concrete Approaches

   An overall, broad approach (from [0]):

      As for an overall policy mechanism, we observe that leveraging a
      combination of CSP [16] and ABE [19], or their employment in
      tandem, as a starting point for a multi-vendor approach may be
      reasonable.  For a near-term policy delivery mechanism, we
      advocate use of both HTTP headers and a policy file at a well-
      known location.  Leveraging DNSSEC is attractive in the
      intermediate term, i.e. as it becomes more widely deployed.

10.  Security Considerations

   Security considerations go here.

11.  References

   [[TODO1: re-code refs into xml and place in proper refs section.

   [0] J. Hodges, A. Steingruebl, "The Need for Coherent Web Security
   Policy Framework(s)", Web 2.0 Security & Privacy, Oakland CA, 20 May

Hodges                   Expires March 11, 2012                [Page 18]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

   2010. http://w2spconf.com/2010/papers/p11.pdf

   [1] Breach Security, "THE WEB HACKING INCIDENTS DATABASE 2009," Aug.
   2009. http://www.breach.com/resources/whitepapers/downloads/

   [2] R. Auger, The Cross-Site Request Forgery (CSRF/XSRF) FAQ, 2007.

   [3] A. Barth, J. Caballero, and D. Song, "Secure Content Sniffing for
   Web Browsers--or How to Stop Papers from Reviewing Themselves,"
   Proceedings of the 30th IEEE Symposium on Security & Privacy,
   Oakland, CA: 2009.

   [4] D. Goodin, "Major IE8 flaw makes 'safe' sites unsafe -
   Microsoft's XSS buster busted," The Register, Nov. 2009. http://

   [5] J. Grossman, "Clickjacking: Web pages can see and hear you," Oct.
   2008. http://jeremiahgrossman.blogspot.com/2008/10/

   [6] W. Salusky, Malvertising, 2007.

   [7] T. Dierks and E. Rescorla, "The Transport Layer Security (TLS)
   Protocol Version 1.2," RFC5246, Internet Engineering Task Force, Aug.
   2008. http://www.ietf.org/rfc/rfc5246.txt

   [8] M. Marlinspike, SSLSTRIP, 2009.

   [9] Scope of HTTPOnly Cookies.

   [10] E. Lawrence, IE8 Security Part VII: ClickJacking Defenses, 2009.

   [11] J. Hodges, C. Jackson, and A. Barth, "Strict Transport
   Security," Work-in-progress, Internet-Draft, Jul. 2010.

   [12] A. Barth, C. Jackson, and I. Hickson, "The Web Origin Concept,"
   Internet-Draft, work in progress, Internet Engineering Task Force,
   2009. http://tools.ietf.org/html/draft-abarth-origin

   [13] E. Lawrence, IE8 Security Part VI: Beta 2 Update, 2008. http://

Hodges                   Expires March 11, 2012                [Page 19]

Internet-Draft            WebSec Framework Reqs                 Sep 2011


   [14] G. Markham, Content restrictions, 2007.

   [15] T. Jim, N. Swamy, and M. Hicks, "BEEP: Browser-Enforced Embedded
   Policies," Proceedings of the 16th International World Wide Web
   Conference, Banff, Alberta, Canada, 2007.

   [16] B. Sterne, "Content Security Policy (CSP)," 2011. https://

   [17] A.V. Kesteren, "Cross-Origin Resource Sharing (CORS)," Mar.
   2009. http://www.w3.org/TR/2009/WD-cors-20090317/

   [18] Adobe Systems, "Cross-domain policy file specification." http://

   [19] G. Maone, ABE - Application Boundaries Enforcer, 2009.

   [20] G. Maone, NoScript. http://noscript.net/

   [21] G. Maone, ABE for Web Authors, 2009.

   [22] Microsoft, "Event 1046 - Cross-Site Scripting Filter," MSDN
   Library, undated.

   [23] A. Barth, C. Jackson, and W. Li, "Attacks on JavaScript Mashup
   Communication," Proceedings of the Web 2.0 Security and Privacy
   Workshop, 2009.

   [24] M. Ter Louw, P. Bisht, and V. Venkatakrishnan, "Analysis of
   Hypertext Isolation Techniques for XSS Prevention," Proceedings of
   the Web 2.0 Security and Privacy Workshop, 2008 .

   [25] A. Ozment, S.E. Schechter, and R. Dhamija, "Web Sites Should Not
   Need to Rely on Users to Secure Communications," W3C Workshop on
   Transparency and Usability of Web Authentication, 2006.

   [26] C. Reis, A. Barth, and C. Pizano, "Browser Security: Lessons
   from Google Chrome," ACM Queue, 2009, pp. 1-8.

Hodges                   Expires March 11, 2012                [Page 20]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

   [27] H.J. Wang, C. Grier, A. Moshchuk, S.T. King, P. Choudhury, and
   H. Venter, "The Multi-Principal OS Construction of the Gazelle Web
   Browser," USENIX Security Symposium, 2009.

   [28] M. Zalewski, Browser Security Handbook.

   [29] A. Stamos, D. Thiel, and J. Osborne, Living in the RIA World:
   Blurring the Line between Web and Desktop Security, BlackHat
   presentation, iSecPartners, 2008.

   [30] Mary Shelley, "Frankenstein, or The Modern Prometheus," ca.
   1831. http://en.wikipedia.org/wiki/Frankenstein%27s_monster

   [31] D. Goodin, "cPanel, Netgear and Linksys susceptible to nasty
   attack - Unholy Trinity," The Register, 2009.

   [32] R. Arends, R. Austein, M. Larson, D. Massey, and S. Rose, "DNS
   security introduction and requirements," RFC4033, Internet
   Engineering Task Force, Mar. 2005.

   [33] J.H. Saltzer and M.D. Schroeder, "The Protection of Information
   in Computer Systems," Communications of the ACM, vol. 17, Jul. 1974.

   [34] I. Hickson and many others, "Comments on the Content Security
   Policy specification," discussion on mozilla.dev.security newsgroup.

   [35] S. Egelman, L.F. Cranor, and J. Hong, "You've Been Warned: An
   Empirical Study of the Effectiveness of Web Browser Phishing
   Warnings," CHI 2008, April 5 - 10, 2008, Florence, Italy, 2008.

   [36] S.E. Schechter, R. Dhamija, A. Ozment, and I. Fischer, "The
   Emperor's New Security Indicators," Proceedings of the 2007 IEEE
   Symposium on Security and Privacy.

   [37] R. Dhamija and J.D. Tygar, "The Battle Against Phishing: Dynamic
   Security Skins," Proceedings of the 2005 Symposium on Usable Privacy
   and Security (SOUPS).

   [38] J. Sobey, T. Whalen, R. Biddle, P.V. Oorschot, and A.S. Patrick,
   Browser Interfaces and Extended Validation SSL Certificates: An

Hodges                   Expires March 11, 2012                [Page 21]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

   Empirical Study, Ottawa, Canada: School of Computer Science, Carleton
   University, 2009.

   [39] J. Sunshine, S. Egelman, H. Almuhimedi, N. Atri, and L.F.
   Cranor, "Crying Wolf: An Empirical Study of SSL Warning
   Effectiveness," USENIX Security Symposium, 2009.

   [40] C. Jackson and A. Barth, "ForceHTTPS: Protecting High-Security
   Web Sites from Network Attacks," Proceedings of the 17th
   International World Wide Web Conference (WWW), 2008.

   [41] Microsoft, "Packaging Wizard."

   [42] Mozilla, "Options window."

   [43] S. Yegulalp, "Hacking Firefox: The secrets of about:config,"
   ComputerWorld, May. 2007. http://www.computerworld.com/s/article/

   [44] Web Application Security Consortium, "The WASC Threat
   Classification v2.0," 2009.

12.  Informative References

   [WebSec]   "Web HTTP Application Security Minus Authentication and

              "public-web-security@w3.org: Improving standards and
              implementations to advance the security of the Web.",

Hodges                   Expires March 11, 2012                [Page 22]

Internet-Draft            WebSec Framework Reqs                 Sep 2011

Author's Address

   Jeff Hodges
   2211 North First Street
   San Jose, California  95131

   Email: Jeff.Hodges@PayPal.com

Hodges                   Expires March 11, 2012                [Page 23]