Permissions Policy

W3C Working Draft,

More details about this document
This version:
https://www.w3.org/TR/2024/WD-permissions-policy-1-20240925/
Latest published version:
https://www.w3.org/TR/permissions-policy/
Editor's Draft:
https://w3c.github.io/webappsec-permissions-policy/
Previous Versions:
History:
https://www.w3.org/standards/history/permissions-policy-1/
Feedback:
GitHub
Inline In Spec
Editor:
(Google)

Abstract

This specification defines a mechanism that allows developers to selectively enable and disable use of various browser features and APIs.

Status of this document

This section describes the status of this document at the time of its publication. A list of current W3C publications and the latest revision of this technical report can be found in the W3C technical reports index at https://www.w3.org/TR/.

This document was published by the Web Application Security Working Group as a Working Draft using the Recommendation track. This document is intended to become a W3C Recommendation.

The (archived) public mailing list public-webappsec@w3.org (see instructions) is preferred for discussion of this specification. When sending e-mail, please put the text “permissions-policy” in the subject, preferably like this: “[permissions-policy] …summary of comment…

Publication as a Working Draft does not imply endorsement by W3C and its Members. This is a draft document and may be updated, replaced or obsoleted by other documents at any time. It is inappropriate to cite this document as other than work in progress.

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

This document was produced by a group operating under the 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 03 November 2023 W3C Process Document.

1. Introduction

The web platform provides an ever-expanding set of features and APIs, offering richer functionality, better developer ergonomics, and improved performance. However, a missing piece is the ability for the developer to selectively enable, disable, or modify the behavior of some of these browser features and APIs within their application:

  1. The developer may want to selectively disable access to certain browser features and APIs to "lock down" their application, as a security or performance precaution, to prevent own and third-party content executing within their application from introducing unwanted or unexpected behaviors within their application.
  2. The developer may want to selectively enable access to certain browser features and APIs which may be disabled by default - e.g. some features may be disabled by default in embedded context unless explicitly enabled; some features may be subject to other policy requirements.
  3. The developer may want to use the policy to assert a promise to a client or an embedder about the use—or lack of thereof—of certain features and APIs. For example, to enable certain types of "fast path" optimizations in the browser, or to assert a promise about conformance with some requirements set by other embedders - e.g. various social networks, search engines, and so on.

This specification defines a policy mechanism that addresses the above use cases.

This specification used to be named Feature Policy.

2. Examples

SecureCorp Inc. wants to disable use of Fullscreen and Geolocation APIs within their application. It can do so by delivering the following HTTP response header to define a permissions policy:

Permissions-Policy: fullscreen=(), geolocation=()

By specifying an empty origin list, the specified features will be disabled for all documents, including nested documents, regardless of their origin.

Geolocation is disabled by default in all cross-origin frames. FastCorp Inc. has a specific cross-origin iframe on their site for which it wants to enable geolocation. It can do so by including an "allow" attribute on the iframe element:

<iframe src="https://other.com/map" allow="geolocation"></iframe>

Iframe attributes can selectively enable features in certain frames, and not in others, even if those contain documents from the same origin.

SecureCorp Inc. wants to completely disable use of the Geolocation API within all descendant navigables except for its own origin and those whose origin is "https://example.com", even in the presence of an attacker who can embed their own iframes on SecureCorp’s pages. It can do this by delivering the following HTTP response header to define a restricted permissions policy for Geolocation:

Permissions-Policy: geolocation=(self "https://example.com")

The allowlist is a list of one or more origins, which can include the application’s origin, optionally with the keyword "self", and any third-party origin.

With this policy in effect, it can then use the "allow" iframe attribute as usual to grant geolocation to certain frames, but only those frames hosting content from http://example.com or SecureCorp itself will actually be granted the ability to use that API.

SecureCorp Inc. restructured its domains and now needs to delegate use of the Geolocation API to its origin ("https://example.com") as well as three subdomains ("https://geo.example.com", "https://geo2.example.com", and "https://new.geo2.example.com"). This needs to be accomplished while still disabling the use of the Geolocation API within all other browsing contexts. It can do this by delivering the following HTTP response header:

Permissions-Policy: geolocation=(self "https://example.com" "https://geo.example.com" "https://geo2.example.com" "https://new.geo2.example.com")

This works, but if SecureCorp Inc. feels safe delegating to any subdomains on "https://example.com" the HTTP response header could instead be:

Permissions-Policy: geolocation=(self "https://example.com" "https://*.example.com")

Not only would the above header permit "https://geo.example.com", "https://geo2.example.com", and "https://new.geo2.example.com" to use the Geolocation API, but any other subdomains of "https://example.com" could use it too. Note that "https://example.com" is not covered by the allowlist entry "https://*.example.com" and must also be added.

SecureCorp Inc. restructured its services and now needs to needs to delegate use of the Geolocation API to its origin ("https://example.com") as well as three non-default ports ("https://example.com:444", "https://example.com:445", and "https://example.com:446"). This needs to be accomplished while still disabling the use of the Geolocation API within all other browsing contexts. It can do this by delivering the following HTTP response header:

Permissions-Policy: geolocation=(self "https://example.com" "https://example.com:444" "https://example.com:445" "https://example.com:446")

This works, but if SecureCorp Inc. feels safe delegating to any ports on "https://example.com" the HTTP response header could instead be:

Permissions-Policy: geolocation=(self "https://example.com:*")

Not only would the above header permit "https://example.com:444", "https://example.com:444", and "https://example.com:445" to use the Geolocation API, but any other ports on "https://example.com" could use it too.

JSPlaygroundCorp Inc. wants to host user-generated web applications, but wants the browser to manage their permissions to use powerful features in isolation of each other. This can be accomplished by creating discrete subdomains for each piece of web-content or web-content creator, and navigating them as top-level documents (framework and user-content can still be separated using same-origin iframes).

This is necessary since users grant permissions to the domain they perceive they are interacting with in the browser, which is the top-level domain.

JSPlaygroundCorp should avoid iframing user-generated web applications using the allow attribute from its own domain in this case, as this would grant its domain permissions to all of them.

PlatformCorp Inc. wants to offer a marketplace of embeddable third-party components to build from or games to play under its top-level domain. It wants to delegate the use of powerful features like the getUserMedia() API responsibly. It accepts responsibility for tracking which of its component applications need a feature, using bespoke "install" UX to keep end-users in charge.

Camera and microphone are disabled by default in all cross-origin frames. Each third-party component has a subdomain, and can be embedded in a cross-origin iframe. PlatformCorp can use the allow attribute on the iframe element to control whether to delegate camera or microphone access or not to each subdomain.

An iframe where the component "app1" should have camera access, "app2" should have microphone access, and "app3" should have both might look like this:

<iframe
  allow="camera https://app1.site.com https://app3.site.com;
         microphone https://app2.site.com https://app3.site.com"
  src="https://doc1.site.com"
  sandbox="allow-same-origin allow-scripts">
</iframe>

Iframe attributes can selectively enable features in certain frames, and not in others, even if those contain documents from the same origin. The list of sandbox tokens might be longer in practice.

Since browsers generally ask users to grant permissions to the top-level domain, there might not be any additional permission prompt for the components to request camera or microphone access if the user already trusts PlatformCorp.

[HTML5] defines a sandbox attribute for iframe elements that allows developers to reduce the risk of including potentially untrusted content by imposing restrictions on content’s abilities - e.g. prevent it from submitting forms, running scripts and plugins, and more. The sandbox directive defined by [CSP2] extends this capability to any resource, framed or not, to ask for the same set of restrictions - e.g. via an HTTP response header (Content-Security-Policy: sandbox). These mechanisms enable the developer to:

However, there are several limitations to the above mechanism: the developer cannot automatically apply a policy across all contexts, which makes it hard or impossible to enforce consistently in some cases (e.g. due to third-party content injecting frames, which the developer does not control); there is no mechanism to selectively enable features that may be off by default; the sandbox mechanism automatically disables all sandbox features, and requires the developer to opt back in to each of them, so it is impossible to extend the set of sandbox features without significant compatibility risk.

Permissions Policy is intended to be used in combination with the sandbox mechanism (i.e. it does not duplicate feature controls already covered by sandbox), and provides an extensible mechanism that addresses the above limitations.

4. Framework

4.1. Policy-controlled Features

A policy-controlled feature is an API or behaviour which can be enabled or disabled in a document by referring to it in a permissions policy.

For brevity, policy-controlled features will often be referred to in this document simply as "Features". Unless otherwise indicated, the term "feature" refers to policy-controlled features. Other specifications, defining such features, should use the longer term to avoid any ambiguity.
This spec currently only deals with features defined in Documents. We should figure out how to word this to include the possibility of features and permissions policies in Workers and Worklets as well.

Policy-controlled features are identified by tokens, which are character strings used in policy directives.

Each policy-controlled feature has a default allowlist, which defines whether that feature is available in documents in top-level traversables, and how access to that feature is inherited in child navigables.

A user agent has a set of supported features, which is the set of features which it allows to be controlled through policies. User agents are not required to support every feature.

The policy-controlled features themselves are not themselves part of this framework. A non-normative list of currently-defined features is maintained as a companion document alongside this specification.

4.2. Policies

A declared policy is a struct with the following items:

declarations

an ordered map from features to allowlists

reporting configuration

an ordered map from features to strings

A permissions policy is a struct with the following items:

inherited policy

an ordered map from features to "Enabled" or "Disabled"

declared policy

a declared policy

An empty permissions policy is a permissions policy that has an inherited policy which contains "Enabled" for every supported feature, a declared policy whose declarations and reporting configuration are both empty ordered maps.

4.3. Inherited policies

The inherited policy for a feature feature is the value in the inherited policy whose key is feature. After a permissions policy has been initialized, its inherited policy will contain a value for each supported feature.

Upon both creation and navigation, Each Document inherits a set of policies from its parent frame, or in the case of the Document in a top-level traversable, from the defined defaults for each policy-controlled feature. This inherited policy determines the initial state ("Enabled" or "Disabled") of each feature, and whether it can be controlled by a declared policy in the Document.

In a Document in a top-level traversable, the inherited policy is based on defined defaults for each feature.

In a Document in a child navigable, the inherited policy is based on the parent document’s permissions policy, as well as the child navigable's container policy.

4.4. Header policies

A header policy is a list of policy directives delivered via an HTTP header with a document. This forms the document’s permissions policy’s declared policy.

4.5. Container policies

In addition to the header policy, each child navigable has a container policy, which is a policy directive, which may be empty. The container policy can be set by attributes on the navigable container.

The container policy for a child navigable influences the inherited policy of any Document loaded into that navigable. (See § 9.7 Define an inherited policy for feature in container at origin).

Currently, the container policy cannot be set directly, but is indirectly set by the iframe allowfullscreen, and allow attributes. Future revisions to this spec may introduce a mechanism to explicitly declare the full container policy.

4.6. Policy directives

A policy directive is an ordered map, mapping policy-controlled features to corresponding allowlists of origins.

A policy directive is represented in HTTP headers as the serialization of an sf-dictionary structure, and in HTML attributes as its ASCII serialization.

4.7. Allowlists

A permissions policy allowlist is conceptually a set of origins. An allowlist may be either:

  • The special value *, which represents every origin, or
  • A struct containing:
The keywords 'self', 'src', and 'none' can appear in the text representation of allowlists in headers and attribute strings. These keywords are always interpreted in context during parsing, and only the origins which they refer to are stored in the allowlist. The keywords themselves are not part of the allowlist.

To determine whether an allowlist matches an origin origin, run these steps:

  1. If the allowlist is the special value *, then return true.

Note: We are not using the CSP variant of wildcard matching as it requires the HTTPS scheme.

  1. If the allowlist’s self-origin is not null and it is same origin-domain with origin, then return true.

  2. If the allowlist’s src-origin is not null and it is same origin-domain with origin, then return true.

  3. If origin is an opaque origin, return false.

  4. Let url be the result of calling the url parser on the serialization of origin.

  5. For each permissions-source-expression item in the allowlist’s expressions:

    1. If the result of running Does url match expression in origin with redirect count? on url, item, origin, and 0 is true then return true.

  6. Return false.

4.8. Default Allowlists

Every policy-controlled feature has a default allowlist. The default allowlist determines whether the feature is allowed in a Document with no declared policy in a top-level traversable, and also whether access to the feature is automatically delegated to documents in child navigables.

The default allowlist for a feature is one of these values:

*
The feature is allowed in Documents in top-level traversables by default, as well as those in all child navigables. It can be disallowed in child navigables by explicitly supplying a container policy on the navigable container that overrides this default (or in any navigable, by delivering the Document with a suitable Permissions-Policy header).
'self'
The feature is allowed in documents in top-level traversables by default, as well as those in child navigables whose document is same origin with its parent's document, when allowed in that Document. It is disallowed by default in child navigables whose document is cross-origin with its parent's document.

5. Permissions Policy Serialization

5.1. HTML attribute serialization

Policy Directives in HTML attributes are represented as their ASCII serialization, with the following ABNF:

serialized-permissions-policy = serialized-policy-directive *(";" serialized-policy-directive)
serialized-policy-directive = feature-identifier RWS allow-list
feature-identifier = 1*( ALPHA / DIGIT / "-")
allow-list = allow-list-value *(RWS allow-list-value)
allow-list-value = permissions-source-expression / "*" / "'self'" / "'src'" / "'none'"
permissions-source-expression = scheme-source / host-source
The string "'self'" may be used as an origin in an allowlist. When it is used in this way, it will refer to the origin of the Document which contains the permissions policy.

5.2. Structured header serialization

Policy Directives in HTTP headers are represented as Structured Fields. [RFC8941]

In this representation, a policy directive is represented by a Dictionary.

Each Dictionary Member associates a feature with an allowlist. The Member Names must be Tokens. If a token does not name one of the user agent’s supported features, then the Dictionary Member will be ignored by the processing steps.

The Member Values represent allowlists, and must be one of:

  • a String containing the ASCII permissions-source-expression

  • the Token *

  • the Token self

  • an Inner List containing zero or more of the above items.

Member Values may have a Parameter named "report-to", whose value must be a String. Any other parameters will be ignored.

Any other items inside of an Inner List will be ignored by the processing steps, and the Member Value will be processed as if they were not present. Member Values of any other form will cause the entire Dictionary Member to be ignored by the processing steps.

6. Delivery

6.1. `Permissions-Policy` HTTP Header Field

The `Permissions-Policy` HTTP header field can be used in the response (server to client) to communicate the permissions policy that should be enforced by the client.

`Permissions-Policy` is a structured header. Its value must be a dictionary. It’s ABNF is:

PermissionsPolicy = sf-dictionary

The semantics of the dictionary are defined in § 5.2 Structured header serialization.

The processing steps are defined in § 9.2 Construct policy from dictionary and origin.

6.2. The allow attribute of the iframe element

iframe elements have an allow attribute, which contains an ASCII-serialized policy directive.

The allowlist for the features named in the attribute may be empty; in that case, the default value for the allowlist is 'src', which represents the origin of the URL in the iframe’s src attribute.

When not empty, the allow attribute will result in adding an allowlist for each recognized feature to the iframe element’s content navigable's container policy, when it is constructed.

6.3. Additional attributes to support legacy features

Some features controlled by Permissions Policy have existing iframe attributes defined. This specification redefines these attributes to influence the iframe's content navigable's container policy.

6.3.1. allowfullscreen

The allowfullscreen iframe attribute controls access to requestFullscreen().

If the iframe element has an allow attribute whose value contains the token "fullscreen", then the allowfullscreen attribute must have no effect.

Otherwise, the presence of an allowfullscreen attribute on an iframe will result in adding an allowlist of * for the "fullscreen" feature to the iframe element’s content navigable's container policy, when it is constructed.

This is different from the behaviour of <iframe allow="fullscreen">, and is for compatibility with existing uses of allowfullscreen. If allow="fullscreen" and allowfullscreen are both present on an iframe element, then the more restrictive allowlist of allow="fullscreen" will be used.

7. Policy Introspection from Scripts

7.1. Overview

The current policy which is in effect in a document can be observed by scripts. This can be used to make decisions, for instance, about what user interface to display, in cases where it is not possible to determine otherwise whether a feature is enabled or not. (Some features may not have any observable failure mode, or may have unwanted side effects to feature detection.)

Documents and iframes both provide a PermissionsPolicy object which can be used to inspect the permissions policies which apply to them.

7.1.1. Document policies

To retreive the currently effective policy, use document.permissionsPolicy. This returns a PermissionsPolicy object, which can be used to:

  • query the state (allowed or denied) in the current document for a given feature,

  • get a list of all available features (allowed or not) in the current document,

  • get a list of all allowed features in the current document, or

  • get the allowlist for a given feature in the current document.

<!doctype html>
<script>
 const policy = document.permissionsPolicy;

 // This will be true if this document can use WebUSB.
 const can_use_usb = policy.allowsFeature('usb');

 // True if a new frame at https://example.com will be allowed to use WebXR.
 if (policy.allowsFeature('xr-spatial-tracking', 'https://example.com')) {
   // Show UI to create frame at https://example.com.
 } else {
   // Show an alternative UI.
 }

 // Get the list of origins which are allowed to request payment. The result
 // will be a list of explicit origins, or the single element ['*'] if all
 // origins are allowed.
 const allowed_payment_origins = policy.getAllowlistForFeature('payment');

 // Get the list of all features supported in this document (even those
 // which are not allowed). The result will be an array of strings, each
 // representing a feature.
 const all_features = policy.features();
 if (all_features.includes('geolocation')) {
   // Append a child frame to a third-party map service.
 }
</script>

7.1.2. Frame policies

It is also possible to inspect the policy on an iframe element, from the document which contains it. The policy object in this case represents the observable policy for the frame, which depends only on the current document and the attributes of the iframe element. It does not reveal whether a feature is actually currently allowed in the frame, as the document in the frame may have applied its own policy via an HTTP header, or may have navigated away from its initial location to a new origin. Revealing the effective policy in the iframe element’s nested navigable in that case could leak information about the behaviour of a cross-origin document.

<!doctype html>
<iframe id="frame" allow="fullscreen; xr-spatial-tracking"></iframe>
<script>
 const iframe_element = document.getElementById("frame");
 const iframe_policy = iframe_element.permissionsPolicy;

 // True if the framed document will be allowed to use WebXR
 if (iframe_policy.allowsFeature('xr-spatial-tracking')) {
  // display virtual reality controls
 }
</script>

The observable policy on an iframe element is independent of any actual content loaded into the frame (to avoid cross-origin information leakage,) or even whether it is in a document tree.

<!doctype html>
<!-- this frame should not be allowed to use fullscreen when the document
    in its src attribute is loaded in it -->
<iframe id="frame" allow="fullscreen https://example.com" src="https://example.net/" ></iframe>
<script>
 const iframe_element = document.getElementById("frame");
 const iframe_policy = iframe_element.permissionsPolicy;
 // This will be false, as the URL listed in the src attribute is not allowed
 // by policy to use fullscreen.
 const is_fullscreen_allowed_in_frame = iframe_policy.allowsFeature('fullscreen');

 const new_frame = document.createElement('iframe');
 new_frame.allow = 'sync-xhr';
 // This will be true, as the iframe is allowed to use sync-xhr at whatever URL is
 // mentioned in its src attribute, even though that attribute is not yet set.
 const is_sync_xhr_allowed = new_frame.permissionsPolicy.allowsFeature('sync-xhr');
</script>

7.2. The permissionsPolicy object

[Exposed=Window]
interface PermissionsPolicy {
  boolean allowsFeature(DOMString feature, optional DOMString origin);
  sequence<DOMString> features();
  sequence<DOMString> allowedFeatures();
  sequence<DOMString> getAllowlistForFeature(DOMString feature);
};

partial interface Document {
    [SameObject] readonly attribute PermissionsPolicy permissionsPolicy;
};

partial interface HTMLIFrameElement {
    [SameObject] readonly attribute PermissionsPolicy permissionsPolicy;
};

A PermissionsPolicy object has an associated node, which is a Node. The associated node is set when the PermissionsPolicy object is created.

A PermissionsPolicy object has a default origin, which is an origin, whose value depends on the state of the PermissionsPolicy object’s associated node:

Each Document has a policy object, which is a PermissionsPolicy instance whose associated node is that Document.

A Document's permissionsPolicy IDL attribute, on getting, must return the Document's policy object.

Each iframe element has a policy object, which is a PermissionsPolicy instance whose associated node is that element.

An iframe's permissionsPolicy IDL attribute, on getting, must return the iframe's policy object.

The allowsFeature(feature, origin) method must run the following steps:

  1. If origin is omitted, set origin to this PermissionsPolicy object’s default origin.

  2. Let policy be the observable policy for this PermissionsPolicy object’s associated node.

  3. If feature is allowed by policy for origin, return true.

  4. Otherwise, return false.

The features() method must run the following steps:

  1. Set result to an empty ordered set.

  2. For each supported feature feature:

    1. Append feature to result.

  3. return result

The allowedFeatures() method must run the following steps:

  1. Set result to an empty ordered set.

  2. Let origin be this PermissionsPolicy object’s default origin.

  3. Let policy be the observable policy for this PermissionsPolicy object’s associated node.

  4. For each supported feature feature:

    1. If feature is allowed by policy for origin, append feature to result.

  5. return result

The getAllowlistForFeature(feature) method must run the following steps:

  1. Set result to an empty list.

  2. Let origin be this PermissionsPolicy object’s default origin.

  3. Let policy be the observable policy for this PermissionsPolicy object’s associated node.

  4. If feature is not allowed in policy for origin, return result

  5. Let allowlist be policy’s declared policy[feature]'s declarations.

  6. If allowlist is the special value *:

    1. Append "*" to result

    2. Return result.

  7. If the allowlist’s self-origin is not null, append the serialization of it to result.

  8. If the allowlist’s src-origin is not null, append the serialization of it to result.

  9. Otherwise, for each permissions-source-expression item in allowlist’s expressions:

    1. Append item to result

  10. Return result.

The observable policy for any Node is a permissions policy, which contains the information about the policy in the navigable represented by that Node which is visible from the current document.

To get the observable policy for a Document document, return document’s permissions policy.

To get the observable policy for an Element node, run the following steps:

  1. Let inherited policy be an empty ordered map.

  2. For each supported feature feature:

    1. Let isInherited be the result of running Define an inherited policy for feature in container at origin on feature, node and node’s declared origin.

    2. Set inherited policy[feature] to isInherited.

  3. Return a new permissions policy with inherited policy inherited policy, declared policy a struct with both declarations and reporting configuration new ordered maps.

To get the declared origin for an Element node, run the following steps:

  1. If node’s node document’s sandboxed origin browsing context flag is set, then return a new opaque origin.

  2. If node’s sandbox attribute is set, and does not contain the allow-same-origin keyword, then return a new opaque origin.

  3. If node’s srcdoc attribute is set, then return node’s node document’s origin.

  4. If node’s src attribute is set:

    1. Let url be the result of parsing node’s src attribute, relative to node’s node document.

    2. If url is not failure, return url’s origin.

  5. Return node’s node document’s origin.

The declared origin concept is intended to represent the origin of the document which the embedding page intends to load into a frame. This means, for instance, that if the browser does not support the sandbox or srcdoc attributes, it should not take those attributes into account when computing the declared origin.

8. Reporting

Permissions policy violation reports indicate that some behavior of the Document has violated a permissions policy. It is up to the specification of each individual policy-controlled feature to define what it means to violate that policy, and how to determine when such a violation has occurred.

Permissions policy violation reports have the report type "permissions-policy-violation".

Permissions policy violation reports are visible to ReportingObservers.

[Exposed=Window]
interface PermissionsPolicyViolationReportBody : ReportBody {
  [Default] object toJSON();
  readonly attribute DOMString featureId;
  readonly attribute DOMString? sourceFile;
  readonly attribute long? lineNumber;
  readonly attribute long? columnNumber;
  readonly attribute DOMString disposition;
  readonly attribute DOMString? allowAttribute;
};

A permissions policy violation report’s body, represented in JavaScript by PermissionsPolicyViolationReportBody, contains the following fields:

8.1. `Permissions-Policy-Report-Only` HTTP Header Field

The `Permissions-Policy-Report-Only` HTTP header field can be used in the response (server to client) to communicate a permissions policy that should not be enforced by the client, but instead should be used to trigger reports to be sent if any policy declared within it would have been violated, had the policy been active.

`Permissions-Policy-Report-Only` is a structured header. Its value must be a dictionary.

The semantics of the dictionary are defined in § 5.2 Structured header serialization.

The processing steps are defined in § 9.2 Construct policy from dictionary and origin.

9. Algorithms

9.1. Process response policy

Given a response (response), an origin (origin), and a boolean (report-only), this algorithm returns a declared policy.
  1. Let header name be "Permissions-Policy-Report-Only" if report-only is True, or "Permissions-Policy" otherwise.

  2. Let parsed header be the result of executing get a structured field value given header name and "dictionary" from response’s header list.

  3. If parsed header is null, return an empty ordered map.

  4. Let policy be the result of executing Construct policy from dictionary and origin on parsed header and origin.

  5. Return policy.

9.2. Construct policy from dictionary and origin

Given an ordered map (dictionary) and an origin (origin), this algorithm will return a declared policy.
  1. Let declarations be an empty ordered map.

  2. Let reporting-config be an empty ordered map.

  3. For each feature-name → (value, params) of dictionary:

    1. If feature-name does not identify any recognized policy-controlled feature, then continue.

    2. Let feature be the policy-controlled feature identified by feature-name.

    3. If params["report-to"] exists, and is a string, then set reporting-config[feature] to params["report-to"].

    4. Let allowlist be a new allowlist.

    5. If value is the token *, or if value is a list which contains the token *, set allowlist to the special value *.

    6. Otherwise:

      1. If value is the token self, let allowlist’s self-origin be origin.

      2. Otherwise if value is a list, then for each element in value:

        1. If element is the token self, let allowlist’s self-origin be origin.

        2. If element is a valid permissions-source-expression, append element to allowlist’s expressions.

    7. Set declarations[feature] to allowlist.

  4. Return «declarations, reporting-config».

9.3. Parse policy directive

Given a string (value), an origin (container origin), and an optional origin (target origin), this algorithm returns a policy directive.
  1. Let directive be an empty ordered map.

  2. For each serialized-declaration returned by strictly splitting value on the delimiter U+003B (;):

    1. Let tokens be the result of splitting serialized-declaration on ASCII whitespace.

    2. If tokens is an empty list, then continue.

    3. Let feature-name be the first element of tokens.

    4. If feature-name does not identify any recognized policy-controlled feature, then continue.

    5. Let feature be the policy-controlled feature identified by feature-name.

    6. Let targetlist be the remaining elements, if any, of tokens.

    7. Let allowlist be a new allowlist.

    8. If any element of targetlist is the string "*", set allowlist to the special value *.

    9. Otherwise:

      1. If targetlist is empty and target origin is given, let allowlist’s src-origin be target origin.

      2. For each element in targetlist:

        1. If element is an ASCII case-insensitive match for "'self'":

          1. Let allowlist’s self-origin be container origin.

          2. Continue to the next element.

        2. If target origin is given, and element is an ASCII case-insensitive match for "'src'":

          1. Let allowlist’s src-origin be target origin.

          2. Continue to the next element.

        3. Let result be the result of executing the URL parser on element.

        4. If result is not failure:

          1. Let target be the origin of result.

          2. If target is not an opaque origin, append the serialization of target to allowlist’s expressions.

    10. Set directive[feature] to allowlist.

  3. Return directive

9.4. Process permissions policy attributes

Given an element (element), this algorithm returns a container policy, which may be empty.
  1. If element is not an iframe element, then return an empty policy directive.

  2. Let container policy be the result of running Parse policy directive given the value of element’s allow attribute, the origin of element’s node document, and element’s declared origin.

  3. If element’s allowfullscreen attribute is specified, and container policy does not contain an entry for the fullscreen feature.

    1. Set container policy[fullscreen] = the special value *.

  4. Return container policy.

9.5. Create a Permissions Policy for a navigable

Given null or an element (container) and an origin (origin) this algorithm returns a new Permissions Policy.
  1. Assert: If not null, container is a navigable container.

  2. Let inherited policy be a new ordered map.

  3. For each feature supported,

    1. Let isInherited be the result of running Define an inherited policy for feature in container at origin on feature, container and origin.

    2. Set inherited policy[feature] to isInherited.

  4. Let policy be a new permissions policy, with inherited policy inherited policy and declared policy «[], []».

  5. Return policy.

9.6. Create a Permissions Policy for a navigable from response

Given null or a navigable container (container), an origin (origin), a response (response), and an optional boolean (report-only), with a default value of False, this algorithm returns a new permissions policy.
  1. Let policy be the result of running Create a Permissions Policy for a navigable given container and origin.

  2. Let d be the result of running Process response policy given response, origin and report-only.

  3. For each featureallowlist of d’s declarations:

    1. If policy’s inherited policy[feature] is true, then set policy’s declared policy’s declarations[feature] to allowlist.

  4. Set policy’s declared policy[feature]'s reporting configuration to d’s reporting configuration.

  5. Return policy.

9.7. Define an inherited policy for feature in container at origin

Given a feature (feature), null or a navigable container (container), an origin for a Document in that container (origin), and an optional boolean (report-only), with a default value of False, this algorithm returns the inherited policy value for feature.
  1. If container is null, return "Enabled".

  2. If the result of executing Get feature value for origin on feature, container’s node document, container’s node document’s origin, and report-only is "Disabled", return "Disabled".

  3. If the result of executing Get feature value for origin on feature, container’s node document, origin, and report-only is "Disabled", return "Disabled".

  4. Let container policy be the result of running Process permissions policy attributes on container.

  5. If feature exists in container policy:

    1. If the allowlist for feature in container policy matches origin, return "Enabled".

    2. Otherwise return "Disabled".

  6. If feature’s default allowlist is *, return "Enabled".

  7. If feature’s default allowlist is 'self', and origin is same origin with container’s node document’s origin, return "Enabled".

  8. Otherwise return "Disabled".

9.8. Get feature value for origin

Given a feature (feature), a Document object (document), an origin (origin), and a boolean (report-only), this algorithm returns "Disabled" if feature should be considered disabled, and "Enabled" otherwise.
  1. Let policy be document’s report-only permissions policy if report-only is True, or document’s permissions policy otherwise.

  2. If policy’s inherited policy for feature is "Disabled", return "Disabled".

  3. If feature is present in policy’s declared policy:

    1. If policy’s declared policy’s declarations[feature] matches origin, then return "Enabled".

    2. Otherwise return "Disabled".

  4. Return "Enabled".

9.9. Check permissions policy

To check a permissions policy, given permissions policy (policy), a feature (feature), an origin (origin) and another origin (document origin), this algorithm returns "Disabled" if feature should be considered disabled, and "Enabled" otherwise.
  1. If policy’s inherited policy for feature is "Disabled", return "Disabled".

  2. If feature is present in policy’s declared policy:

    1. If policy’s declared policy’s declarations[feature] matches origin, then return "Enabled".

    2. Otherwise return "Disabled".

  3. If feature’s default allowlist is *, return "Enabled".

  4. If feature’s default allowlist is 'self', and origin is same origin with document origin, return "Enabled".

  5. Return "Disabled".

9.10. Is feature enabled in document for origin?

Given a feature (feature), a Document object (document), an origin (origin), and an optional boolean (report), with a default value of True, this algorithm returns "Disabled" if feature should be considered disabled, and "Enabled" otherwise. If report is True, then it will also generate and queue a report if the feature is not enabled in either document’s permissions policy or document’s report-only permissions policy.

Note: The default value of True for report means that most permissions policy checks will generate a violation report if the feature is not enabled. This is the expected result, as most checks are for an actual attempted use of the feature. If a call to this algorithm is performed just to query the state of a feature, and does not represent an actual attempt to use the feature, then report should be set to False.

  1. Let policy be document’s permissions policy.

  2. Let report-only policy be document’s report-only permissions policy.

  3. Let result be the result of calling Check permissions policy, given policy, feature, origin, and document’s origin.

  4. Let report-only result be the result of calling Check permissions policy, given report-only policy, feature, origin, and document’s origin.

  5. If report is True:

    1. Let settings be document’s environment settings object.

    2. If result is "Disabled":

      1. Let endpoint be the result of calling Get the reporting endpoint for a feature given feature and policy.

      2. Call Generate report for violation of permissions policy on settings given feature, settings, "Enforce", and endpoint.

    3. Else, if report-only result is "Disabled":

      1. Let report-only endpoint be the result of calling Get the reporting endpoint for a feature given feature and report-only policy.

      2. Call Generate report for violation of permissions policy on settings given feature, settings, "Report", and report-only endpoint.

  6. Return result

9.11. Get the reporting endpoint for a feature

Given a feature (feature) and a permissions policy (policy), this algorithm returns a string naming the endpoint to send violation reports to, or null if no such endpoint has been declared in policy.
  1. Let config be policy’s declared policy's reporting configuration.

  2. If config[feature] exists, return config[feature].

  3. Return null.

9.12. Check potential violation of permissions policy in container

Given a navigable container (container), this algorithm sends potential violation reports.
  1. Let document be container’s node document.

  2. Let settings be document’s environment settings object.

  3. For each supported feature feature:

    1. If the result of running Define an inherited policy for feature in container at origin on feature, container and container’s declared origin is "Disabled":

      1. Let endpoint be the result of calling Get the reporting endpoint for a feature given feature and document’s permissions policy.

      2. Call Generate report for potential violation of permissions policy on settings given feature, settings, "Enforce", and endpoint.

    2. Else, if the result of running Define an inherited policy for feature in container at origin on feature, container, container’s declared origin and True is "Disabled":

      1. Let report-only endpoint be the result of calling Get the reporting endpoint for a feature given feature and document’s report-only permissions policy.

      2. Call Generate report for potential violation of permissions policy on settings given feature, settings, "Report", and report-only endpoint.

9.13. Generate report for violation of permissions policy on settings

Given a feature (feature), an environment settings object (settings), a string (disposition), and a string-or-null (endpoint), this algorithm generates a report about the violation of the policy for feature.
  1. Let body be a new PermissionsPolicyViolationReportBody, initialized as follows:

    featureId

    feature’s string representation.

    sourceFile

    null

    lineNumber

    null

    columnNumber

    null

    disposition

    disposition

  2. If the user agent is currently executing script, and can extract the source file’s URL, line number, and column number from settings, then set body’s sourceFile, lineNumber, and columnNumber accordingly.

  3. Execute generate and queue a report with body, "permissions-policy-violation", endpoint, and settings.

9.14. Generate report for potential violation of permissions policy on settings

Given a feature (feature), an environment settings object (settings), a string (disposition), a string-or-null (endpoint), and a string-or-null (allowAttribute), this algorithm generates a report about the violation of the policy for feature.
  1. Let body be a new PermissionsPolicyViolationReportBody, initialized as follows:

    featureId

    feature’s string representation.

    sourceFile

    null

    lineNumber

    null

    columnNumber

    null

    disposition

    disposition

    allowAttribute

    allowAttribute

  2. If the user agent is currently executing script, and can extract the source file’s URL, line number, and column number from settings, then set body’s sourceFile, lineNumber, and columnNumber accordingly.

  3. Execute generate and queue a report with body, "potential-permissions-policy-violation", endpoint, and settings.

9.15. Should request be allowed to use feature?

Given a feature (feature) and a request (request), this algorithm returns true if the request should be allowed to use feature, and false otherwise.
  1. Set window to request’s window.

  2. If window is not a Window, return false.

    Permissions Policy within non-Window contexts (WorkerGlobalScope or WorkletGlobalScope) is being figured out in issue #207. After that’s resolved, update this algorithm to allow fetches initiated within these contexts to use policy-controlled features. Until that’s resolved, disallow all policy-controlled features (e.g., sending Client Hints to third parties) in these contexts.
  3. Set document to window’s associated Document.

  4. Let origin be request’s origin.

  5. Let result be the result of executing Is feature enabled in document for origin? on feature, document, and origin.

  6. If result is "Enabled", return true.

  7. Otherwise, return false.

10. Changes to other specifications

10.1. Changes to the HTML specification

Every Document has a report-only permissions policy, which is a permissions policy, which is initially empty.

In 7.5.1 Shared document creation infrastructure, after step 3, insert the following step:

  1. Let reportOnlyPermissionsPolicy be the result of calling Create a Permissions Policy for a navigable from response given navigationParams’s navigable’s container, navigationParams’s origin, navigationParams’s response, and True.

And in the same section, in step 10, set the new Document's report-only permissions policy to reportOnlyPermissionsPolicy.

And in the same section, in step 19 before the return, insert the following step:

  1. If navigationParams’s navigable’s container is not null, call Check potential violation of permissions policy in container given navigationParams’s navigable’s container.

11. IANA Considerations

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

Header field name
Permissions-Policy
Applicable protocol
http
Status
standard
Author/Change controller
W3C
Specification document
https://www.w3.org/TR/permissions-policy/

12. Privacy and Security

This specification standardizes a mechanism for an embedding page to set a policy which will be enforced on an embedded page. Similar to iframe sandbox, this can be done without the express permission of the embedded page, which means that behaviors of existing features can be changed in published web sites, by embedding them in another document with an appropriate container policy.

As such, the biggest privacy and security concerns are:

To a degree, these concerns are already present in the web platform, and this specification attempts to at least not make them needlessly worse.

Security and privacy issues may also be caused by the design of individual features, so care must be taken when integrating with this specification. This section attempts to provide some guidance as to what kinds of behaviors could cause such issues.

12.1. Exposure of cross-origin behavior

Features should be designed such that a violation of the policy in a framed document is not observable by documents in other frames. For instance, a hypothetical feature which caused a event to be fired in the embedding document if it is used while disabled by policy, could be used to extract information about the state of an embedded document. If the feature is known only to be used while a user is logged in to the site, for instance, then the embedder could disable that feature for the frame, and then listen for the resulting events to determine whether or not the user is logged in.

The introspection API is designed to only show information about a subframe’s policy which could already be deduced by the embedding document. This observable policy is not affected by any HTTP headers delivered with the framed document, and does not change when the frame navigates itself, even if such navigation is to a different origin, where a different policy applies. Only navigations caused by setting the src attribute of the <iframe> element will cause the observable policy to be updated.

12.2. Unanticipated behavior changes

The Permissions Policy mechanism grants a document the ability to control which features will and will not be availble in a subframe at the time it is loaded. When a feature represents an existing, long-standing behavior of the web platform, this may mean that existing published content on the web was not written with the expectation that a particular API could fail.

As a practical (though contrived) example, consider a document which uses synchronous XMLHttpRequest to determine whether a user has sufficient privileges to access the page:

<!DOCTYPE html>
<h1>Welcome to SecureCorp!</h1>
<script>
  var req = new XMLHttpRequest();
  req.open("GET", "/api/security_check.json", false);
  req.send();
  if (req.response == "untrusted user") {
    // User is not logged in; redirect to a safe page
    location.href = "/security_check_failed.html";
  }
</script>
<!-- Page continues with assumption that user is logged in -->

If this document is embedded by a page which disables the "sync-xhr" feature, the call to XMLHttpRequest.open() would fail, and the security check would be bypassed.

Note that this sort of behavior forcing is already possible on the web: some features are only allowed in top-level documents, and not in any iframes, and iframe sandboxing can be used in a similar way to embed a frame without access to features which it may be depending on.

In general, this concern is mitigated in two ways:

Authors integrating their features with Permissions Policy can decide when and how the feature will fail when a document attempts to use it while it is disabled. Authors should attempt to make use of existing failure modes, when they exist, to increase the chance that existing content will already be correctly handling such failures.

12.3. Exposure of embedding policy

Care has been taken to limit the information which an page can infer about the behavior of cross-origin pages which it embeds. It may be possible in some scenarios however, for the embedded page to infer information about its embedder, by examining the policy which the embedder has enforced on it.

This is similar to the existing document.fullscreenEnabled property, which can be used by the embedded document to infer whether its embedder has granted it the ability to use the Fullscreen API. If this is only granted in certain cases — when the user is logged in to the embedding site, for instance — then the embedded site can learn something about the state of its embedder.

13. Change log

13.1. Changes since FPWD

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.

Index

Terms defined by this specification

Terms defined by reference

References

Normative References

[CSP3]
Mike West; Antonio Sartori. Content Security Policy Level 3. 9 September 2024. WD. URL: https://www.w3.org/TR/CSP3/
[DOM]
Anne van Kesteren. DOM Standard. Living Standard. URL: https://dom.spec.whatwg.org/
[FETCH]
Anne van Kesteren. Fetch Standard. Living Standard. URL: https://fetch.spec.whatwg.org/
[FULLSCREEN]
Philip Jägenstedt. Fullscreen API Standard. Living Standard. URL: https://fullscreen.spec.whatwg.org/
[HTML]
Anne van Kesteren; et al. HTML Standard. Living Standard. URL: https://html.spec.whatwg.org/multipage/
[INFRA]
Anne van Kesteren; Domenic Denicola. Infra Standard. Living Standard. URL: https://infra.spec.whatwg.org/
[REPORTING-1]
Douglas Creager; Ian Clelland; Mike West. Reporting API. 13 August 2024. WD. URL: https://www.w3.org/TR/reporting-1/
[RFC2119]
S. Bradner. Key words for use in RFCs to Indicate Requirement Levels. March 1997. Best Current Practice. URL: https://datatracker.ietf.org/doc/html/rfc2119
[RFC3864]
G. Klyne; M. Nottingham; J. Mogul. Registration Procedures for Message Header Fields. September 2004. Best Current Practice. URL: https://www.rfc-editor.org/rfc/rfc3864
[RFC8941]
M. Nottingham; P-H. Kamp. Structured Field Values for HTTP. February 2021. Proposed Standard. URL: https://httpwg.org/specs/rfc8941.html
[URL]
Anne van Kesteren. URL Standard. Living Standard. URL: https://url.spec.whatwg.org/
[WEBIDL]
Edgar Chen; Timothy Gu. Web IDL Standard. Living Standard. URL: https://webidl.spec.whatwg.org/

Informative References

[CSP2]
Mike West; Adam Barth; Daniel Veditz. Content Security Policy Level 2. 15 December 2016. REC. URL: https://www.w3.org/TR/CSP2/
[HTML5]
Ian Hickson; et al. HTML5. 27 March 2018. REC. URL: https://www.w3.org/TR/html5/
[MEDIACAPTURE-STREAMS]
Cullen Jennings; et al. Media Capture and Streams. 27 June 2024. CR. URL: https://www.w3.org/TR/mediacapture-streams/
[PERMISSIONS]
Marcos Caceres; Mike Taylor. Permissions. 19 March 2024. WD. URL: https://www.w3.org/TR/permissions/

IDL Index

[Exposed=Window]
interface PermissionsPolicy {
  boolean allowsFeature(DOMString feature, optional DOMString origin);
  sequence<DOMString> features();
  sequence<DOMString> allowedFeatures();
  sequence<DOMString> getAllowlistForFeature(DOMString feature);
};

partial interface Document {
    [SameObject] readonly attribute PermissionsPolicy permissionsPolicy;
};

partial interface HTMLIFrameElement {
    [SameObject] readonly attribute PermissionsPolicy permissionsPolicy;
};

[Exposed=Window]
interface PermissionsPolicyViolationReportBody : ReportBody {
  [Default] object toJSON();
  readonly attribute DOMString featureId;
  readonly attribute DOMString? sourceFile;
  readonly attribute long? lineNumber;
  readonly attribute long? columnNumber;
  readonly attribute DOMString disposition;
  readonly attribute DOMString? allowAttribute;
};

Issues Index

This spec currently only deals with features defined in Documents. We should figure out how to word this to include the possibility of features and permissions policies in Workers and Worklets as well.
Permissions Policy within non-Window contexts (WorkerGlobalScope or WorkletGlobalScope) is being figured out in issue #207. After that’s resolved, update this algorithm to allow fetches initiated within these contexts to use policy-controlled features. Until that’s resolved, disallow all policy-controlled features (e.g., sending Client Hints to third parties) in these contexts.
MDN

Headers/Feature-Policy

In only one current engine.

FirefoxNoneSafariNoneChrome60+
Opera?Edge79+
Edge (Legacy)?IENone
Firefox for Android?iOS Safari?Chrome for Android?Android WebView?Samsung Internet?Opera Mobile?

Headers/Permissions-Policy

In only one current engine.

FirefoxNoneSafariNoneChrome88+
Opera?Edge88+
Edge (Legacy)?IENone
Firefox for Android?iOS Safari?Chrome for Android?Android WebView?Samsung Internet?Opera Mobile?