Content Security Policy

Editor’s Draft,

This version:
https://w3c.github.io/webappsec/specs/content-security-policy/
Latest version:
http://www.w3.org/TR/CSP2/
Previous Versions:
http://www.w3.org/TR/2014/WD-CSP2-20140703/
http://www.w3.org/TR/2014/WD-CSP11-20140211/
http://www.w3.org/TR/2012/CR-CSP-20121115/
Feedback:
[email protected] with subject line “[CSP3] … message topic …” (archives)
Issue Tracking:
GitHub
Editors:
(Google Inc.)
(Mozilla Corporation)
Former Editors:
(Google Inc.)
(formerly of Mozilla Corporation)

Abstract

This document defines a policy language used to declare a set of content restrictions for a web resource, and a mechanism for transmitting the policy from a server to a client where the policy is enforced.

Status of this document

This is a public copy of the editors’ draft. It is provided for discussion only and may change at any moment. Its publication here does not imply endorsement of its contents by W3C. Don’t cite this document other than as work in progress.

Changes to this document may be tracked at https://github.com/w3c/webappsec.

The (archived) public mailing list [email protected] (see instructions) is preferred for discussion of this specification. When sending e-mail, please put the text “CSP3” in the subject, preferably like this: “[CSP3] …summary of comment…

This document was produced by the Web Application Security Working Group.

This document was produced by a group operating under the 5 February 2004 W3C Patent Policy. W3C maintains a public list of any patent disclosures made in connection with the deliverables of the group; that page also includes instructions for disclosing a patent. An individual who has actual knowledge of a patent which the individual believes contains Essential Claim(s) must disclose the information in accordance with section 6 of the W3C Patent Policy.

This document is governed by the 1 August 2014 W3C Process Document.

The following features are at-risk, and may be dropped during the CR period:

“At-risk” is a W3C Process term-of-art, and does not necessarily imply that the feature is in danger of being dropped or delayed. It means that the WG believes the feature may have difficulting being interoperably implemented in a timely manner, and marking it as such allows the WG to drop the feature if necessary when transitioning to the Proposed Rec stage, without having to publish a new Candidate Rec without the feature first.

Table of Contents

1. Introduction

This section is not normative.

This document defines Content Security Policy, a mechanism web applications can use to mitigate a broad class of content injection vulnerabilities, such as cross-site scripting (XSS). Content Security Policy is a declarative policy that lets the authors (or server administrators) of a web application inform the client about the sources from which the application expects to load resources.

To mitigate XSS attacks, for example, a web application can declare that it only expects to load script from specific, trusted sources. This declaration allows the client to detect and block malicious scripts injected into the application by an attacker.

Content Security Policy (CSP) is not intended as a first line of defense against content injection vulnerabilities. Instead, CSP is best used as defense-in-depth, to reduce the harm caused by content injection attacks. As a first line of defense against content injection, server operators should validate their input and encode their output.

There is often a non-trivial amount of work required to apply CSP to an existing web application. To reap the greatest benefit, authors will need to move all inline script and style out-of-line, for example into external scripts, because the user agent cannot determine whether an inline script was injected by an attacker.

To take advantage of CSP, a web application opts into using CSP by supplying a Content-Security-Policy HTTP header. Such policies apply to the current resource representation only. To supply a policy for an entire site, the server needs to supply a policy with each resource representation.

1.1. Changes from Level 2

This document describes an evolution of the Content Security Policy specification. Level 3 makes the following changes from Level 2:

  1. The following directives are brand new in this revision:
    1. manifest-src controls the protected resource’s ability to load application manifests [MANIFEST].
    2. reflected-xss controls the user agent’s built-in heuristics to actively protect against XSS. It is meant to supplant the X-XSS-Protection HTTP request header.
A

1.2. Changes from Level 1

Level 2 made two breaking changes from Level 1, and added support for a number of new directives and capabilities which are summarized below:

  1. The following changes are backwards incompatible with the majority of user agent’s implementations of Level 1:
    1. The path component of a source expression is now ignored if the resource being loaded is the result of a redirect, as described in §4.2.2.3 Paths and Redirects.

      Note: Paths are technically new in Level 2, but they were already implemented in many user agents before this revision of CSP was completed, so noting the change here seems reasonable.

    2. A protected resource’s ability to load Workers [WORKERS] is now controlled via child-src rather than script-src.
    3. Workers now have their own policy, separate from the protected resource which loaded them. This is described in §5.1 Workers.
  2. The following directives are brand new in this revision:
    1. base-uri controls the protected resource’s ability to specify the document base URL.
    2. child-src deprecates and replaces frame-src, controlling the protected resource’s ability to embed frames, and to load Workers.
    3. form-action controls the protected resource’s ability to submit forms.
    4. frame-ancestors controls the protected resource’s ability be embedded in other documents. It is meant to supplant the X-Frame-Options HTTP request header. [RFC7034]
    5. plugin-types controls the protected resource’s ability to load specific types of plugins.
  3. Individual inline scripts and stylesheets may be whitelisted via nonces (as described in §4.2.4 Valid Nonces) and hashes (as described in §4.2.5 Valid Hashes).
  4. A CSP request header is now sent with relevant requests, as described in §3.4 The CSP HTTP Request Header.
  5. A SecurityPolicyViolationEvent is fired upon violations, as described in §6.4 Firing Violation Events.
  6. A number of new fields were added to violation reports (both those POSTED via report-uri, and those handed to the DOM via SecurityPolicyViolationEvent events. These include effectiveDirective, statusCode, sourceFile, lineNumber, and columnNumber.
  7. Certain flags present in the sandbox directive now affect Worker creation, as described in §7.16.1 Sandboxing and Workers.

2. Key Concepts and Terminology

The Augmented Backus-Naur Form (ABNF) notation used in this document is specified in RFC5234 [ABNF]. We use the #rule extension defined in [RFC7230].

security policy
security policy directive
security policy directive name
security policy directive value
A security policy refers to both a set of security preferences for restrictions within which content can operate, and to a fragment of text that codifies or transmits these preferences. For example, the following string is a policy which restricts script and object content:
script-src 'self'; object-src 'none'

Security policies contain a set of security policy directives (script-src and object-src in the example above), each responsible for declaring the restrictions for a particular resource type, or manipulating a specific aspect of the policy’s restrictions. The list of directives defined by this specification can be found in §7 Directives.

Each directive has a name and value; value may be optional for some directives. A detailed grammar can be found in §4 Syntax and Algorithms.

protected resource
A security policy is applied by a user agent to a specific resource representation, known as the protected resource. See §3 Policy Delivery for details regarding the mechanisms by which policies may be applied to a protected resource.

3. Policy Delivery

The server delivers a policy to the user agent via an HTTP response header (defined in §3.1 Content-Security-Policy Header Field and §3.2 Content-Security-Policy-Report-Only Header Field) or an HTML meta element (defined in §3.3 HTML meta Element).

Servers are informed that requests are coming from a protected resource via an HTTP request header (defined in §3.4 The CSP HTTP Request Header).

3.1. Content-Security-Policy Header Field

The Content-Security-Policy header field is the preferred mechanism for delivering a policy. The grammar is as follows:

"Content-Security-Policy:" 1#policy-token

For example, a response might include the following header field:

Content-Security-Policy: script-src 'self'

A server MUST NOT send more than one HTTP header field named Content-Security-Policy with a given resource representation.

A server MAY send different Content-Security-Policy header field values with different representations of the same resource or with different resources.

Upon receiving an HTTP response containing at least one Content-Security-Policy header field, the user agent MUST enforce each of the policies contained in each such header field.

3.2. Content-Security-Policy-Report-Only Header Field

The Content-Security-Policy-Report-Only header field lets servers experiment with policies by monitoring (rather than enforcing) a policy. The grammar is as follows:

"Content-Security-Policy-Report-Only:" 1#policy-token

For example, server operators might wish to develop their security policy iteratively. The operators can deploy a report-only policy based on their best estimate of how their site behaves:

Content-Security-Policy-Report-Only: script-src 'self';
                                     report-uri /csp-report-endpoint/

If their site violates this policy the user agent will send violation reports to the URL specified in the policy’s report-uri directive, but allow the violating resources to load regardless. Once a site has confidence that the policy is appropriate, they can start enforcing the policy using the Content-Security-Policy header field.

A server MUST NOT send more than one HTTP header field named Content-Security-Policy-Report-Only with a given resource representation.

A server MAY send different Content-Security-Policy-Report-Only header field values with different representations of the same resource or with different resources.

Upon receiving an HTTP response containing at least one Content-Security-Policy-Report-Only header field, the user agent MUST monitor each of the policies contained in each such header field.

Note: The Content-Security-Policy-Report-Only header is not supported inside a meta element.

3.3. HTML meta Element

The server MAY supply policy via one or more HTML meta elements with http-equiv attributes that are an ASCII case-insensitive match for the string "Content-Security-Policy". For example:

<meta http-equiv="Content-Security-Policy" content="script-src 'self'">

Add the following entry to the pragma directives for the meta element:

Content security policy (http-equiv="content-security-policy")
  1. If the Document’s head element is not an ancestor of the meta element, abort these steps.
  2. If the meta element lacks a content attribute, abort these steps.
  3. Let policy be the value of the content attribute of the meta element.
  4. Let directive-set be the result of parsing policy.
  5. Remove all occurrences of reflected-xss, report-uri, frame-ancestors, and sandbox directives from directive-set.

    Note: User agents are encouraged to issue a warning to developers if one or more of these directives are included in a policy delivered via meta.

  6. Enforce each of the directives in directive-set, as defined for each directive type.

Authors are strongly encouraged to place meta elements as early in the document as possible, because policies in meta elements are not applied to content which preceds them. In particular, note that resources fetched or prefetched using the Link HTTP response header field, and resources fetched or prefetched using link and script elements which precede a meta-delivered policy will not be blocked.

Note: A policy specified via a meta element will be enforced along with any other policies active for the protected resource, regardless of where they’re specified. The general impact of enforcing multiple policies is described in §3.5 Enforcing multiple policies.

Note: Modifications to the content attribute of a meta element after the element has been parsed will be ignored.

Note: The Content-Security-Policy-Report-Only header is not supported inside a meta element.

3.4. The CSP HTTP Request Header

The CSP header field indicates that a particular request is subject to a policy, and its value is defined by the following ABNF grammar:

"CSP:" 1#csp-header-value

csp-header-value = *WSP "active" *WSP

If the user agent is monitoring or enforcing a policy that includes directives whose value is a source list, then the user agent MUST send a header field named CSP along with requests for resources whose origin does not match the protected resource’s origin. The value of this header MUST be active.

The user agent MAY choose to send this header only if the request is for a resource type which the active policy would effect. That is, given a policy of img-src example.com, the user agent would send CSP: active along with requests for images, but might choose not to send the header with requests for script.

Note: The central reason for including this header is that it hints to a server that information about redirects might be leaked as a side-effect of a page’s active policy. If this header is present, a server might decline to redirect a logged-out user from example.com to accounts.example.com, for example, as a malicious embedder might otherwise be able to determine the user’s logged-in status.

3.5. Enforcing multiple policies

This section is not normative.

The above sections note that when multiple policies are present, each must be enforced or reported, according to its type. An example will help clarify how that ought to work in practice. The behavior of an XMLHttpRequest might seem unclear given a site that, for whatever reason, delivered the following HTTP headers:

Content-Security-Policy: default-src 'self' https://example.com https://example.net;
                         connect-src 'none';
Content-Security-Policy: connect-src https://example.com/;
                         script-src https://example.com/

Is a connection to example.com allowed or not? The short answer is that the connection is not allowed. Enforcing both policies means that a potential connection would have to pass through both unscathed. Even though the second policy would allow this connection, the first policy contains connect-src 'none', so its enforcement blocks the connection. The impact is that adding additional policies to the list of policies to enforce can only further restrict the capabilities of the protected resource.

To demonstrate that further, consider a script tag on this page. The first policy would lock scripts down to 'self', https://example.com and https://example.net via the default-src directive. The second, however, would only allow script from https://example.com/. Script will only load if it meets both policy’s criteria: in this case, the only origin that can match is https://example.com, as both policies allow it.

3.6. Policy applicability

This section is not normative.

Policies are associated with an protected resource, and enforced or monitored for that resource. If a resource does not create a new execution context (for example, when including a script, image, or stylesheet into a document), then any policies delivered with that resource are discarded without effect. Its execution is subject to the policy or policies of the including context. The following table outlines examples of these relationships:

Resource Type What policy applies?
Top-level Contexts HTML as a new, top-level browsing context The policy delivered with the resource applies.
SVG, as a top-level document The policy delivered with the resource applies.
Embedded Contexts Any resource included via iframe, object, or embed Unless the embedded resource is a globally unique identifier (or a srcdoc iframe), the embedded resource is controlled by the policy delivered with the resource. If the embedded resource is a globally unique identifier or srcdoc iframe, it inherits the policy of the context creating it. (The frame-src and child-src directives of the embedding context only control what resources are eligible for embedding in that context, not the behavior of the resource once embedded.)
SVG, as an embedded document Unless the resource is a globally unique identifier, it is controlled by the policy delivered with the resource. If a globally unique identifier, it inherits the policy of the context creating it.
JavaScript, as a Worker, Shared Worker, or Service Worker Unless the resource is a globally unique identifier, it is controlled by the policy delivered with the resource. If a globally unique identifier, it inherits the policy of the context creating it.
Subresources SVG, inlined via svg The policy of the including context applies.
SVG, as a resource document The policy of the including context applies.
HTML via XMLHttpRequest The policy of the context that performed the fetch applies.
Image via img element The policy of the including context applies.
JavaScript via a script element The policy of the including context applies.
SVG, via img No policy applies; this should be just as safe as JPG.
SVG, as a WebFont No policy applies; this should be just as safe as WOFF.

4. Syntax and Algorithms

4.1. Policy Syntax

A Content Security Policy consists of a U+003B SEMICOLON (;) delimited list of directives. Each directive consists of a directive name and (optionally) a directive value, defined by the following ABNF:

policy-token    = [ directive-token *( ";" [ directive-token ] ) ]
directive-token = *WSP [ directive-name [ WSP directive-value ] ]
directive-name  = 1*( ALPHA / DIGIT / "-" )
directive-value = *( WSP / <VCHAR except ";" and ","> )

4.1.1. Parsing Policies

To parse the policy policy, the user agent MUST use an algorithm equivalent to the following:

  1. Let the set of directives be the empty set.
  2. For each non-empty token returned by strictly splitting the string policy on the character U+003B SEMICOLON (;):
    1. Skip whitespace.
    2. Collect a sequence of characters that are not space characters. The collected characters are the directive name.
    3. If there are characters remaining in token, skip ahead exactly one character (which must be a space character).
    4. The remaining characters in token (if any) are the directive value.
    5. If the set of directives already contains a directive whose name is a case insensitive match for directive name, ignore this instance of the directive and continue to the next token.
    6. Add a directive to the set of directives with name directive name and value directive value.
  3. Return the set of directives.

4.2. Source List Syntax

Many CSP directives use a value consisting of a source list, defined in the ABNF grammar below.

Each source expression in the source list represents a location from which content of the specified type can be retrieved. For example, the source expression 'none' represents the empty set of URLs, and the source expression 'unsafe-inline' represents content supplied inline in the resource itself.

source-list       = *WSP [ source-expression *( 1*WSP source-expression ) *WSP ]
                  / *WSP "'none'" *WSP
source-expression = scheme-source / host-source / keyword-source / nonce-source / hash-source
scheme-source     = scheme-part ":"
host-source       = [ scheme-part "://" ] host-part [ port-part ] [ path-part ]
keyword-source    = "'self'" / "'unsafe-inline'" / "'unsafe-eval'"
base64-value      = 1*( ALPHA / DIGIT / "+" / "/" )*2( "=" )
nonce-value       = base64-value
hash-value        = base64-value
nonce-source      = "'nonce-" nonce-value "'"
hash-algo         = "sha256" / "sha384" / "sha512"
hash-source       = "'" hash-algo "-" hash-value "'"
scheme-part       = <scheme production from RFC 3986, section 3.1>
host-part         = "*" / [ "*." ] 1*host-char *( "." 1*host-char )
host-char         = ALPHA / DIGIT / "-"
path-part         = <path production from RFC 3986, section 3.3>
port-part         = ":" ( 1*DIGIT / "*" )

If the policy contains a nonce-source expression, the server MUST generate a fresh value for the nonce-value directive at random and independently each time it transmits a policy. The generated value SHOULD be at least 128 bits long (before encoding), and generated via a cryptographically secure random number generator. This requirement ensures that the nonce-value is difficult for an attacker to predict.

Note: Using a nonce to whitelist inline script or style is less secure than not using a nonce, as nonces override the restrictions in the directive in which they are present. An attacker who can gain access to the nonce can execute whatever script they like, whenever they like. That said, nonces provide a substantial improvement over 'unsafe-inline' when layering a content security policy on top of old code. When considering 'unsafe-inline', authors are encouraged to consider nonces (or hashes) instead.

The host-char production intentionally contains only ASCII characters; internationalized domain names cannot be entered directly into a policy string, but instead MUST be Punycode-encoded [RFC3492]. For example, the domain üüüüüü.de would be encoded as xn--tdaaaaaa.de.

NOTE: Though IP addresses do match the grammar above, only 127.0.0.1 will actually match a URL when used in a source expression (see §4.2.2 Matching Source Expressions for details). The security properties of IP addresses are suspect, and authors ought to prefer hostnames to IP addresses whenever possible.

4.2.1. Parsing Source Lists

To parse a source list source list, the user agent MUST use an algorithm equivalent to the following:

  1. Strip leading and trailing whitespace from source list.
  2. If source list is an ASCII case-insensitive match for the string 'none' (including the quotation marks), return the empty set.
  3. Let set of source expressions be the empty set.
  4. For each token returned by splitting source list on spaces, if the token matches the grammar for source-expression, add the token to the set of source expressions.
  5. Return the set of source expressions.

Note: Characters like U+003B SEMICOLON (;) and U+002C COMMA (,) cannot appear in source expressions directly: if you’d like to include these characters in a source expression, they must be percent encoded as %3B and %2C respectively.

4.2.2. Matching Source Expressions

A URL url is said to match a source expression for a protected resource if the following algorithm returns does match:

  1. Let url be the result of processing the URL through the URL parser.
  2. If the source expression a consists of a single U+002A ASTERISK character (*), and url’s scheme is not one of blob, data, filesystem, then return does match.
  3. If the source expression matches the grammar for scheme-source:
    1. If url’s scheme is an ASCII case-insensitive match for the source expression’s scheme-part, return does match.
    2. Otherwise, return does not match.
  4. If the source expression matches the grammar for host-source:
    1. If url’s host is null, return does not match.
    2. Let url-scheme, url-host, and url-port be the scheme, host, and port of url’s origin, respectively.

      Note: If url doesn’t specify a port, then its origin’s port will be the default port for url’s scheme.

    3. Let url-path-list be the path of url.
    4. If the source expression has a scheme-part that is not a case insensitive match for url-scheme, then return does not match.
    5. If the source expression does not have a scheme, return does not match if any of the following are true:
      1. The scheme of the protected resource’s URL is a case insensitive match for HTTP, and url-scheme is not a case insensitive match for either HTTP or HTTPS.
      2. The scheme of the protected resource’s URL is not a case insensitive match for HTTP, and url-scheme is not a case insensitive match for the scheme of the protected resource’s URL.
    6. If the first character of the source expression’s host-part is an U+002A ASTERISK character (*) and the remaining characters, including the leading U+002E FULL STOP character (.), are not a case insensitive match for the rightmost characters of url-host, then return does not match.
    7. If the first character of the source expression’s host-part is not an U+002A ASTERISK character (*) and url-host is not a case insensitive match for the source expression’s host-part, then return does not match.
    8. If the source expression’s host-part matches the IPv4address production from [RFC3986], and is not 127.0.0.1, or is an IPv6 address, return does not match.

      Note: A future version of this specification may allow literal IPv6 and IPv4 addresses, depending on usage and demand. Given the weak security properties of IP addresses in relation to named hosts, however, authors are encouraged to prefer the latter whenever possible.

    9. If the source expression does not contain a port-part and url-port is not the default port for url-scheme, then return does not match.
    10. If the source expression does contain a port-part, then return does not match if both of the following are true:
      1. The port-part does not contain an U+002A ASTERISK character. (*)
      2. The port-part does not represent the same number as url-port.
    11. If the source expression contains a non-empty path-part, and the URL is not the result of a redirect, then:
      1. Let exact-match be true if the final character of path-part is not the U+002F SOLIDUS character (/), and false otherwise.
      2. Let source-expression-path-list be the result of splitting path-part on the U+002F SOLIDUS character (/).
      3. If source-expression-path-list’s length is greater than url-path-list’s length, return does not match.
      4. For each entry in source-expression-path-list:
        1. Percent decode entry.
        2. Percent decode the first item in url-path-list.
        3. If entry is not an ASCII case-insensitive match for the first item in url-path-list, return does not match.
        4. Pop the first item in url-path-list off the list.
      5. If exact-match is true, and url-path-list is not empty, return does not match.
    12. Otherwise, return does match.
  5. If the source expression is a case insensitive match for 'self' (including the quotation marks), then:
    1. Return does match if the origin of url matches the origin of protected resource’s URL.

      Note: This includes IP addresses. That is, a document at https://111.111.111.111/ with a policy of img-src 'self' can load the image https://111.111.111.111/image.png, as the origins match.

  6. Otherwise, return does not match.

Note: This algorithm treats the URLs https://example.com/ and https://example.com./ as non-matching. This is consistent with browser behavior which treats documents served from these URLs as existing in distinct origins.

A URL url is said to match a source list for protected resource if at least one source expression in the set of source expressions obtained by parsing the source list matches url for protected resource.

Note: No URLs match an empty set of source expressions, such as the set obtained by parsing the source list 'none'.

4.2.2.1. Security Considerations for GUID URL schemes

This section is not normative.

As defined above, special URL schemes that refer to specific pieces of unique content, such as "data:", "blob:" and "filesystem:" are excluded from matching a policy of * and must be explicitly listed. Policy authors should note that the content of such URLs is often derived from a response body or execution in a Document context, which may be unsafe. Especially for the default-src and script-src directives, policy authors should be aware that allowing "data:" URLs is equivalent to unsafe-inline and allowing "blob:" or "filesystem:" URLs is equivalent to unsafe-eval.

4.2.2.2. Path Matching

This section is not normative.

The rules for matching source expressions that contain paths are simpler than they look: paths that end with the '/' character match all files in a directory and its subdirectories. Paths that do not end with the '/' character match only one specific file. A few examples should make this clear:

  1. The source expression example.com has no path, and therefore matches any file served from that host.
  2. The source expression example.com/scripts/ matches any file in the scripts directory of example.com, and any of its subdirectories. For example, both https://example.com/scripts/file.js and https://example.com/scripts/js/file.js would match.
  3. The source expression example.com/scripts/file.js matches only the file named file.js in the scripts directory of example.com.
  4. Likewise, the source expression example.com/js matches only the file named js. In particular, note that it would not match files inside a directory named js. Files like example.com/js/file.js would be matched only if the source expression ended with a trailing "/", as in example.com/js/.

Note: Query strings have no impact on matching: the source expression example.com/file matches each of https://example.com/file, https://example.com/file?key=value, https://example.com/file?key=notvalue, and https://example.com/file?notkey=notvalue.

4.2.2.3. Paths and Redirects

To avoid leaking path information cross-origin (as discussed in Egor Homakov’s Using Content-Security-Policy for Evil), the matching algorithm ignores the path component of a source expression if the resource being loaded is the result of a redirect. For example, given a page with an active policy of img-src example.com not-example.com/path:

  • Directly loading https://not-example.com/not-path would fail, as it doesn’t match the policy.
  • Directly loading https://example.com/redirector would pass, as it matches example.com.
  • Assuming that https://example.com/redirector delivered a redirect response pointing to https://not-example.com/not-path, the load would succeed, as the initial URL matches example.com, and the redirect target matches not-example.com/path if we ignore its path component.

This restriction reduces the granularity of a document’s policy when redirects are in play, a necessary compromise to avoid brute-forced information leaks of this type.

The relatively long thread "Remove paths from CSP?" from [email protected] has more detailed discussion around alternate proposals.

4.2.3. The nonce attribute

Nonce sources require a new nonce attribute to be added to both script and style elements.

partial interface HTMLScriptElement {
  attribute DOMString nonce;
};
nonce, of type DOMString
This attribute reflects the value of the element’s nonce content attribute.
partial interface HTMLStyleElement {
  attribute DOMString nonce;
};
nonce, of type DOMString
This attribute reflects the value of the element’s nonce content attribute.

4.2.4. Valid Nonces

An element has a valid nonce for a set of source expressions if the value of the element’s nonce attribute after stripping leading and trailing whitespace is a case-sensitive match for the nonce-value component of at least one nonce-source expression in set of source expressions.

4.2.5. Valid Hashes

An element’s content is the script block’s source for script elements, or the value of the element’s textContent IDL attribute for non-script elements such as style.

The digest of element’s content for is the result of applying a hashing algorithm to the element’s content.

To determine whether element has a valid hash for a set of source expressions, execute the following steps:

  1. Let hashes be a list of all hash-source expressions in set of source expressions.
  2. For each hash in hashes:
    1. Let hashing algorithm be:
    2. Let expected be the hash-value component of hash.
    3. Let actual be the base64 encoding of the binary digest of element’s content using the hashing algorithm.
    4. If actual is a case-sensitive match for expected, return true and abort these steps.
  3. Return false.

Note: If an element has an invalid hash, it would be helpful if the user agent reported the failure to the author by adding a warning message containing the actual hash value.

4.3. Media Type List Syntax

The plugin-types directive uses a value consisting of a media type list.

Each media type in the media type list represents a specific type of resource that can be retrieved and used to instantiate a plugin in the protected resource.

media-type-list   = media-type *( 1*WSP media-type )
media-type        = <type from RFC 2045> "/" <subtype from RFC 2045>

4.3.1. Parsing

To parse a media type list media type list, the user agent MUST use an algorithm equivalent to the following:

  1. Let the set of media types be the empty set.
  2. For each token returned by splitting media type list on spaces, if the token matches the grammar for media-type, add the token to the set of media types. Otherwise ignore the token.
  3. Return the set of media types.

4.3.2. Matching

A media type matches a media type list if, and only if, the media type is an ASCII case-insensitive match for at least one token in the set of media types obtained by parsing the media type list.

4.4. Reporting

To strip uri for reporting, the user agent MUST use an algorithm equivalent to the following:

  1. If the origin of uri is a globally unique identifier (for example, uri has a scheme of data, blob, or filesystem), then abort these steps, and return the ASCII serialization of uri’s scheme.
  2. If the origin of uri is not the same as the origin of the protected resource, then abort these steps, and return the ASCII serialization of uri’s origin.
  3. Return uri, with any fragment component removed.

To generate a violation report object, the user agent MUST use an algorithm equivalent to the following:

  1. Prepare a JSON object violation with the following keys and values:
    blocked-uri
    The originally requested URL of the resource that was prevented from loading, stripped for reporting, or the empty string if the resource has no URL (inline script and inline style, for example).
    document-uri
    The address of the protected resource, stripped for reporting.
    effective-directive
    The name of the policy directive that was violated. This will contain the directive whose enforcement triggered the violation (e.g. "script-src") even if that directive does not explicitly appear in the policy, but is implicitly activated via the default-src directive.
    original-policy
    The original policy, as received by the user agent.
    referrer
    The referrer attribute of the protected resource, or the empty string if the protected resource has no referrer.
    status-code
    The status-code of the HTTP response that contained the protected resource, if the protected resource was obtained over HTTP. Otherwise, the number 0.
    violated-directive
    The policy directive that was violated, as it appears in the policy. This will contain the default-src directive in the case of violations caused by falling back to the default sources when enforcing a directive.
  2. If a specific line or a specific file can be identified as the cause of the violation (for example, script execution that violates the script-src directive), the user agent MAY add the following keys and values to violation:
    source-file
    The URL of the resource where the violation occurred, stripped for reporting.
    line-number
    The line number in source-file on which the violation occurred.
    column-number
    The column number in source-file on which the violation occurred.
  3. Return violation.

Note: blocked-uri will not contain the final location of a resource that was blocked after one or more redirects. It instead will contain only the location that the protected resource requested, before any redirects were followed.

To send violation reports, the user agent MUST use an algorithm equivalent to the following:

  1. Prepare a JSON object report object with a single key, csp-report, whose value is the result of generating a violation report object.
  2. Let report body be the JSON stringification of report object.
  3. For each report URL in the set of report URLs:
    1. If the user agent has already sent a violation report for the protected resource to report URL, and that report contained an entity body that exactly matches report body, the user agent MAY abort these steps and continue to the next report URL.
    2. Queue a task to fetch report URL from the origin of the protected resource, with the synchronous flag not set, using HTTP method POST, with a Content-Type header field of application/csp-report, and an entity body consisting of report body. If the origin of report URL is not the same as the origin of the protected resource, the block cookies flag MUST also be set. The user agent MUST NOT follow redirects when fetching this resource. (Note: The user agent ignores the fetched resource.) The task source for these tasks is the Content Security Policy task source.

To report a violation, the user agent MUST:

  1. Fire a violation event at the protected resource’s Document.
  2. If the set of report URLs is non-empty, send violation reports to each.

Note: This section of the specification should not be interpreted as limiting user agents' ability to apply restrictions to violation reports in order to limit data leakage above and beyond what these algorithms specify. For example, a user agent might offer users the option of disabling reporting entirely.

5. Processing Model

To enforce a policy, the user agent MUST parse the policy and enforce each of the directives contained in the policy, where the specific requirements for enforcing each directive are defined separately for each directive (See §7 Directives, below).

Generally speaking, enforcing a directive prevents the protected resource from performing certain actions, such as loading scripts from URLs other than those indicated in a source list. These restrictions make it more difficult for an attacker to abuse an injection vulnerability in the resource because the attacker will be unable to usurp the resource’s privileges that have been restricted in this way.

Note: User agents may allow users to modify or bypass policy enforcement through user preferences, bookmarklets, third-party additions to the user agent, and other such mechanisms.

To monitor a policy, the user agent MUST parse the policy and monitor each of the directives contained in the policy.

Monitoring a directive does not prevent the protected resource from undertaking any actions. Instead, any actions that would have been prevented by the directives are allowed, but a violation report is generated and reported to the developer of the web application. Monitoring a policy is useful for testing whether enforcing the policy will cause the web application to malfunction.

A server MAY cause user agents to monitor one policy while enforcing another policy by returning both Content-Security-Policy and Content-Security-Policy-Report-Only header fields. For example, if a server operator may wish to enforce one policy but experiment with a stricter policy, she can monitor the stricter policy while enforcing the original policy. Once the server operator is satisfied that the stricter policy does not break the web application, the server operator can start enforcing the stricter policy.

If the user agent monitors or enforces a policy that does not contain any directives, the user agent SHOULD report a warning message in the developer console.

If the user agent monitors or enforces a policy that contains an unrecognized directive, the user agent SHOULD report a warning message in the developer console indicating the name of the unrecognized directive.

If the user agent monitors or enforces a policy that contains a directive that contains a source list, then the user agent MUST set a CSP Request Header when requesting cross-origin resources, as described in §3.4 The CSP HTTP Request Header.

5.1. Workers

Whenever a user agent runs a worker:

  • If the worker’s script’s origin is a globally unique identifier (for example, the worker’s script’s URL has a scheme of data, blob, or filesystem), then:
    • If the user agent is enforcing a CSP policy for the owner document or parent worker, the user agent MUST enforce the CSP policy for the worker.
    • If the user agent is monitoring a CSP policy for the owner document or parent worker, the user agent MUST monitor the CSP policy for the worker.
  • Otherwise:
    • If the worker’s script is delivered with a Content-Security-Policy HTTP header containing the value policy, the user agent MUST enforce policy for the worker.
    • If the worker’s script is delivered with a Content-Security-Policy-Report-Only HTTP header containing the value policy, the user agent MUST monitor policy for the worker.

5.2. srcdoc iframes

Whenever a user agent creates an iframe srcdoc document in a browsing context nested in the protected resource, if the user agent is enforcing any policies for the protected resource, the user agent MUST enforce those policies on the iframe srcdoc document as well.

Whenever a user agent creates an iframe srcdoc document in a browsing context nested in the protected resource, if the user agent is monitoring any policies for the protected resource, the user agent MUST monitor those policies on the iframe srcdoc document as well.

6. Script Interfaces

6.1. Strawman DOM API

This is a sketch of what an API might look like. It is neither complete nor finalized; take it as an outline for discussion, nothing more. The general idea is to model CSP as a set of SecurityPolicy objects, each composed of a sequence of directives.

The following is what I think we’ll need for source-list directives (§4.2 Source List Syntax). It’ll need some adjustment to deal well with the other directive types.

[ NoInterfaceObject ]
interface SecurityPolicySource {
};

[Constructor(DOMString sourceExpression)]
interface SecurityPolicySourceURL : SecurityPolicySource {
    boolean matchesURL(DOMString url);

    readonly attribute DOMString scheme;
    readonly attribute DOMString host;
    readonly attribute DOMString port;
    readonly attribute DOMString path;

    readonly attribute boolean hostWildcard;
    readonly attribute boolean portWildcard;
};

enum SecurityPolicyHashAlgorithm { "SHA-256", "SHA-384", "SHA-512" };

[Constructor(DOMString sourceExpression)]
interface SecurityPolicySourceHash : SecurityPolicySource {
    boolean matchesNode(Node node);

    readonly attribute SecurityPolicyHashAlgorithm algorithm;
    readonly attribute DOMString hash;
};

[Constructor(DOMString sourceExpression)]
interface SecurityPolicySourceNonce : SecurityPolicySource {
    boolean matchesNode(Node node);

    readonly attribute DOMString nonce;
};
[ NoInterfaceObject ]
interface SecurityPolicyDirective {
};

[Constructor(DOMString directiveName, DOMString directiveValue)]
interface SecurityPolicySourceListDirective : SecurityPolicyDirective {
    boolean matchesURL(DOMString url);
    boolean matchesNode(Node node);

    readonly attribute boolean allowRedirects;
    readonly attribute boolean allowEval;

    // TODO: Bikeshed doesn’t like `Sequence`?
    readonly attribute SecurityPolicySource[] sources;
    readonly attribute DOMString directive;
};
enum SecurityPolicyMode {
    "EnforceMode",
    "ReportOnlyMode"
};

[Constructor(DOMString policy, SecurityPolicyMode mode)]
interface SecurityPolicy {
    // Fetch’s 'Request' object has a URL and a RequestContext, which
    // should be everything we need here to check outgoing network
    // requests. It lacks a redirect flag; not sure how to model that
    // yet. TODO.
    boolean allowRequest(Request request);

    // Inline checks can be done on a Node basis; given an HTMLScriptElement,
    // we’d skim through the `script-src` directive’s sources, checking for
    // hash and nonce matches.
    boolean allowNode(Node node);

    // One-offs. Maybe there’s a good way to fold these in somehow?
    boolean allowEval();
    boolean allowInlineEventHandler();

    // TODO: Bikeshed doesn’t like `Sequence`?
    readonly attribute SecurityPolicyDirective[] directives;

    readonly attribute SecurityPolicyMode mode;
    readonly attribute URL reportingEndpoint;
};

6.2. SecurityPolicyViolationEvent Interface

[Constructor(DOMString type, optional SecurityPolicyViolationEventInit eventInitDict)]
interface SecurityPolicyViolationEvent : Event {
    readonly    attribute DOMString documentURL;
    readonly    attribute DOMString referrer;
    readonly    attribute DOMString blockedURL;
    readonly    attribute DOMString violatedDirective;
    readonly    attribute DOMString effectiveDirective;
    readonly    attribute DOMString originalPolicy;
    readonly    attribute DOMString sourceFile;
    readonly    attribute DOMString statusCode;
    readonly    attribute long      lineNumber;
    readonly    attribute long      columnNumber;
};
documentURL, of type DOMString, readonly
Refer to the document-uri property of violation reports for a description of this property.
referrer, of type DOMString, readonly
Refer to the referrer property of violation reports for a description of this property.
blockedURL, of type DOMString, readonly
Refer to the blocked-uri property of violation reports for a description of this property.
violatedDirective, of type DOMString, readonly
Refer to the violated-directive property of violation reports for a description of this property.
effectiveDirective, of type DOMString, readonly
Refer to the effective-directive property of violation reports for a description of this property.
originalPolicy, of type DOMString, readonly
Refer to the original-policy property of violation reports for a description of this property.
statusCode, of type DOMString, readonly
Refer to the status-code property of violation reports for a description of this property.
sourceFile, of type DOMString, readonly
Refer to the source-file property of violation reports for a description of this property.
lineNumber, of type long, readonly
Refer to the line-number property of violation reports for a description of this property.
columnNumber, of type long, readonly
Refer to the column-number property of violation reports for a description of this property.

6.3. SecurityPolicyViolationEventInit Interface

dictionary SecurityPolicyViolationEventInit : EventInit {
    DOMString documentURL;
    DOMString referrer;
    DOMString blockedURL;
    DOMString violatedDirective;
    DOMString effectiveDirective;
    DOMString originalPolicy;
    DOMString sourceFile;
    long      lineNumber;
    long      columnNumber;
};
documentURL, of type DOMString
Refer to the document-uri property of violation reports for a description of this property.
referrer, of type DOMString
Refer to the referrer property of violation reports for a description of this property.
blockedURL, of type DOMString
Refer to the blocked-uri property of violation reports for a description of this property.
violatedDirective, of type DOMString
Refer to the violated-directive property of violation reports for a description of this property.
effectiveDirective, of type DOMString
Refer to the effective-directive property of violation reports for a description of this property.
originalPolicy, of type DOMString
Refer to the original-policy property of violation reports for a description of this property.
sourceFile, of type DOMString
Refer to the source-file property of violation reports for a description of this property.
lineNumber, of type long
Refer to the line-number property of violation reports for a description of this property.
columnNumber, of type long
Refer to the column-number property of violation reports for a description of this property.

6.4. Firing Violation Events

To fire a violation event, the user agent MUST use an algorithm equivalent to the following:

  1. Let report object be the result of generating a violation report object.
  2. Queue a task to fire an event named securitypolicyviolation using the SecurityPolicyViolationEvent interface with the following initializations:
    • blockedURL MUST be initialized to the value of report object’s blocked-uri key.
    • documentURL MUST be initialized to the value of report object’s document-uri key.
    • effectiveDirective MUST be initialized to the value of report object’s effective-directive key.
    • originalPolicy MUST be initialized to the value of report object’s original-policy key.
    • referrer MUST be initialized to the value of report object’s referrer key.
    • violatedDirective MUST be initialized to the value of report object’s violated-directive key.
    • sourceFile MUST be initialized to the value of report object’s source-file key.
    • lineNumber MUST be initialized to the value of report object’s line-number key.
    • columnNumber MUST be initialized to the value of report object’s column-number key.

The task source for these tasks is the Content Security Policy task source.

7. Directives

This section describes the content security policy directives introduced in this specification. Directive names are case insensitive.

In order to protect against Cross-Site Scripting (XSS), web application authors SHOULD include:

In either case, authors SHOULD NOT include either 'unsafe-inline' or data: as valid sources in their policies. Both enable XSS attacks by allowing code to be included directly in the document itself; they are best avoided completely.

7.1. base-uri

The base-uri directive restricts the URLs that can be used to specify the document base URL. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "base-uri"
directive-value   = source-list

The term allowed base URLs refers to the result of parsing the base-uri directive’s value as a source list.

Note: base-uri does not fall back to the default sources.

Step 4 of the algorithm defined in HTML5 to obtain a document’s base URL (resolution of the href attribute of the base element) MUST be changed to:

  1. If the previous step was not successful, or the result of the previous step does not match the allowed base URLs for the protected resource, then the document base URL is fallback base URL. Otherwise, it is the result of the previous step.

7.2. child-src

The child-src directive governs the creation of nested browsing contexts as well as Worker execution contexts. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "child-src"
directive-value   = source-list

The term allowed child sources refers to the result of parsing the child-src directive’s value as a source list if a child-src directive is explicitly specified, and otherwise to the default sources.

7.2.1. Nested Browsing Contexts

To enforce the child-src directive the user agent MUST enforce the frame-src directive.

7.2.2. Workers

Whenever the user agent fetches a URL while processing the Worker or SharedWorker constructors [WORKERS], the user agent MUST act as if there was a fatal network error and no resource was obtained, and report a violation if the URL does not match the allowed child sources for the protected resource.

7.3. connect-src

The connect-src directive restricts which URLs the protected resource can load using script interfaces. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "connect-src"
directive-value   = source-list

The term allowed connection targets refers to the result of parsing the connect-src directive’s value as a source list if the policy contains an explicit connect-src directive, or otherwise to the default sources.

Whenever the user agent fetches a URL in the course of one of the following activities, if the URL does not match the allowed connection targets for the protected resource, the user agent MUST act as if there was a fatal network error and no resource was obtained, and report a violation:

7.3.1. Usage

This section is not normative.

JavaScript offers a few mechanisms that directly connect to an external server to send or receive information. EventSource maintains an open HTTP connection to a server in order to receive push notifications, WebSockets open a bidirectional communication channel between your browser and a server, and XMLHttpRequest allows arbitrary HTTP requests. These are powerful APIs that enable useful functionality, but also provide tempting avenues for data exfiltration.

The connect-src directive allows restricting these sorts of connections to the those matching the source expressions list. Sending a policy that defines a list of source expressions for this directive is straightforward. For example, to limit connections to only example.com, send the following header:

Content-Security-Policy: connect-src example.com

Each of the following will fail with the preceding directive in place:

  • new WebSocket("wss://evil.com/");
  • (new XMLHttpRequest()).open("GET", "https://evil.com/", true);
  • new EventSource("https://evil.com");

7.4. default-src

The default-src directive sets a default source list for a number of directives. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "default-src"
directive-value   = source-list

Let the default sources be the result of parsing the default-src directive’s value as a source list if a default-src directive is explicitly specified, and otherwise the U+002A ASTERISK character (*).

To enforce the default-src directive, the user agent MUST enforce the following directives:

If not specified explicitly in the policy, the directives listed above will use the default sources as their source list.

7.4.1. Usage

This section is not normative.

default-src, as the name implies, serves as a default source list which the other source list-style directives will use as a fallback if they’re not otherwise explicitly set. That is, consider the following policy declaration:

Content-Security-Policy: default-src 'self'

Under this policy, fonts, frames, images, media, objects, scripts, and styles will all only load from the same origin as the protected resource, and connections will only be made to the same origin. Adding a more specific declaration to the policy would completely override the default source list for that resource type.

Content-Security-Policy: default-src 'self'; script-src example.com

Under this new policy, fonts, frames, and etc. continue to be load from the same origin, but scripts will only load from example.com. There’s no inheritance; the script-src directive sets the allowed sources of script, and the default list is not used for that resource type.

Given this behavior, one good way of building a policy for a site would be to begin with a default-src of 'none', and to build up a policy from there that contains only those resource types which are actually in use for the page you’d like to protect. If you don’t use webfonts, for instance, there’s no reason to specify a source list for font-src; specifying only those resource types a page uses ensures that the possible attack surface for that page remains as small as possible.

7.5. font-src

The font-src directive restricts from where the protected resource can load fonts. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "font-src"
directive-value   = source-list

The term allowed font sources refers to the result of parsing the font-src directive’s value as a source list if the policy contains an explicit font-src, or otherwise to the default sources.

Whenever the user agent fetches a URL in the course of one of the following activities, if the URL does not match the allowed font sources for the protected resource, the user agent MUST act as if there was a fatal network error and no resource was obtained, and report a violation:

  • Requesting data for display in a font, such as when processing the <<@font-face>> Cascading Style Sheets (CSS) rule.

7.6. form-action

The form-action restricts which URLs can be used as the action of HTML form elements. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "form-action"
directive-value   = source-list

The term allowed form actions refers to the result of parsing the form-action directive’s value as a source list.

Whenever the user agent fetches a URL in the course of processing an HTML form element, if the URL does not match the allowed form actions for the protected resource, the user agent MUST act as if there was a fatal network error and no resource was obtained, and report a violation.

Note: form-action does not fall back to the default sources when the directive is not defined. That is, a policy that defines default-src 'none' but not form-action will still allow form submissions to any target.

7.7. frame-ancestors

The frame-ancestors directive indicates whether the user agent should allow embedding the resource using a frame, iframe, object, embed or applet element, or equivalent functionality in non-HTML resources. Resources can use this directive to avoid many UI Redressing [UIREDRESS] attacks by avoiding being embedded into potentially hostile contexts.

The syntax for the name and value of the directive are described by the following ABNF grammar:

ancestor-source-list = [ ancestor-source *( 1*WSP ancestor-source ) ] / "'none'"
ancestor-source      = scheme-source / host-source / "'self'"

directive-name  = "frame-ancestors"
directive-value = ancestor-source-list

The term allowed frame ancestors refers to the result of parsing the frame-ancestors directive’s value as a source list. If a frame-ancestors directive is not explicitly included in the policy, then allowed frame ancestors is "*".

To enforce the frame-ancestors directive, whenever the user agent would load the protected resource into a nested browsing context, the user agent MUST perform the following steps:

  1. Let nestedContext be the nested browsing context into which the protected resource is being loaded.
  2. Let ancestorList be the list of all ancestors of nestedContext.
  3. For each ancestorContext in ancestorList:
    1. Let document be ancestorContext’s active document.
    2. If document’s URL does not match the allowed frame ancestors for the protected resource, the user agent MUST:
      1. Abort loading the protected resource.
      2. Take one of the following actions:
      3. Parse a sandboxing directive using the empty string as the input and the newly created document’s forced sandboxing flag set as the output.
      4. Report a violation.
      5. Abort these steps.

Steps 3.2.2 and 3.2.3 ensure that the blocked frame appears to be a normal cross-origin document load. If these steps are ignored, leakage of a document’s policy state is possible.

The frame-ancestors directive MUST be ignored when monitoring a policy, and when a contained in a policy defined via a meta element.

Note: frame-ancestors does not fall back to the default sources when the directive is not defined. That is, a policy that defines default-src 'none' but not frame-ancestors will still allow the resource to be framed from anywhere.

When generating a violation report for a frame-ancestors violation, the user agent MUST NOT include the value of the embedding ancestor as a blocked-uri value unless it is same-origin with the protected resource, as disclosing the value of cross-origin ancestors is a violation of the Same-Origin Policy.

7.7.1. Relation to X-Frame-Options

This directive is similar to the X-Frame-Options header that several user agents have implemented. The 'none' source expression is roughly equivalent to that header’s DENY, 'self' to SAMEORIGIN, and so on. The major difference is that many user agents implement SAMEORIGIN such that it only matches against the top-level document’s location. This directive checks each ancestor. If any ancestor doesn’t match, the load is cancelled. [RFC7034]

The frame-ancestors directive obsoletes the X-Frame-Options header. If a resource has both policies, the frame-ancestors policy SHOULD be enforced and the X-Frame-Options policy SHOULD be ignored.

7.7.2. Multiple Host Source Values

This section is not normative.

Multiple source-list expressions are allowed in a single policy (in contrast to X-Frame-Options, which allows only one) to enable scenarios involving embedded application components that are multiple levels below the top-level browsing context.

Many common scenarios for embedding (e.g. embeddable payment, sharing or social apps) involve potentially many hundreds or thousands of valid source-list expressions, but it is strongly recommended against accommodating such scenarios with a static frame-ancestors directive listing multiple values. In such cases it is beneficial to generate this value dynamically, based on an HTTP Referer header or an explicitly passed-in value, to allow only the sources necessary for each given embedding of the resource.

Consider a service providing a payments application at https://payments/makeEmbedded. The service allows this resource to be embedded by both merchant Alice and merchant Bob, who compete with each other. Sending:

Content-Security-Policy: frame-ancestors https://alice https://bob

would allow Bob to re-frame Alice’s resource and create fraudulent clicks, perhaps discrediting Alice with her customers or the payments service. If the payments service used additional information (e.g. as part of a URL like https://payments/makeEmbedded?merchant=alice) to send individually-tailored headers listing only the source-list expressions needed by each merchant, this attack would be eliminated.

7.8. frame-src

The frame-src directive is deprecated. Authors who wish to govern nested browsing contexts SHOULD use the child-src directive instead.

The frame-src directive restricts from where the protected resource can embed frames. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "frame-src"
directive-value   = source-list

The term allowed frame sources refers to the result of parsing the frame-src directive’s value as a source list if the policy contains an explicit frame-src, or otherwise to the list of allowed child sources.

Whenever the user agent fetches a URL in the course of one of the following activities, if the URL does not match the allowed frame sources for the protected resource, the user agent MUST act as if there was a fatal network error and no resource was obtained, and report a violation:

7.9. img-src

The img-src directive restricts from where the protected resource can load images. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "img-src"
directive-value   = source-list

The term allowed image sources refers to the result of parsing the img-src directive’s value as a source list if the policy contains an explicit img-src, or otherwise to the list of default sources.

Whenever the user agent fetches a URL in the course of one of the following activities, if the URL does not match the allowed image sources for the protected resource, the user agent MUST act as if there was a fatal network error and no resource was obtained, and report a violation:

  • Requesting data for an image, such as when processing the src or srcset attributes of an img element, the src attribute of an input element with a type of image, the poster attribute of a video element, the url(), image() or image-set() values on any Cascading Style Sheets (CSS) property that is capable of loading an image [CSS4-IMAGES], or the href attribute of a link element with an image-related rel attribute, such as icon.

7.10. manifest-src

The manifest-src directive restricts which application manifest [MANIFEST] the user agent may apply to the protected resource. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "manifest-src"
directive-value   = source-list

The term allowed manifest sources refers to the result of parsing the manifest-src directive’s value as a source list if the policy contains an explicit manifest-src, or otherwise to the default sources.

Whenever the user agent fetches a URL that does not match the allowed manifest sources in the course of executing the steps for obtaining a manifest, the user agent MUST act as if there was a fatal network error and no resource was obtained, and report a violation.

7.11. media-src

The media-src directive restricts from where the protected resource can load video, audio, and associated text tracks. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "media-src"
directive-value   = source-list

The term allowed media sources refers to the result of parsing the media-src directive’s value as a source list if the policy contains an explicit media-src, or otherwise to the list of default sources.

Whenever the user agent fetches a URL in the course of one of the following activities, if the URL does not match the allowed media sources for the protected resource, the user agent MUST act as if there was a fatal network error and no resource was obtained, and report a violation:

  • Requesting data for a video or audio clip, such as when processing the src attribute of a video, audio, source, or track element.

7.12. object-src

The object-src directive restricts from where the protected resource can load plugins. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "object-src"
directive-value   = source-list

The term allowed object sources refers to the result of parsing the object-src directive’s value as a source list if the policy contains an explicit object-src, or otherwise to the list of default sources.

Whenever the user agent fetches a URL in the course of one of the following activities, if the URL does not match the allowed object sources for the protected resource, the user agent MUST act as if there was a fatal network error and no resource was obtained, and report a violation:

It is not required that the consumer of the element’s data be a plugin in order for the object-src directive to be enforced. Data for any object, embed, or applet element MUST match the allowed object sources in order to be fetched. This is true even when the element data is semantically equivalent to content which would otherwise be restricted by one of the other §7 Directives, such as an object element with a text/html MIME type.

Whenever the user agent would load a plugin without an associated URL (e.g., because the object element lacked a data attribute), if the protected resource’s URL does not match the allowed object sources for the protected resource, the user agent MUST NOT load the plugin.

7.13. plugin-types

The plugin-types directive restricts the set of plugins that can be invoked by the protected resource by limiting the types of resources that can be embedded. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "plugin-types"
directive-value   = media-type-list

The term allowed plugin media types refers to the result of parsing the plugin-types directive’s value as a media type list.

Whenever the user agent would instantiate a plugin to handle resource while enforcing the plugin-types directive, the user agent MUST instead act as though the plugin reported an error and report a violation if any of the following conditions hold:

Note: In any of these cases, acting as though the plugin reported an error will cause the user agent to display the fallback content.

Whenever the user agent creates a plugin document as the active document of a child browsing context of the protected resource, if the user agent is enforcing any plugin-types directives for the protected resource, the user agent MUST enforce those plugin-types directives on the plugin document as well.

Whenever the user agent creates a plugin document as the active document of a child browsing context of the protected resource, if the user agent is monitoring any plugin-types directives for the protected resource, the user agent MUST monitor those plugin-types directives on the plugin document as well.

7.13.1. Usage

This section is not normative.

The plugin-types directive whitelists a certain set of MIME types that can be embedded in a protected resource. For example, a site might want to ensure that PDF content loads, but that no other plugins can be instantiated. The following directive would satisfy that requirement:

Content-Security-Policy: plugin-types application/pdf

Resources embedded via an embed or object element delivered with an application/pdf content type would be rendered in the appropriate plugin; resources delivered with some other content type would be blocked. Multiple types can be specified, in any order. If the site decided to additionally allow Flash at some point in the future, it could do so with the following directive:

Content-Security-Policy: plugin-types application/pdf application/x-shockwave-flash

Note: Wildcards are not accepted in the plugin-types directive. Only the resource types explicitly listed in the directive will be allowed.

7.13.2. Predeclaration of expected media types

This section is not normative.

Enforcing the plugin-types directive requires that object and embed elements declare the expected media type of the resource they include via the type attribute. If an author expects to load a PDF, she could specify this as follows:

<object data="resource" type="application/pdf"></object>

If resource isn’t actually a PDF file, it won’t load. This prevents certain types of attacks that rely on serving content that unexpectedly invokes a plugin other than that which the author intended.

Note: resource will not load in this scenario even if its media type is otherwise whitelisted: resources will only load when their media type is whitelisted and matches the declared type in their containing element.

7.14. reflected-xss

The reflected-xss directive instructs a user agent to activate or deactivate any heuristics used to filter or block reflected cross-site scripting attacks. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "reflected-xss"
directive-value   = "allow" / "block" / "filter"

A user agent with support for XSS protection MUST enforce this directive as follows:

  • If the value of the directive is allow, the user agent MUST disable its active protections against reflected cross-site scripting attacks for the protected resource.
  • If the value of the directive is filter, the user agent MUST enable its active protections against reflected cross-site scripting attacks for the protected resource. This might result in filtering script that is believed to be reflected being filtered or selectively blocking script execution.
  • If the value of the directive is block, the user agent MUST stop rendering the protected resource upon detection of reflected script, and instead act as if there was a fatal network error and no resource was obtained, and report a violation:

If the user agent’s active protections against reflected cross-site scripting attacks detect or prevent script execution, the user agent MUST report a violation.

Note: The reflected-xss directive will be ignored if contained within a meta element.

7.14.1. Relationship to X-XSS-Protection

This directive is meant to subsume the functionality provided by the proprietary X-XSS-Protection HTTP header which is supported by a number of user agents. Roughly speaking:

  • reflected-xss allow is equivalent to X-XSS-Protection: 0
  • reflected-xss filter is equivalent to X-XSS-Protection: 1
  • reflected-xss block is equivalent to X-XSS-Protection: 1; mode=block

7.15. report-uri

The report-uri directive specifies a URL to which the user agent sends reports about policy violation. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "report-uri"
directive-value   = uri-reference *( 1*WSP uri-reference )
uri-reference     = <URL-reference from RFC 3986>

The set of report URLs is the value of the report-uri directive, each resolved relative to the protected resource’s URL.

The process of sending violation reports to the URLs specified in this directive’s value is defined in this document’s §4.4 Reporting section.

Note: The report-uri directive will be ignored if contained within a meta element.

7.16. sandbox

The sandbox directive specifies an HTML sandbox policy that the user agent applies to the protected resource. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "sandbox"
directive-value   = "" / sandbox-token *( 1*WSP sandbox-token )
sandbox-token     = <token from RFC 7230>

When enforcing the sandbox directive, the user agent MUST parse a sandboxing directive using the directive-value as the input and protected resource’s forced sandboxing flag set as the output. [HTML5]

The sandbox directive will be ignored when monitoring a policy, and when contained in a policy defined via a meta element. Moreover, this directive has no effect when monitored, and has no reporting requirements.

7.16.1. Sandboxing and Workers

When delivered via an HTTP header, a Content Security Policy may indicate that sandboxing flags ought to be applied to a JavaScript execution environment that is not a Document. Of particular interest is the script content intended for use as a Worker, Shared Worker, or Service Worker. Many of the sandboxing flags do not apply to such environments, but allow-scripts and allow-same-origin have special requirements.

When a resource is loaded while executing the runs a Worker algorithm, the user agent MUST act as if there was a fatal network error and no resource could be obtained if either of the following conditions holds:

  1. The sandbox directive delivered with the resource does not contain the allow-scripts flag.
  2. The sandbox directive delivered with the resource does not contain the allow-same-origin flag, and the creation of the new execution context requires it to be same-origin with its creating context.

7.16.2. Usage

This section is not normative.

HTML5 defines a sandbox attribute for iframe elements, intended to allow web authors to reduce the risk of including potentially untrusted content by imposing restrictions on that content’s abilities. When the attribute is set, the content is forced into a unique origin, prevented from submitting forms, running script, creating or navigating other browsing contexts, and prevented from running plugins. These restrictions can be loosened by setting certain flags as the attribute’s value.

The sandbox directive allows any resource, framed or not, to ask for the same sorts of restrictions to be applied to itself.

For example, a message board or email system might provide downloads of arbitrary attachments provided by other users. Attacks that rely on tricking a client into rendering one of these attachments could be mitigated by requesting that resources only be rendered in a very restrictive sandbox. Sending the sandbox directive with an empty value establishes such an environment:

Content-Security-Policy: sandbox

More trusted resources might be allowed to run in an environment with fewer restrictions by adding allow-* flags to the directive’s value. For example, you can allow a page that you trust to run script, while ensuring that it isn’t treated as same-origin with the rest of your site. This can be accomplished by sending the sandbox directive with the allow-scripts flag:

Content-Security-Policy: sandbox allow-scripts

The set of flags available to the CSP directive should match those available to the iframe attribute. Currently, those include:

Note: Like the rest of Content Security Policy, the sandbox directive is meant as a defense-in-depth. Web authors would be well-served to use it in addition to standard sniffing-mitigation and privilege-reduction techniques.

7.17. script-src

The script-src directive restricts which scripts the protected resource can execute. The directive also controls other resources, such as XSLT style sheets [XSLT], which can cause the user agent to execute script. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "script-src"
directive-value   = source-list

The term allowed script sources refers to the result of parsing the script-src directive’s value as a source list if the policy contains an explicit script-src, or otherwise to the default sources.

If 'unsafe-inline' is not in the list of allowed script sources, or if at least one nonce-source or hash-source is present in the list of allowed script sources:

  • Whenever the user agent would execute an inline script from a script element that lacks a valid nonce and lacks a valid hash for the allowed script sources, instead the user agent MUST NOT execute script, and MUST report a violation.
  • Whenever the user agent would execute an inline script from an inline event handler, instead the user agent MUST NOT execute script, and MUST report a violation.
  • Whenever the user agent would execute script contained in a javascript URL, instead the user agent MUST NOT execute the script, and MUST report a violation.

If 'unsafe-eval' is not in allowed script sources:

  • Instead of evaluating their arguments, both operator eval and function eval [ECMA-262] MUST throw an EvalError exception.
  • When called as a constructor, the function Function [ECMA-262] MUST throw an EvalError exception.
  • When called with a first argument which returns false when passed into IsCallable() (a string, for example), the setTimeout() function MUST return zero without creating a timer.
  • When called with a first argument which returns false when passed into IsCallable() (a string, for example), the setInterval() function MUST return zero without creating a timer.

Whenever the user agent fetches a URL (including when following redirects) in the course of one of the following activities, if the URL does not match the allowed script sources for the protected resource, the user agent MUST act as if there was a fatal network error and no resource was obtained, and report a violation:

  • Requesting a script while processing the src attribute of a script element that lacks a valid nonce for the allowed script sources.
  • Requesting a script while invoking the importScripts method on a WorkerGlobalScope object. [WORKERS]
  • Requesting an HTML component, such as when processing the href attribute of a link element with a rel attribute containing the token import. [HTML-IMPORTS]
  • Requesting an Extensible Stylesheet Language Transformations (XSLT) [XSLT], such as when processing the <?xml-stylesheet?> processing directive in an XML document [XML11], the href attributes on <xsl:include> and <xsl:import> elements.

7.17.1. Nonce usage for script elements

This section is not normative.

The script-src directive lets developers specify exactly which script elements on a page were intentionally included for execution. Ideally, developers would avoid inline script entirely and whitelist scripts by URL. However, in some cases, removing inline scripts can be difficult or impossible. For those cases, developers can whitelist scripts using a randomly generated nonce.

Usage is straightforward. For each request, the server generates a unique value at random, and includes it in the Content-Security-Policy header:

Content-Security-Policy: default-src 'self';
                         script-src 'self' https://example.com 'nonce-$RANDOM'

This same value is then applied as a nonce attribute to each script element that ought to be executed. For example, if the server generated the random value Nc3n83cnSAd3wc3Sasdfn939hc3, the server would send the following policy:

Content-Security-Policy: default-src 'self';
                         script-src 'self' https://example.com 'nonce-Nc3n83cnSAd3wc3Sasdfn939hc3'

Script elements can then execute either because their src URLs are whitelisted or because they have a valid nonce:

<script>
alert("Blocked because the policy doesn’t have 'unsafe-inline'.")
</script>

<script nonce="EDNnf03nceIOfn39fn3e9h3sdfa">
alert("Still blocked because nonce is wrong.")
</script>

<script nonce="Nc3n83cnSAd3wc3Sasdfn939hc3">
alert("Allowed because nonce is valid.")
</script>

<script src="https://example.com/allowed-because-of-src.js"></script>

<script nonce="EDNnf03nceIOfn39fn3e9h3sdfa"
    src="https://elsewhere.com/blocked-because-nonce-is-wrong.js"></script>

<script nonce="Nc3n83cnSAd3wc3Sasdfn939hc3"
    src="https://elsewhere.com/allowed-because-nonce-is-valid.js"></script>

Note that the nonce’s value is not a hash or signature that verifies the contents of the script resources. It’s quite simply a random string that informs the user agent which scripts were intentionally included in the page.

Script elements with the proper nonce execute, regardless of whether they’re inline or external. Script elements without the proper nonce don’t execute unless their URLs are whitelisted. Even if an attacker is able to inject markup into the protected resource, the attack will be blocked by the attacker’s inability to guess the random value.

7.17.2. Hash usage for script elements

This section is not normative.

The script-src directive lets developers whitelist a particular inline script by specifying its hash as an allowed source of script.

Usage is straightforward. The server computes the hash of a particular script block’s contents, and includes the base64 encoding of that value in the Content-Security-Policy header:

Content-Security-Policy: default-src 'self';
                         script-src 'self' https://example.com 'sha256-base64 encoded hash'

Each inline script block’s contents are hashed, and compared against the whitelisted value. If there’s a match, the script is executed. For example, the SHA-256 digest of alert('Hello, world.'); is qznLcsROx4GACP2dm0UCKCzCG+HiZ1guq6ZZDob/Tng=.

You can obtain the digest of a string on the command line simply via the openssl program. For example:
echo -n "alert('Hello, world.');" | openssl dgst -sha256 -binary | openssl enc -base64

If the server sent the following header:

Content-Security-Policy: script-src 'sha256-qznLcsROx4GACP2dm0UCKCzCG+HiZ1guq6ZZDob/Tng='

Then the following script tag would result in script execution:

<script>alert('Hello, world.');</script>

Whitespace is significant. The following scripts blocks would not hash to the same value, and would therefore not execute:

<script> alert('Hello, world.');</script>
<script>alert('Hello, world.'); </script>
<script> alert('Hello, world.'); </script>
<script>
alert('Hello, world.');
</script>

Note also that the hash applies only to inline script. An externalized script containing the value alert('Hello, world.'); would not execute if its origin was not whitelisted as a valid source of script.

7.18. style-src

The style-src directive restricts which styles the user agent applies to the protected resource. The syntax for the name and value of the directive are described by the following ABNF grammar:

directive-name    = "style-src"
directive-value   = source-list

The term allowed style sources refers to the result of parsing the style-src directive’s value as a source list if the policy contains an explicit style-src, or otherwise to the default sources.

If 'unsafe-inline' is not in the list of allowed style sources, or if at least one nonce-source or hash-source is present in the list of allowed style sources:

Note: These restrictions on inline do not prevent the user agent from applying style from an external stylesheet (e.g., found via <link rel="stylesheet" ...>).

If 'unsafe-eval' is not in allowed style sources, then:

Whenever the user agent fetches a URL in the course of one of the following activities, if the URL does not match the allowed style sources for the protected resource, the user agent MUST act as if there was a fatal network error and no resource was obtained, and report a violation:

  • Requesting an external stylesheet when processing the href of a link element whose rel attribute contains the token stylesheet.
  • Requesting an external stylesheet when processing the <<@import>> directive.
  • Requesting an external stylesheet when processing a Link HTTP response header field. [RFC5988]

    Note: As this stylesheet might be prefetched before a Document actually exists, user agents will need to carefully consider how to instantiate a meaningful policy against which to compare this request. See §10.1 Processing Complications for more detail.

Note: The style-src directive does not restrict the use of XSLT. XSLT is restricted by the script-src directive because the security consequences of including an untrusted XSLT stylesheet are similar to those incurred by including an untrusted script.

7.18.1. Nonce usage for style elements

This section is not normative.

See the script-src nonce usage information for detail; the application of nonces to style elements is similar enough to avoid repetition here.

7.18.2. Hash usage for style elements

This section is not normative.

See the script-src hash usage information for detail; the application of hashes to style elements is similar enough to avoid repetition here.

8. Examples

8.1. Sample Policy Definitions

This section provides some sample use cases and supporting policies.

A server wishes to load resources only from its own origin:
Content-Security-Policy: default-src 'self'
An auction site wishes to load images from any URL, plugin content from a list of trusted media providers (including a content distribution network), and scripts only from a server under its control hosting sanitized ECMAScript:
Content-Security-Policy:
    default-src 'self'; img-src *;
    object-src media1.example.com media2.example.com *.cdn.example.com;
    script-src trustedscripts.example.com
An online banking site wishes to ensure that all of the content in its pages is loaded over TLS to prevent attackers from eavesdropping on insecure content requests:
Content-Security-Policy: default-src https: 'unsafe-inline' 'unsafe-eval'

This policy allows inline content (such as inline script elements), use of eval, and loading resources over https. Note: This policy does not provide any protection from cross-site scripting vulnerabilities.

A website that relies on inline script elements wishes to ensure that script is only executed from its own origin, and those elements it intentionally inserted inline:
Content-Security-Policy: script-src 'self' 'nonce-$RANDOM';

The inline script elements would then only execute if they contained a matching nonce attribute:

<script nonce="$RANDOM">...</script>

8.2. Sample Violation Report

This section contains an example violation report the user agent might send to a server when the protected resource violates a sample policy.

In the following example, the user agent rendered a representation of the resource http://example.org/page.html with the following policy:

default-src 'self'; report-uri http://example.org/csp-report.cgi

The protected resource loaded an image from http://evil.example.com/image.png, violating the policy.

{
  "csp-report": {
    "document-uri": "http://example.org/page.html",
    "referrer": "http://evil.example.com/haxor.html",
    "blocked-uri": "http://evil.example.com/image.png",
    "violated-directive": "default-src 'self'",
    "effective-directive": "img-src",
    "original-policy": "default-src 'self'; report-uri http://example.org/csp-report.cgi"
  }
}

9. Security Considerations

9.1. Cascading Style Sheet (CSS) Parsing

The style-src directive restricts the locations from which the protected resource can load styles. However, if the user agent uses a lax CSS parsing algorithm, an attacker might be able to trick the user agent into accepting malicious "stylesheets" hosted by an otherwise trustworthy origin.

These attacks are similar to the CSS cross-origin data leakage attack described by Chris Evans in 2009. User agents SHOULD defend against both attacks using the same mechanism: stricter CSS parsing rules for style sheets with improper MIME types.

9.2. Violation Reports

The violation reporting mechanism in this document has been designed to mitigate the risk that a malicious web site could use violation reports to probe the behavior of other servers. For example, consider a malicious web site that white lists https://example.com as a source of images. If the malicious site attempts to load https://example.com/login as an image, and the example.com server redirects to an identity provider (e.g., identityprovider.example.net), CSP will block the request. If violation reports contained the full blocked URL, the violation report might contain sensitive information contained in the redirected URL, such as session identifiers or purported identities. For this reason, the user agent includes only the origin of the blocked URL.

10. Implementation Considerations

The Content-Security-Policy header is an end-to-end header. It is processed and enforced at the client and, therefore, SHOULD NOT be modified or removed by proxies or other intermediaries not in the same administrative domain as the resource.

The originating administrative domain for a resource might wish to apply a Content-Security-Policy header outside of the immediate context of an application. For example, a large organization might have many resources and applications managed by different individuals or teams but all subject to a uniform organizational standard. In such situations, a Content-Security-Policy header might be added or combined with an existing one at a network-edge security gateway device or web application firewall. To enforce multiple policies, the administrator SHOULD combine the policy into a single header. An administrator might wish to use different combination algorithms depending on his or her intended semantics.

One sensible policy combination algorithm is to start by allowing a default set of sources and then letting individual upstream resource owners expand the set of allowed sources by including additional origins. In this approach, the resultant policy is the union of all allowed origins in the input policies.

Another sensible policy combination algorithm is to intersect the given policies. This approach enforces that content comes from a certain whitelist of origins, for example, preventing developers from including third-party scripts or content in violation of organizational standards and practices. In this approach, the combination algorithm forms the combined policy by removing disallowed hosts from the policies supplied by upstream resource owners.

Interactions between the default-src and other directives SHOULD be given special consideration when combining policies. If none of the policies contains a default-src directive, adding new src directives results in a more restrictive policy. However, if one or more of the input policies contain a default-src directive, adding new src directives might result in a less restrictive policy, for example, if the more specific directive contains a more permissive set of allowed origins.

Using a more restrictive policy than the input policy authored by the resource owner might prevent the resource from rendering or operating as intended.

Note: Migration to HTTPS from HTTP may require updates to the policy in order to keep things running as before. Source expressions like http://example.com do not match HTTPS resources. For example, administrators SHOULD carefully examine existing policies before rolling out HTTP Strict Transport Security headers for an application. [RFC6797]

Server administrators MAY wish to send multiple policies if different reporting options are desired for subsets of an overall policy. For instance, the following headers:
Content-Security-Policy: frame-ancestors https://example.com/ 
Content-Security-Policy: default-src https:; report-uri https://example.com/

would send violation reports for http resources, but would not send violation reports for frame-ancestors violations. Note also that combining them via ',' into the single header

Content-Security-Policy: frame-ancestors https://example.com/, default-src https:; report-uri https://example.com/

would have the same effect, as the comma splits the header during parsing.

10.1. Processing Complications

Many user agents implement some form of optimistic resource fetching algorithm to speed up page loads. In implementing these features, user agents MUST ensure that these optimizations do not alter the behavior of the page’s security policy.

Here, we’ll note a few potential complications that could cause bugs in implementations:

  1. The frame-ancestor directive MUST take effect before a document is loaded into a nested browsing context, and certainly before script is potentially executed. One way to approach this constraint is to perform the ancestor check defined in §7.7 frame-ancestors while parsing the document’s headers. This might mean that no document object is available at all, which can complicate checks against 'self', and scheme- or port-relative source expressions.
  2. Likewise, the Link HTTP response header could generate requests for stylesheet resources before a document is available. User agents MUST ensure that any policy contained in the response headers is parsed and effective before these requests are generated. For example, a response returning the following headers:
    Content-Security-Policy: style-src 'none'
    Link: <awesome.css>; rel=stylesheet
    

    MUST have the same behavior as a response returning the following headers:

    Link: <awesome.css>; rel=stylesheet
    Content-Security-Policy: style-src 'none'
    

    namely, both must block requests for the stylesheet. To meet this requirement user agents MUST wait until all headers have been processed before beginning to prefetch resources.

11. IANA Considerations

The permanent message header field registry should be updated with the following registrations: [RFC3864]

11.1. Content-Security-Policy

Header field name
Content-Security-Policy
Applicable protocol
http
Status
standard
Author/Change controller
W3C
Specification document
This specification (See Content-Security-Policy Header Field)

11.2. Content-Security-Policy-Report-Only

Header field name
Content-Security-Policy-Report-Only
Applicable protocol
http
Status
standard
Author/Change controller
W3C
Specification document
This specification (See Content-Security-Policy-Report-Only Header Field)

11.3. CSP

Header field name
CSP
Applicable protocol
http
Status
standard
Author/Change controller
W3C
Specification document
This specification (See §3.4 The CSP HTTP Request Header)

12. Acknowledgements

In addition to the documents in the W3C Web Application Security working group, the work on this document is also informed by the work of the IETF websec working group, particularly that working group’s requirements document: draft-hodges-websec-framework-reqs.

A portion of the frame-ancestors directive was originally developed as X-Frame-Options. [RFC7034]

Brian Smith, Neil Matatall, Anne van Kesteren, and Sigbjørn Vik provided particularly insightful feedback to keep this specification sane.

Conformance

Document conventions

Conformance requirements are expressed with a combination of descriptive assertions and RFC 2119 terminology. The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in the normative parts of this document are to be interpreted as described in RFC 2119. However, for readability, these words do not appear in all uppercase letters in this specification.

All of the text of this specification is normative except sections explicitly marked as non-normative, examples, and notes. [RFC2119]

Examples in this specification are introduced with the words "for example" or are set apart from the normative text with class="example", like this:

This is an example of an informative example.

Informative notes begin with the word "Note" and are set apart from the normative text with class="note", like this:

Note, this is an informative note.

Conformant Algorithms

Requirements phrased in the imperative as part of algorithms (such as "strip any leading space characters" or "return false and abort these steps") are to be interpreted with the meaning of the key word ("must", "should", "may", etc) used in introducing the algorithm.

Conformance requirements phrased as algorithms or specific steps can be implemented in any manner, so long as the end result is equivalent. In particular, the algorithms defined in this specification are intended to be easy to understand and are not intended to be performant. Implementers are encouraged to optimize.

Conformance Classes

A conformant user agent must implement all the requirements listed in this specification that are applicable to user agents.

A conformant server must implement all the requirements listed in this specification that are applicable to servers.

Index

Terms defined by this specification

Terms defined by reference

References

Normative References

[ABNF]
Dave Crocker; Paul Overell. Augmented BNF for Syntax Specifications: ABNF. RFC. URL: http://www.ietf.org/rfc/rfc5234.txt
[BEACON]
Jatinder Mann; Alois Reitbauer. Beacon. WD. URL: http://www.w3.org/TR/beacon/
[CSS21]
Bert Bos; et al. Cascading Style Sheets Level 2 Revision 1 (CSS 2.1) Specification. 7 June 2011. REC. URL: http://www.w3.org/TR/CSS2
[ECMA-262]
Allen Wirfs-Brock. ECMA-262 6th Edition, The ECMAScript 2015 Language Specification. June 2015. Standard. URL: http://www.ecma-international.org/ecma-262/6.0/
[FETCH]
Anne van Kesteren. Fetch. Living Standard. URL: https://fetch.spec.whatwg.org/
[FIPS180]
Secure Hash Standard. URL: http://csrc.nist.gov/publications/fips/fips180-4/fips-180-4.pdf
[HTML-IMPORTS]
Dmitri Glazkov; Hajime Morrita. HTML Imports. WD. URL: http://www.w3.org/TR/html-imports/
[MANIFEST]
Marcos Caceres; Anssi Kostiainen; Kenneth Rohde Christiansen; et al. Manifest for web application. WD. URL: http://w3c.github.io/manifest/
[RFC3492]
Adam M. Costello. Punycode: A Bootstring encoding of Unicode for Internationalized Domain Names in Applications (IDNA). REC. URL: http://www.ietf.org/rfc/rfc3492.txt
[RFC3864]
Graham Klyne; Mark Nottingham; Jeffrey C. Mogul. Registration Procedures for Message Header Fields. RFC. URL: http://www.ietf.org/rfc/rfc3864.txt
[RFC6454]
Adam Barth. The Web Origin Concept. RFC. URL: http://www.ietf.org/rfc/rfc6454.txt
[RFC7034]
David Ross; Tobias Gondrom. HTTP Header Field X-Frame-Options. RFC. URL: http://www.ietf.org/rfc/rfc7034.txt
[RFC7230]
Roy T. Fielding; Julian F. Reschke. HTTP/1.1 Message Syntax and Routing. RFC. URL: http://www.ietf.org/rfc/rfc7230.txt
[RFC7231]
Roy T. Fielding; Julian F. Reschke. HTTP/1.1 Semantics and Content. RFC. URL: http://www.ietf.org/rfc/rfc7231.txt
[SVG2]
Nikos Andronikos; et al. Scalable Vector Graphics (SVG) 2. 9 July 2015. WD. URL: http://www.w3.org/TR/SVG2/
[URL]
Anne van Kesteren; Sam Ruby. URL. WD. URL: http://www.w3.org/TR/url/
[WEBIDL]
Cameron McCormack; Boris Zbarsky. Web IDL. ED. URL: https://heycam.github.io/webidl/
[CSS-IMAGES-3]
CSS Image Values and Replaced Content Module Level 3 URL: http://www.w3.org/TR/css3-images/
[CSS-IMAGES-4]
CSS Image Values and Replaced Content Module Level 4 URL: http://www.w3.org/TR/css4-images/
[CSS4-IMAGES]
Elika Etemad; Tab Atkins Jr.. CSS Image Values and Replaced Content Module Level 4. 11 September 2012. WD. URL: http://www.w3.org/TR/css4-images/
[CSSOM]
Simon Pieters; Glenn Adams. CSS Object Model (CSSOM). 5 December 2013. WD. URL: http://www.w3.org/TR/cssom/
[DOM]
Anne van Kesteren; et al. W3C DOM4. 18 June 2015. LCWD. URL: http://www.w3.org/TR/dom/
[HTML5]
Ian Hickson; et al. HTML5. 28 October 2014. REC. URL: http://www.w3.org/TR/html5/
[RFC2119]
S. Bradner. Key words for use in RFCs to Indicate Requirement Levels. March 1997. Best Current Practice. URL: https://tools.ietf.org/html/rfc2119
[RFC3986]
T. Berners-Lee; R. Fielding; L. Masinter. Uniform Resource Identifier (URI): Generic Syntax. January 2005. Internet Standard. URL: https://tools.ietf.org/html/rfc3986
[RFC5234]
D. Crocker, Ed.; P. Overell. Augmented BNF for Syntax Specifications: ABNF. January 2008. Internet Standard. URL: https://tools.ietf.org/html/rfc5234
[RFC5988]
M. Nottingham. Web Linking. October 2010. Proposed Standard. URL: https://tools.ietf.org/html/rfc5988
[RFC7159]
T. Bray, Ed.. The JavaScript Object Notation (JSON) Data Interchange Format. March 2014. Proposed Standard. URL: https://tools.ietf.org/html/rfc7159
[WORKERS]
Ian Hickson. Web Workers. 1 May 2012. CR. URL: http://www.w3.org/TR/workers/
[XML11]
Tim Bray; et al. Extensible Markup Language (XML) 1.1 (Second Edition). 16 August 2006. REC. URL: http://www.w3.org/TR/xml11/
[XSLT]
James Clark. XSL Transformations (XSLT) Version 1.0. 16 November 1999. REC. URL: http://www.w3.org/TR/xslt

Informative References

[RFC6797]
Jeff Hodges; Collin Jackson; Adam Barth. HTTP Strict Transport Security (HSTS). RFC. URL: http://www.ietf.org/rfc/rfc6797.txt
[UIREDRESS]
Giorgio Maone; et al. User Interface Security Directives for Content Security Policy. WD. URL: http://www.w3.org/TR/UISecurity/

IDL Index

partial interface HTMLScriptElement {
  attribute DOMString nonce;
};

partial interface HTMLStyleElement {
  attribute DOMString nonce;
};

[ NoInterfaceObject ]
interface SecurityPolicySource {
};

[Constructor(DOMString sourceExpression)]
interface SecurityPolicySourceURL : SecurityPolicySource {
    boolean matchesURL(DOMString url);

    readonly attribute DOMString scheme;
    readonly attribute DOMString host;
    readonly attribute DOMString port;
    readonly attribute DOMString path;

    readonly attribute boolean hostWildcard;
    readonly attribute boolean portWildcard;
};

enum SecurityPolicyHashAlgorithm { "SHA-256", "SHA-384", "SHA-512" };

[Constructor(DOMString sourceExpression)]
interface SecurityPolicySourceHash : SecurityPolicySource {
    boolean matchesNode(Node node);

    readonly attribute SecurityPolicyHashAlgorithm algorithm;
    readonly attribute DOMString hash;
};

[Constructor(DOMString sourceExpression)]
interface SecurityPolicySourceNonce : SecurityPolicySource {
    boolean matchesNode(Node node);

    readonly attribute DOMString nonce;
};

[ NoInterfaceObject ]
interface SecurityPolicyDirective {
};

[Constructor(DOMString directiveName, DOMString directiveValue)]
interface SecurityPolicySourceListDirective : SecurityPolicyDirective {
    boolean matchesURL(DOMString url);
    boolean matchesNode(Node node);

    readonly attribute boolean allowRedirects;
    readonly attribute boolean allowEval;

    // TODO: Bikeshed doesn’t like `Sequence`?
    readonly attribute SecurityPolicySource[] sources;
    readonly attribute DOMString directive;
};

enum SecurityPolicyMode {
    "EnforceMode",
    "ReportOnlyMode"
};

[Constructor(DOMString policy, SecurityPolicyMode mode)]
interface SecurityPolicy {
    // Fetch’s 'Request' object has a URL and a RequestContext, which
    // should be everything we need here to check outgoing network
    // requests. It lacks a redirect flag; not sure how to model that
    // yet. TODO.
    boolean allowRequest(Request request);

    // Inline checks can be done on a Node basis; given an HTMLScriptElement,
    // we’d skim through the `script-src` directive’s sources, checking for
    // hash and nonce matches.
    boolean allowNode(Node node);

    // One-offs. Maybe there’s a good way to fold these in somehow?
    boolean allowEval();
    boolean allowInlineEventHandler();

    // TODO: Bikeshed doesn’t like `Sequence`?
    readonly attribute SecurityPolicyDirective[] directives;

    readonly attribute SecurityPolicyMode mode;
    readonly attribute URL reportingEndpoint;
};

[Constructor(DOMString type, optional SecurityPolicyViolationEventInit eventInitDict)]
interface SecurityPolicyViolationEvent : Event {
    readonly    attribute DOMString documentURL;
    readonly    attribute DOMString referrer;
    readonly    attribute DOMString blockedURL;
    readonly    attribute DOMString violatedDirective;
    readonly    attribute DOMString effectiveDirective;
    readonly    attribute DOMString originalPolicy;
    readonly    attribute DOMString sourceFile;
    readonly    attribute DOMString statusCode;
    readonly    attribute long      lineNumber;
    readonly    attribute long      columnNumber;
};

dictionary SecurityPolicyViolationEventInit : EventInit {
    DOMString documentURL;
    DOMString referrer;
    DOMString blockedURL;
    DOMString violatedDirective;
    DOMString effectiveDirective;
    DOMString originalPolicy;
    DOMString sourceFile;
    long      lineNumber;
    long      columnNumber;
};