Trusted Types

1. Introduction

This section is not normative.

DOM-Based Cross-Site Scripting (DOM XSS) occurs when a web application takes a string value from an attacker-controlled source (e.g. the document URL parameter, or postMessage channel) and passes that value to one of the injection sinks, that eventually causes execution of the script payload controlled by the attacker.

This vulnerability type is prevalent in the web applications for two reasons. For one, it’s easy to introduce - there are over 60 different injection sinks (e.g. Element.innerHTML, or Location.href setters). A lot of those sinks are widely used, and are often passed an attacker-controlled value without the developer realizing it. Secondly, DOM XSS is difficult to prevent. Due to the dynamic nature of JavaScript it’s difficult to ascertain that this vulnerability is not present in a given program - as such, DOM XSS is often missed during manual code reviews, and automated code analysis. As an example, foo[bar] = aString is a statement that potentially introduces DOM XSS.

This document defines Trusted Types - an API that allows applications to lock down DOM XSS injection sinks to only accept non-spoofable, typed values in place of strings. These values can in turn only be created from application-defined policies, allowing the authors to define rules guarding writing values to the DOM, and reducing the DOM XSS attack surface to small, isolated parts of the web application codebase, which are substantially easier to safeguard, monitor and review.

1.1. Goals

• Minimize the likelihood of DOM XSS vulnerabilities.

• Encourage a design in which XSS-related security decisions are encapsulated within a small part of the application.

• Reduce DOM XSS security review surface for complex web application codebases.

• Allow the detection of DOM XSS vulnerabilities similar to how regular programming errors are detected and surfaced to the developers, with the assist of dynamic and static analysis tools.

1.2. Non-goals

• Prevent, or mitigate the result of injections into server-side generated markup, in specific reflections into the body of the scripts running in a document. To address server-side XSS vectors, we recommend existing solutions like templating systems or CSP script-src.

• Address resource confinement, e.g. to prevent data exfiltration, or connecting to external sources via [Fetch].

• Control subresource loading. Trusted Types aim to allow the authors to control loading resources that can script the current document, but not other subresources.

• Prevent cross-origin JavaScript execution (for example, Trusted Types don’t guard loading new documents with JavaScript code via data: URLs).

• Prevent malicious authors of the web application’s JavaScript code from being able to trigger DOM XSS; attempting to protect against malicious authors would result in an overtly complex and not-practical design; such authors already have code execution capabilities, creating DOM XSS flaws from this starting position is not a privilege escalation.

1.3. Use cases

• An author maintains a complex web application written in a framework that uses a secure templating system to generate the UI components. The application also depends on 3rd party client-side libraries that perform auxiliary tasks (e.g. analytics, performance monitoring). To ensure that none of these components introduce DOM XSS vulnerabilities, author defines a Trusted Type policy in the templating policy and enables the enforcement for the DOM sinks.

• An existing web application interacts with the DOM mostly using XSS-safe patterns (i.e. withour using injection sinks). In a few places, however, it resorts to using risky patterns like loading additional script using JSONP, calling into innerHTML or eval.

  Review finds that those places do not cause XSS (e.g. because user-controlled data is not part of the input to those sinks), but it’s hard to migrate the application off using these patterns.

  As such, CSP cannot be enforced on this application (without resorting to an unsafe version using 'unsafe-eval' 'unsafe-inline'). Additionally, it’s possible some codebase with DOM XSS flaws was not included in a review, or will be introduced in the future.

  To address this risk, the author converts the reviewed parts to using Trusted Types, and enables Trusted Type enforcement. Addditional places using the injection sinks, should they exist in the future, are correctly blocked and reported.

• A security team is tasked with assuring that the client-side heavy application code does not contain XSS vulnerabilities. Since the server side code is homogeneous (it’s moslty an API backend), and the application enforces Trusted Types, the review only focuses on the Trusted Type policies and their rules. Later on the reviewed policy names are allowed in the 'trusted-types' CSP directive, safe for the developers to use.

  Any additonal code, including the code of often-changing dependencies, can be excluded from the review, unless it creates a Trusted Type policy. Without it, the code cannot cause a DOM XSS.

2. Framework

2.1. Injection sinks

This section is not normative.

A DOM XSS injection sink is a function that evaluates an input string value in a way that could result in XSS if that value is untrusted.

Examples of injection sinks include:

• Functions that parse & insert HTML strings into the document like Element.innerHTML setter,

• Setters for Element attributes that accept a URL of the code to load like HTMLScriptElement.src,

• Setters for Element attributes that accept a code to execute like HTMLScriptElement.text,

• Functions that execute code directly like eval,

• Functions that create a new same-origin Document with caller-controlled markup like parseFromString(),

• Navigation to 'javascript:' URLs.

An application is vulnerable to DOM XSS if it permits a flow of data from an attacker-controlled source and permits that data to reach an injection sink without appropriate validation, sanitization or escaping.

It’s difficult to determine if DOM XSS is present by analyzing the invocations of injection sinks, as the strings do not carry the information about the provenance of their value. To allow the authors to control values reaching sensitive DOM and JavaScript functions, we introduce Trusted Types.

Note: The exact list of injection sinks covered by this document is defined in § 4 Integrations.

2.2. Trusted Types

We introduce the following list of Trusted Types indicating that a given value is trusted by the authors to be used with an injection sink in a certain context.

Note: Trusted in this context signifies the fact that the application author is confident that a given value can be safely used with an injection sink - she trusts it does not introduce a vulnerability. That does not imply that the value is indeed safe.

Note: This allows the authors to specify the intention when creating a given value, and the user agents to introduce checks based on the type of such value to preserve the authors' intent. For example, if authors intend a value to be used as an HTML snippet, an attempt to load a script from that value would fail.

Note: All Trusted Types wrap over an immutable string, specified when the objects are created. These objects are unforgeable in a sense that there is no JavaScript-exposed way to replace the inner string value of a given object - it’s stored in an internal slot with no setter exposed.

Note: All Trusted Types stringifiers return the inner string value. This makes it easy to incrementally migrate the application code into using Trusted Types in place of DOM strings (it’s possible to start producing types in parts of the application, while still using and accepting strings in other parts of the codebase). In that sense, Trusted Types are backwards-compatible with the regular DOM APIs.

2.2.1. TrustedHTML

The TrustedHTML interface represents a string that a developer can confidently insert into an injection sink that will render it as HTML. These objects are immutable wrappers around a string, constructed via a TrustedTypePolicy's createHTML method.

[Exposed=Window]
interface TrustedHTML {
  stringifier;
};

TrustedHTML objects have a [[Data]] internal slot which holds a DOMString. The slot’s value is set when the object is created, and will never change during its lifetime.

To stringify a TrustedHTML object, return the DOMString from its [[Data]] internal slot.

2.2.2. TrustedScript

The TrustedScript interface represents a string with an uncompiled script body that a developer can confidently pass into an injection sink that might lead to executing that script. These objects are immutable wrappers around a string, constructed via a TrustedTypePolicy's createScript method.

[Exposed=Window]
interface TrustedScript {
  stringifier;
};

TrustedScript objects have a [[Data]] internal slot which holds a DOMString. The slot’s value is set when the object is created, and will never change during its lifetime.

To stringify a TrustedScript object, return the DOMString from its [[Data]] internal slot.

2.2.3. TrustedScriptURL

The TrustedScriptURL interface represents a string that a developer can confidently pass into an injection sink that will parse it as a URL of an external script resource. These objects are immutable wrappers around a string, constructed via a TrustedTypePolicy's createScriptURL method.

[Exposed=Window]
interface TrustedScriptURL {
  stringifier;
};

TrustedScriptURL objects have a [[Data]] internal slot which holds a USVString. The slot’s value is set when the object is created, and will never change during its lifetime.

To stringify a TrustedScriptURL object, return the USVString from its [[Data]] internal slot.

2.3. Policies

Trusted Types can only be created via user-defined and immutable policies that define rules for converting a string into a given Trusted Type object. Policies allows the authors to specify custom, programmatic rules that Trusted Types must adhere to.

const sanitizingPolicy = trustedTypes.createPolicy('sanitize-html', {
  createHTML: (input) => myTrustedSanitizer(input, { superSafe: 'ok'}),
});

myDiv.innerHTML = sanitizingPolicy.createHTML(untrustedValue);

Note: Trusted Type objects wrap values that are explicitly trusted by the author. As such, creating a Trusted Type object instance becomes a de facto DOM XSS injection sink, and hence code that creates Trusted Type instances is security-critical. To allow for strict control over Trusted Type object creation we don’t expose the constructors of those directly, but require policy usage.

Multiple policies can be created in a given Realm, allowing the applications to define different rules for different parts of the codebase.

const cdnScriptsPolicy = trustedTypes.createPolicy('cdn-scripts', {
  createScriptURL(url) {
    const parsed = new URL(url, document.baseURI);
    if (parsed.origin == 'https://mycdn.example') {
      return url;
    }
    throw new TypeError('invalid URL');
  },
});

myLibrary.init({policy: cdnScriptsPolicy});

Note: As Trusted Type objects can only be created via policies, if enforcement is enabled, only the policy code can introduce a DOM XSS, and hence call-sites of the policies' create* functions are the only security-sensitive code in the entire program. Only this typically small subset of the entire code base needs to be security-reviewed for DOM XSS - there’s no need to monitor or review the injection sinks themselves, as User Agents enforce that those sinks will only accept Trusted Type objects, and these in turn can only be created via policies.

The createPolicy function returns a policy object which create* functions will create Trusted Type objects after applying the policy rules.

Note: While it’s safe to freely use a policy that sanitizes its input anywhere in the application, there might be a need to create lax policies to be used internally, and only to be called with author-controlled input. For example, a client-side HTML templating library, an HTML sanitizer library, or a JS asynchronous code plugin loading subsystem each will likely need full control over HTML or URLs. The API design facilitates that - each policy may only be used if the callsite can obtain a reference to the policy (a return value from createPolicy()). As such, policy references can be treated as capabilities, access to which can be controlled using JavaScript techniques (e.g. via closures, internal function variables, or modules).

(function renderFootnote() {
  const unsafePolicy = trustedTypes.createPolicy('html', {
    createHTML: input => input,
  });
  const footnote = await fetch('/footnote.html').then(r => r.text());
  footNote.innerHTML = unsafePolicy.createHTML(footnote);
})();

2.3.1. TrustedTypePolicyFactory

TrustedTypePolicyFactory creates policies and verifies that Trusted Type object instances were created via one of the policies.

Note: This factory object is exposed to JavaScript through window.trustedTypes reference - see § 4.1.1 Extensions to the Window interface.

[Exposed=Window] interface TrustedTypePolicyFactory {
    [Unforgeable] TrustedTypePolicy createPolicy(
        DOMString policyName, optional TrustedTypePolicyOptions policyOptions);
    [Unforgeable] sequence<DOMString> getPolicyNames();
    [Unforgeable] boolean isHTML(any value);
    [Unforgeable] boolean isScript(any value);
    [Unforgeable] boolean isScriptURL(any value);
    [Unforgeable] readonly attribute TrustedHTML emptyHTML;
    DOMString? getAttributeType(
        DOMString tagName,
        DOMString attribute,
        optional DOMString elementNs = "",
        optional DOMString attrNs = "");
    DOMString? getPropertyType(
        DOMString tagName,
        DOMString property,
        optional DOMString elementNs = "");
    readonly attribute TrustedTypePolicy? defaultPolicy;
};

Internal slot [[DefaultPolicy]] may contain a TrustedTypePolicy object, and is initially empty.

Internal slot [[CreatedPolicyNames]] is an ordered set of strings, initially empty.

anElement.innerHTML = trustedTypes.emptyHTML; // no need to create a policy

trustedTypes.defaultPolicy === null;  // true
const dp = trustedTypes.createPolicy('default', {});
trustedTypes.defaultPolicy === dp;  // true

2.3.2. TrustedTypePolicy

Policy objects implement a TrustedTypePolicy interface and define a group of functions creating Trusted Type objects. Each of the create* functions converts a string value to a given Trusted Type variant, or throws a TypeError if a conversion of a given value is disallowed.

[Exposed=Window]
interface TrustedTypePolicy {
  [Unforgeable] readonly attribute DOMString name;
  [Unforgeable] TrustedHTML createHTML(DOMString input, any... arguments);
  [Unforgeable] TrustedScript createScript(DOMString input, any... arguments);
  [Unforgeable] TrustedScriptURL createScriptURL(DOMString input, any... arguments);
};

Each policy has a name.

Each TrustedTypePolicy object has an [[options]] internal slot, holding the TrustedTypePolicyOptions object describing the actual behavior of the policy.

2.3.3. TrustedTypePolicyOptions

This dictionary holds author-defined functions for converting string values into trusted values. These functions do not create Trusted Type object instances directly - this behavior is provided by TrustedTypePolicy.

dictionary TrustedTypePolicyOptions {
   CreateHTMLCallback? createHTML;
   CreateScriptCallback? createScript;
   CreateScriptURLCallback? createScriptURL;
};
callback CreateHTMLCallback = DOMString (DOMString input, any... arguments);
callback CreateScriptCallback = DOMString (DOMString input, any... arguments);
callback CreateScriptURLCallback = USVString (DOMString input, any... arguments);

2.3.4. Default policy

This section is not normative.

One of the policies, the policy with a name "default", is special; When an injection sink is passed a string (instead of a Trusted Type object), this policy will be implicitly called by the user agent with the string value as the first argument, and the sink name as a second argument.

This allows the application to define a fallback behavior to use instead of causing a violation. The intention is to allow the applications to recover from an unexpected data flow, and sanitize the potentially attacker-controlled string "as a last resort", or reject a value if a safe value cannot be created. Errors thrown from within a policy are propagated to the application.

If the default policy doesn’t exist, or if its appropriate create* function returns null or undefined, it will cause a CSP violation. In the enforcing mode, an error will be thrown, but in report-only the original value passed to the default policy will be used.

Note: This optional behavior allows for introducing Trusted Type enforcement to applications that are still using legacy code that uses injection sinks. Needless to say, this policy should necessarily be defined with very strict rules not to introduce a DOM XSS vulnerability in unknown parts of the application. In an extreme case, a lax, no-op default policy could bring the application DOM XSS security posture back to the pre-Trusted Types level. If possible, authors should resort to a default policy in a transitional period only, use it to detect and rewrite their dependencies that use injection sinks unsafely and eventually phase out the usage of the default policy entirely.

Note: See § 3.4 Get Trusted Type compliant string for details on how the default policy is applied.

// Content-Security-Policy: trusted-types default

trustedTypes.createPolicy('default', {
  createScriptURL: (s, sink) => {
    console.log("Please refactor.");
    return s + "?default-policy-used&sink=" + encodeURIComponent(sink);
  }
});

aScriptElement.src = "https://cdn.example/script.js";
// Please refactor.
console.log(aScriptElement.src);
// https://cdn.example/script.js?default-policy-used&sink=script.src

2.4. Enforcement

Note: Enforcement is the process of checking that a value has an appropriate type before it reaches an injection sink.

The JavaScript API that allows authors to create policies and Trusted Types objects from them is always available (via trustedTypes). Since injection sinks stringify their security sensitive arguments, and Trusted Type objects stringify to their inner string values, this allows the authors to use Trusted Types in place of strings.

To prevent DOM XSS, on top of the JavaScript code using the Trusted Types, the user agent needs to enforce them i.e. assert that the injection sinks are never called with string values, and Trusted Type values are used instead. This section describes how authors may control this enforcing behavior.

2.4.1. Content Security Policy

Applications may control Trusted Type enforcement via configuring a Content Security Policy. This document defines a new trusted-types Content Security Policy directive that corresponds to Trusted Types rules.

Note: Using CSP mechanisms allows the authors to prepare their application for enforcing Trusted Types via using the Content-Security-Report-Only HTTP Response header.

Note: Most of the enforcement rules are defined as modifications of the algorithms in other specifications, see § 4 Integrations.

2.4.2. TrustedTypes extended attribute

Note: This section describes the implementation details of the Trusted Types mechanism. IDL extended attributes are not exposed to the JavasScript code.

To annotate the injection sink functions that require Trusted Types, we introduce [TrustedTypes] IDL extended attribute. Its presence indicates that the relevant construct is to be supplemented with additional Trusted Types enforcement logic.

The TrustedTypes extended attribute takes an identifier as an argument. The only valid values for the identifier are TrustedHTML, TrustedScript, or TrustedScriptURL. This extended attribute must not appear on anything other than an attribute or an operation argument. Additionally, it must not appear on read only attributes.

When the extended attribute appears on an attribute, the setter for that attribute must run the following steps in place of the ones specified in its description:

1. If context object's relevant global object has a trusted type policy factory:

   Update when workers have TT.

   1. If the context object is an Element, let objectName be the context object's local name; Otherwise, let objectName be the context object's constructor name.

   2. Set sink to the result of concatenating the list « objectName, attribute identifier. » with "." as separator.

      For example, the following annotation and JavaScript code:

      partial interface mixin HTMLScriptElement {
        [CEReactions, TrustedTypes=TrustedScriptURL] attribute ScriptURLString src;
      };
      

      document.createElement('script').src = foo;
      document.createElement('script').setAttribute('SrC', foo);
      

      causes the sink value to be "script.src".

   3. Set value to the result of running the Get Trusted Type compliant string algorithm, passing the following arguments:

      • the new value as input,

      • TrustedTypes extended attribute identifier as expectedType,

      • sink as sink,

      • context object's relevant global object as global.

   4. If an exception was thrown, rethrow exception and abort further steps.

2. Run the originally specified steps for this construct, using value as a new value to set.

Note: If the IDL attribute reflects a content attribute, identical steps should be performed when that content attribute is modified by JavaScript code e.g. via Element.setAttribute() function.

When the extended attribute appears on an operation argument, before its operation is invoked, run the following steps:

1. If context object's relevant global object has a trusted type policy factory:

   Update when workers have TT.

   1. If the context object is an Element, let objectName be the context object's local name; Otherwise, let objectName be the context object's constructor name.

   2. Set sink to the result of concatenating the list « objectName, operation identifier. » with "." as separator.

      For example, the following annotation and JavaScript code:

      partial interface Element {
        void insertAdjacentHTML(DOMString position, [TrustedTypes=TrustedHTML] HTMLString text);
      };
      

      document.createElement('a').insertAdjacentHTML('beforebegin', foo);
      

      causes the sink value to be "a.insertAdjacentHTML".

   3. Set the new argument value to the result of running the Get Trusted Type compliant string algorithm, passing the following arguments:

      • the argument value as input,

      • TrustedTypes extended attribute identifier as expectedType,

      • sink as sink,

      • context object's relevant global object as global.

   4. If an exception was thrown, rethrow exception and abort further steps.

2. Invoke the originally specified steps for the operation.

3. Algorithms

3.1. Create a Trusted Type Policy

To create a TrustedTypePolicy, given a TrustedTypePolicyFactory (factory), a string (policyName), TrustedTypePolicyOptions dictionary (options), and a global object (global) run these steps:

1. Let allowedByCSP be the result of executing Should Trusted Type policy creation be blocked by Content Security Policy? algorithm with global, policyName and factory’s [[CreatedPolicyNames]] slot value.

2. If allowedByCSP is "Blocked", throw a TypeError and abort further steps.

3. Let policy be a new TrustedTypePolicy object.

4. Set policy’s name property value to policyName.

5. Let policyOptions be a new TrustedTypePolicyOptions object.

6. Set policyOptions createHTML property to option’s createHTML property value.

7. Set policyOptions createScript property to option’s createScript property value.

8. Set policyOptions createScriptURL property to option’s createScriptURL property value.

9. Set policy’s [[options]] internal slot value to policyOptions.

10. If the policyName is default, set the factory’s [[DefaultPolicy]] slot value to policy.

11. Append policyName to factory’s [[CreatedPolicyNames]].

12. Return policy.

3.2. Get default policy

1. Return the value of factory’s [[DefaultPolicy]] slot (null if the slot is empty).

3.3. Create a Trusted Type

Given a TrustedTypePolicy policy, a type name trustedTypeName, a string value and a list arguments, execute the following steps:

1. Let functionName be a function name for the given trustedTypeName, based on the following table:

   Function name	Trusted Type name
   "createHTML"	"TrustedHTML"
   "createScript"	"TrustedScript"
   "createScriptURL"	"TrustedScriptURL"

2. Let options be the value of policy’s [[options]] slot.

3. Let function be the value of the property in options named functionName.

4. If function is null, throw a TypeError.

5. Let policyValue be the result of invoking function with value as a first argument, items of arguments as subsequent arguments, and callback **this** value set to null.

6. If policyValue is an error, return policyValue and abort the following steps.

7. If policy’s name is "default" and the policyValue is null or undefined, return policyValue.

   Note: This is used in a Get Trusted Type compliant string algorithm to signal that a value was rejected.

8. Let dataString be the result of stringifying policyValue.

9. Let trustedObject be a new instance of an interface with a type name trustedTypeName, with its [[Data]] internal slot value set to dataString.

10. Return trustedObject.

3.4. Get Trusted Type compliant string

This algorithm will return a string that can be assigned to a DOM injection sink, optionally unwrapping it from a matching Trusted Type. It will ensure that the Trusted Type enforcement rules were respected.

Given a TrustedType type (expectedType), a global object (global), TrustedType or a string (input), and a string (sink), run these steps:

1. Assert: global is a Window.

   Synchronize when TT are added to workers (perhaps use "has a CSP list"?)

2. Let cspList be the global’s CSP list.

3. If cspList does not contain a policy which directive set containing a directive with a name "trusted-types", return stringified input and abort these steps.

4. If input has type expectedType, return stringified input and abort these steps.

5. Let convertedInput be the result of executing Process value with a default policy with the same arguments as this algorithm.

6. If the algorithm threw an error, rethrow the error and abort the following steps.

7. If convertedInput is null or undefined, execute the following steps:

   1. Let disposition be the result of executing Should sink type mismatch violation be blocked by Content Security Policy? algorithm, passing global, input as source, and sink.

   2. If disposition is “Allowed”, return stringified input and abort futher steps.

      Note: This step assures that the default policy rejection will be reported, but ignored in a report-only mode.

   3. Throw a TypeError and abort further steps.

8. Assert: convertedInput has type expectedType.

9. Return stringified convertedInput.

3.5. Process value with a default policy

This algorithm routes a value to be assigned to an injection sink through a default policy, should one exist.

Given a TrustedType type (expectedType), a global object (global), TrustedType or a string (input), and a string (sink), run these steps:

1. Let defaultPolicy be the result of executing Get default policy algorithm on global’s trusted type policy factory.

2. If defaultPolicy is null, return null.

3. Let convertedInput be the result of executing Create a Trusted Type algorithm, with the following arguments:

   • defaultPolicy as policy

   • input as value

   • expectedType’s type name as trustedTypeName

   • « sink » as arguments

4. If the algorithm threw an error, rethrow it. Otherwise, return convertedInput.

4. Integrations

typedef (DOMString or TrustedHTML) HTMLString;
typedef (DOMString or TrustedScript) ScriptString;
typedef (USVString or TrustedScriptURL) ScriptURLString;
typedef (TrustedHTML or TrustedScript or TrustedScriptURL) TrustedType;

typedef (([TreatNullAs=EmptyString] DOMString) or TrustedHTML) HTMLStringDefaultsEmpty;

Note: HTMLStringDefaultsEmpty is a non-nullable variant that accepts null on set. Having this separate type allows Web IDL §3.3.21 [TreatNullAs] to attach to DOMString per heycam/webidl#441.

TreatNullAs=EmptyString is confusing. See note "It should not be used in specifications unless ...". For some sinks a null value will result in "", and "null" for others. This already caused problems in the polyfill. <https://github.com/w3c/webappsec-trusted-types/issues/2>

4.1. Integration with HTML

Window objects have a trusted type policy factory, which is a TrustedTypePolicyFactory object.

Define for workers.

4.1.1. Extensions to the Window interface

This document extends the Window interface defined by HTML:

partial interface mixin Window {
  [Unforgeable] readonly attribute
      TrustedTypePolicyFactory trustedTypes;
};

trustedTypes returns the trusted type policy factory of the current Window.

Note: The implementation in Chrome < 78 uses window.TrustedTypes instead of window.trustedTypes.

Remove the note when the API in Chrome is shipped.

4.1.2. Extensions to the Document interface

This document modifies the Document interface defined by HTML:

partial interface mixin Document {
  [CEReactions] void write([TrustedTypes=TrustedHTML] HTMLString... text);
  [CEReactions] void writeln([TrustedTypes=TrustedHTML] HTMLString... text);
};

4.1.3. Enforcement for scripts

This document modifies script elements. Each script has:

[[ScriptURL]] internal slot.

A string, containing the URL to execute the script from that was set through a TrustedTypes compliant sink. Equivalent to src attribute value. Initially null.

[[ScriptText]] internal slot.

A string, containing the body of the script to execute that was set through a TrustedTypes compliant sink. Equivalent to script’s child text content. Initially null.

The prepare a script algorithm is modified as follows:

1. If the script element is marked as having "already started", then return. The script is not executed.

2. If the element has its "parser-inserted" flag set, then set was-parser-inserted to true and unset the element’s "parser-inserted" flag. Otherwise, set was-parser-inserted to false.

   This is done so that if parser-inserted script elements fail to run when the parser tries to run them, e.g. because they are empty or specify an unsupported scripting language, another script can later mutate them and cause them to run again.

3. If was-parser-inserted is true and the element does not have an async attribute, then set the element’s "non-blocking" flag to true.

   This is done so that if a parser-inserted script element fails to run when the parser tries to run it, but it is later executed after a script dynamically updates it, it will execute in a non-blocking fashion even if the async attribute isn’t set.

4. Execute the Prepare the script URL and text algorithm upon the script element. If that algorithm threw an error, then return. The script is not executed.

5. Let source text be the element’s child text content [[ScriptText]] internal slot value .

6. Let src be the element’s [[ScriptURL]] internal slot value.

7. If the element has no src attribute src is null , and source text is the empty string, then return. The script is not executed.

8. If the element is not connected, then return. The script is not executed.

9. If either:

   • the script element has a type attribute and its value is the empty string, or
   • the script element has no type attribute but it has a language attribute and that attribute’s value is the empty string, or
   • the script element has neither a type attribute nor a language attribute, then

   ...let the script block’s type string for this script element be "text/javascript".

   Otherwise, if the script element has a type attribute, let the script block’s type string for this script element be the value of that attribute with leading and trailing ASCII whitespace stripped.

   Otherwise, the element has a non-empty language attribute; let the script block’s type string for this script element be the concatenation of the string "text/" followed by the value of the language attribute.

   The language attribute is never conforming, and is always ignored if there is a type attribute present.

   Determine the script’s type as follows:

   • If the script block’s type string is a JavaScript MIME type essence match, the script’s type is "classic".
   • If the script block’s type string is an ASCII case-insensitive match for the string "module", the script’s type is "module".
   • If neither of the above conditions are true, then return. No script is executed.

10. If was-parser-inserted is true, then flag the element as "parser-inserted" again, and set the element’s "non-blocking" flag to false.

11. Set the element’s "already started" flag.

12. If the element is flagged as "parser-inserted", but the element’s node document is not the Document of the parser that created the element, then return.

13. If scripting is disabled for the script element, then return. The script is not executed.

    The definition of scripting is disabled means that, amongst others, the following scripts will not execute: scripts in XMLHttpRequest’s responseXML documents, scripts in DOMParser-created documents, scripts in documents created by XSLTProcessor’s transformToDocument feature, and scripts that are first inserted by a script into a Document that was created using the createDocument() API. \[XHR\] \[DOMPARSING\] \[XSLTP\] \[DOM\]

14. If the script element has a nomodule content attribute and the script’s type is "classic", then return. The script is not executed.

    This means specifying nomodule on a module script has no effect; the algorithm continues onward.

15. If the script element does not have a src content attribute src is null , and the Should element’s inline behavior be blocked by Content Security Policy? algorithm returns "Blocked" when executed upon the script element, "script", and source text, then return. The script is not executed. \[CSP\]

16. If the script element has an event attribute and a for attribute, and the script’s type is "classic", then:

    1. Let for be the value of the for attribute.

    2. Let event be the value of the event attribute.

    3. Strip leading and trailing ASCII whitespace from event and for.

    4. If for is not an ASCII case-insensitive match for the string "window", then return. The script is not executed.

    5. If event is not an ASCII case-insensitive match for either the string "onload" or the string "onload()", then return. The script is not executed.

17. If the script element has a charset attribute, then let encoding be the result of getting an encoding from the value of the charset attribute.

    If the script element does not have a charset attribute, or if getting an encoding failed, let encoding be the same as the encoding of the script element’s node document.

    If the script’s type is "module", this encoding will be ignored.

18. Let classic script CORS setting be the current state of the element’s crossorigin content attribute.

19. Let module script credentials mode be the module script credentials mode for the element’s crossorigin content attribute.

20. Let cryptographic nonce be the element’s \[\[CryptographicNonce\]\] internal slot’s value.

21. If the script element has an integrity attribute, then let integrity metadata be that attribute’s value.

    Otherwise, let integrity metadata be the empty string.

22. Let referrer policy be the current state of the element’s referrerpolicy content attribute.

23. Let parser metadata be "parser-inserted" if the script element has been flagged as "parser-inserted", and "not-parser-inserted" otherwise.

24. Let options be a script fetch options whose cryptographic nonce is cryptographic nonce, integrity metadata is integrity metadata, parser metadata is parser metadata, credentials mode is module script credentials mode, and referrer policy is referrer policy.

25. Let settings object be the element’s node document’s relevant settings object.

26. If the element has a src content attribute src is not null , then:

    1. Let src be the value of the element’s src attribute.

    2. If src is the empty string, queue a task to fire an event named error at the element, and return.

    3. Set the element’s from an external file flag.

    4. Parse src relative to the element’s node document.

    5. If the previous step failed, queue a task to fire an event named error at the element, and return. Otherwise, let url be the resulting URL record.

    6. Switch on the script’s type:

       "classic"

       Fetch a classic script given url, settings object, options, classic script CORS setting, and encoding.

       "module"

       Fetch an external module script graph given url, settings object, and options.

       When the chosen algorithm asynchronously completes, set the script’s script to the result. At that time, the script is ready.

       For performance reasons, user agents may start fetching the classic script or module graph (as defined above) as soon as the src attribute is set, instead, in the hope that the element will be inserted into the document (and that the crossorigin attribute won’t change value in the meantime). Either way, once the element is inserted into the document, the load must have started as described in this step. If the UA performs such prefetching, but the element is never inserted in the document, or the src attribute is dynamically changed, or the crossorigin attribute is dynamically changed, then the user agent will not execute the script so obtained, and the fetching process will have been effectively wasted.

27. If the element does not have a src content attribute src is null , run these substeps:

    1. Let base URL be the script element’s node document’s document base URL.

    2. Switch on the script’s type:

       "classic"

       1. Let script be the result of creating a classic script using source text, settings object, base URL, and options.

       2. Set the script’s script to script.

       3. The script is ready.

       "module"

       1. Fetch an inline module script graph, given source text, base URL, settings object, and options. When this asynchronously completes, set the script’s script to the result. At that time, the script is ready.

28. Then, follow the first of the following options that describes the situation:

    If the script’s type is "classic", and the element has a src attribute, and the element has a defer attribute, and the element has been flagged as "parser-inserted", and the element does not have an async attribute

    If the script’s type is "module", and the element has been flagged as "parser-inserted", and the element does not have an async attribute

    Add the element to the end of the list of scripts that will execute when the document has finished parsing associated with the Document of the parser that created the element.

    When the script is ready, set the element’s "ready to be parser-executed" flag. The parser will handle executing the script.

    If the script’s type is "classic", and the element has a src attribute, and the element has been flagged as "parser-inserted", and the element does not have an async attribute

    The element is the pending parsing-blocking script of the Document of the parser that created the element. (There can only be one such script per Document at a time.)

    When the script is ready, set the element’s "ready to be parser-executed" flag. The parser will handle executing the script.

    If the script’s type is "classic", and the element has a src attribute, and the element does not have an async attribute, and the element does not have the "non-blocking" flag set

    If the script’s type is "module", and the element does not have an async attribute, and the element does not have the "non-blocking" flag set

    Add the element to the end of the list of scripts that will execute in order as soon as possible associated with the node document of the script element at the time the prepare a script algorithm started.

    When the script is ready, run the following steps:

    1. If the element is not now the first element in the list of scripts that will execute in order as soon as possible to which it was added above, then mark the element as ready but return without executing the script yet.

    2. Execution: Execute the script block corresponding to the first script element in this list of scripts that will execute in order as soon as possible.

    3. Remove the first element from this list of scripts that will execute in order as soon as possible.

    4. If this list of scripts that will execute in order as soon as possible is still not empty and the first entry has already been marked as ready, then jump back to the step labeled execution.

    If the script’s type is "classic", and the element has a src attribute

    If the script’s type is "module"

    The element must be added to the set of scripts that will execute as soon as possible of the node document of the script element at the time the prepare a script algorithm started.

    When the script is ready, execute the script block and then remove the element from the set of scripts that will execute as soon as possible.

    If the element does not have a src attribute, and the element has been flagged as "parser-inserted", and either the parser that created the script is an XML parser or it’s an HTML parser whose script nesting level is not greater than one, and the Document of the HTML parser or XML parser that created the script element has a style sheet that is blocking scripts

    The element is the pending parsing-blocking script of the Document of the parser that created the element. (There can only be one such script per Document at a time.)

    Set the element’s "ready to be parser-executed" flag. The parser will handle executing the script.

    Otherwise

    Immediately execute the script block, even if other scripts are already executing.

4.1.4. Enforcement in element attributes

This document modifies following IDL attributes of various DOM elements:

partial interface mixin HTMLIFrameElement : HTMLElement {
  [CEReactions, TrustedTypes=TrustedHTML] attribute HTMLString srcdoc;
};

partial interface HTMLEmbedElement : HTMLElement {
  [CEReactions, TrustedTypes=TrustedScriptURL] attribute ScriptURLString src;
};

partial interface HTMLObjectElement : HTMLElement {
  [CEReactions, TrustedTypes=TrustedScriptURL] attribute ScriptURLString data;
  [CEReactions, TrustedTypes=TrustedScriptURL] attribute DOMString codeBase; // obsolete
};

// TODO: Add HTMLPortalElement.src from https://github.com/WICG/portals once it’s specced.

Add base.href enforcement.

4.1.5. Enforcement for scripts

This document modifies how HTMLScriptElement child text content can be set to allow applications to control dynamically created scripts. It does so by adding the innerText and textContent attributes directly on HTMLScriptElement. The behavior of the attributes remains the same as in their original counterparts, apart from additional behavior triggered by the TrustedTypes extended attribute presence.

Note: Using these IDL attributes is the recommended way of dynamically setting URL or a text of a script. Manipulating attribute nodes or text nodes directly will call a default policy on the final value when the script is prepared.

Figure out what to do with script.setAttribute('src'). See DOM#789.

partial interface mixin HTMLScriptElement : HTMLElement {
 [CEReactions, TrustedTypes=TrustedScript] attribute [TreatNullAs=EmptyString] ScriptString innerText;
 [CEReactions, TrustedTypes=TrustedScript] attribute ScriptString? textContent;
 [CEReactions, TrustedTypes=TrustedScriptURL] attribute ScriptURLString src;
 [CEReactions, TrustedTypes=TrustedScript] attribute ScriptString text;
};

On setting, the innerText, textContent and text IDL attributes perform the regular steps, and then set content object's [[ScriptText]] internal slot value with the stringified value.

On setting, the src IDL attribute performs the usual steps, and then sets content object's [[ScriptURL]] internal slot value to its src content attribute value.

4.1.6. Enforcement in timer functions

This document modifies the WindowOrWorkerGlobalScope interface mixin:

typedef (ScriptString or Function) TrustedTimerHandler;

partial interface mixin WindowOrWorkerGlobalScope {
  long setTimeout(TrustedTimerHandler handler, optional long timeout = 0, any... arguments);
  long setInterval(TrustedTimerHandler handler, optional long timeout = 0, any... arguments);
};

To the timer initialization steps algorithm, add this step between 7.1 and 7.2:

1. If the first operation argument is not a Function, or if the first operation argument is a TrustedType, set the first operation argument to the result of executing the Get Trusted Type compliant string algorithm, with

   • global set to the context object's relevant global object.

   • input set to the first method argument, and

   • expectedType set to TrustedScript.

   • sink set to "Window.setInterval" if repeat is true, "Window.setTimeout" otherwise.

     Note: This matches the logic that the extended attribute would apply.

Note: This makes sure that a TrustedScript is passed to timer functions in place of a string when Trusted Types are enforced, but also unconditionally accepts any Function object.

4.1.7. Enforcement in event handler content attributes

This document modifies the attribute change steps for an event handler content attribute.

At the beginning of step 5, insert the following steps:

1. Let value be the result of executing the Get Trusted Type compliant string algorithm, with the following arguments:

   • value as input,

   • TrustedScript as expectedType,

   • sink being the result of concatenating the list « element’s local name, localName » with "." as a separator.

     Note: For example, document.createElement('div').onclick = value will result in sink being 'div.onclick'.

   • eventTarget’s relevant global object as global,

2. If the algorithm throws an error, abort these steps.

Note: This also applies to events in Scalable Vector Graphics (SVG) 2 §EventAttributes.

// Content-Security-Policy: trusted-types *

const img = document.createElement('img');
img.setAttribute('onerror', 'alert(1)'); // TypeError

4.2. Integration with SVG

This document modifies the SVGScriptElement interface to enforce Trusted Types:

partial interface mixin SVGScriptElement : SVGElement {
 // overwrites the definition in SVGURIReference.
 [TrustedTypes=TrustedScriptURL] readonly attribute ScriptURLString href;
};

4.3. Integration with DOM

4.4. Integration with DOM Parsing

This document modifies the following interfaces defined by [DOM-Parsing]:

partial interface Element {
  [CEReactions, TrustedTypes=TrustedHTML] attribute HTMLStringDefaultsEmpty outerHTML;
  [CEReactions] void insertAdjacentHTML(DOMString position, [TrustedTypes=TrustedHTML] HTMLString text);
};

partial interface mixin InnerHTML { // specified in a draft version at https://w3c.github.io/DOM-Parsing/#the-innerhtml-mixin
  [CEReactions, TrustedTypes=TrustedHTML] attribute HTMLStringDefaultsEmpty innerHTML;
};

partial interface Range {
  [CEReactions, NewObject] DocumentFragment createContextualFragment([TrustedTypes=TrustedHTML] HTMLString fragment);
};

[Constructor, Exposed=Window]
interface DOMParser {
  [NewObject] Document parseFromString([TrustedTypes=TrustedHTML] HTMLString str, SupportedType type);
};

4.5. Integration with Content Security Policy

4.5.1. trusted-types directive

This document defines trusted-types - a new Content Security Policy directive.

trusted-types directive configures the Trusted Types framework for all the injection sinks in a current realm. Specifically, it defines which policies (identified by name) can be created, and what should be the behavior when a string value is passed to an injection sink (i.e. should the type-based enforcement be enabled).

The syntax for the directive’s name and value is described by the following ABNF:

directive-name = "trusted-types"
directive-value = serialized-tt-configuration
serialized-tt-configuration = ( tt-expression *( required-ascii-whitespace tt-expression ) )
tt-expression = "*" / tt-policy-name ; In the future, add keywords
tt-policy-name = 1*( ALPHA / DIGIT / "-" / "#" / "=" / "_" / "/" / "@" / "." / "%")

Content-Security-Policy: trusted-types one two

Content-Security-Policy: trusted-types

A value "*" allows for creating policies with any name, including policies with a name that was already used.

Content-Security-Policy: trusted-types *

If the policy named default is present in the list, it refers to the default policy. All strings passed to injection sinks will be passed through it instead of being rejected outright.

Content-Security-Policy: trusted-types one two default

4.5.1.1. trusted-types Pre-Navigation check

Given a request (request), a string navigation type and a policy (policy), this algorithm returns "Blocked" if a navigation violates the trusted-types directive’s constraints and "Allowed" otherwise. This constitutes the trusted-types directive’s pre-navigation check:

Note: This algorithm assures that the code to be executed by a navigation to a javascript: URL will have to pass through a default policy’s createScript function, in addition to all other restrictions imposed by other CSP directives.

1. If request’s url's scheme is not "javascript", return "Allowed" and abort further steps.

2. Let urlString be the result of running the URL serializer on request’s url.

3. Let encodedScriptSource be the result of removing the leading "javascript:" from urlString.

4. Let defaultPolicy be the result of executing Get default policy algorithm on request’s clients's global object's trusted type policy factory.

5. If defaultPolicy is null, return "Blocked" and abort further steps.

6. Let convertedScriptSource be the result of executing Create a Trusted Type algorithm, with the following arguments:

   • defaultPolicy as policy

   • encodedScriptSource as value

   • "TrustedScript as trustedTypeName

   • « "Location.href" » as arguments

7. If convertedScriptSource is not a TrustedScript object, return "Blocked" and abort further steps.

8. Set urlString to be the result of prepending "javascript:" to stringified convertedScriptSource.

9. Let newURL be the result of running the URL parser on urlString. If the parser returns a failure, return "Blocked" and abort further steps.

10. Set request’s url to newURL.

    Note: No other CSP directives operate on javascript: URLs in a pre-navigation check. Other directives that check javascript: URLs will operate on the modified URL later, in the inline check.

11. Return "Allowed".

4.5.2. Should sink type mismatch violation be blocked by Content Security Policy?

Given a global object (global), a string (sink) and a string (source) this algorithm returns "Blocked" if the injection sink requires a Trusted Type, and "Allowed" otherwise.

1. Let result be "Allowed".

2. For each policy in global’s CSP list:

   1. If policy’s directive set does not contain a directive which name is "trusted-types", skip to the next policy.

   2. Let violation be the result of executing Create a violation object for global, policy, and directive on global, policy and "trusted-types"

   3. Set violation’s resource to "trusted-types-sink".

   4. Let trimmedSource be the substring of source, containing its first 40 characters.

   5. Set violation’s sample to be the result of concatenating the list « sink, trimmedSource « using space ("\x20") as a separator.

   6. Execute Report a violation on violation.

   7. If policy’s disposition is "enforce", then set result to "Blocked".

3. Return result.

4.5.3. Should Trusted Type policy creation be blocked by Content Security Policy?

Given a global object (global), a string (policyName) and a list of strings (createdPolicyNames), this algorithm returns "Blocked" if the Trusted Type Policy should not be created, and "Allowed" otherwise.

1. Let result be "Allowed".

2. For each policy in global’s CSP list:

   1. Let createViolation be false.

   2. If policy’s directive set does not contain a directive which name is "trusted-types", skip to the next policy.

   3. Let directive be the policy’s directive set’s directive which name is "trusted-types"

   4. If directive’s value contains a tt-expression which is a match for a value *, skip to the next policy.

      Note: trusted-types * allows authors to create policies with duplicated names.

   5. If createdPolicyNames contains policyName, set createViolation to true.

   6. If directive’s value does not contain a tt-policy-name, which value is policyName, set createViolation to true.

   7. If createViolation is false, skip to the next policy.

   8. Let violation be the result of executing Create a violation object for global, policy, and directive on global, policy and "trusted-types"

   9. Set violation’s resource to "trusted-types-policy".

   10. Set violation’s sample to the substring of policyName, containing its first 40 characters.

   11. Execute Report a violation on violation.

   12. If policy’s disposition is "enforce", then set result to "Blocked".

3. Return result.

4.5.4. Prepare the script URL and text

Given a script (script), this algorithm performs the following steps:

1. If script does not have a src content attribute, set its [[ScriptURL]] internal slot value to null.

2. Otherwise, if script’s [[ScriptURL]] internal slot value is not equal to its src attribute value, set script’s [[ScriptURL]] to the result of executing Get Trusted Type compliant string, passing TrustedScriptURL as expectedType, script’s Document's global object as global, script’s src attribute value as input and script.src as sink. If the algorithm threw an error, rethrow the error and abort further steps.

3. If script’s [[ScriptText]] internal slot value is not equal to its child text content, set script’s [[ScriptText]] to the result of executing Get Trusted Type compliant string, passing TrustedScriptURL as expectedType, script’s Document's global object as global, script’s child text content attribute value and script.text as sink. If the algorithm threw an error, rethrow the error.

4.5.5. Violation object changes

Violation object resource also allows "trusted-types-policy" and "trusted-types-sink" as values.

4.5.6. Support for eval(TrustedScript)

This document modifies the EnsureCSPDoesNotBlockStringCompilation which is reproduced in its entirety below with additions and deletions.

Note: See TC39/ecma262 issue #938 (adding the value to be compiled to algorithm parameters).

let f = new self.top.Function(source);

// Assigning a string to another Realm’s DOM sink uses that Realm’s default policy.
self.top.body.innerHTML = 'Hello, World!';
// Using another Realm’s builtin Function constructor should analogously use that
// Realm’s default policy.
new self.top.Function('alert(1)')()

Given two realms (callerRealm and calleeRealm), and a string value (source), this algorithm returns normally

1. Let sourceString be the result of executing the Get Trusted Type compliant string algorithm, with:

   • calleeRealm as global,

   • source as input,

   • "eval" as sink,

   • TrustedScript as expectedType.

2. If the algorithm throws an error, throw an EvalError.

3. Let globals be a list containing calleeRealm’s global object and calleeRealm’s global object.

4. For each global in globals:

   1. Let result be "Allowed".

   2. For each policy in global’s CSP list:

      1. Let source-list be null.

      2. If policy contains a directive whose name is "script-src", then set source-list to that directive's value.

         Otherwise if policy contains a directive whose name is "default-src", then set source-list to that directive’s value.

      3. If source-list is not null, and does not contain a source expression which is an ASCII case-insensitive match for the string "'unsafe-eval'" then:

         1. Let violation be the result of executing Content Security Policy Level 3 §create-violation-for-global on global, policy, and "script-src".

         2. Set violation’s resource to "inline".

         3. If source-list contains the expression "'report-sample'", then set violation’s sample to the substring of source sourceString containing its first 40 characters.

         4. Execute Content Security Policy Level 3 §report-violation on violation.

         5. If policy’s disposition is "enforce", then set result to "Blocked".

   3. If result is "Blocked", throw an EvalError exception.

5. Return sourceString.

Note: returning sourceString means that the string that gets compiled is that returned by any default policy in the course of executing Get Trusted Type compliant string.

This depends on a solution to issue #144 like TC39 HostBeforeCompile

In some cases, the violation "'report-sample'" contain the result of applying the default policy to a string argument which differs. Specifically when, there is a default policy, isExempt is false, and source there is a CSP policy for either the callerRealm or callerRealm that disallows "'unsafe-eval'". Is this a feature or a bug?

Note: The previous algorithm reports violations via both report-uris where callerRealm != calleeRealm. If Get Trusted Type compliant string reports an error, it only reports it via its calleeRealm’s report-uri.

5. Security Considerations

Trusted Types are not intended to defend against XSS in an actively malicious execution environment. It’s assumed that the application is written by non-malicious authors; the intent is to prevent developer mistakes that could result in security bugs, and not to defend against first-party malicious code actively trying to bypass policy restrictions. Below we enumerate already identified vectors that remain risky even in environments with enforced Trusted Types.

5.1. Cross-document vectors

While the code running in a window in which Trusted Types are enforced cannot dynamically create nodes that would bypass the policy restrictions, it is possible that such nodes can be imported or adopted from documents in other windows, that don’t have the same set of restrictions. In essence - it is possible to bypass Trusted Types if a malicious author creates a setup in which a restricted document colludes with an unrestricted one.

CSP propagation rules (see Content Security Policy Level 3 §initialize-document-csp partially address this issue, as new local scheme documents will inherit the same set of restrictions. To address this issue comprehensively, other mechanisms like Origin Policy should be used to ensure that baseline security rules are applied for the whole origin.

5.2. Deprecated features

Some long-deprecated and rarely used plaform features are not subject to Trusted Types, and could potentially be used by malicious authors to overcome the restrictions:

• HTML imports

5.3. Bypass vectors

Mention anchor element properties bypass. <https://github.com/w3c/webappsec-trusted-types/issues/64>

Mention text/attribute node copy bypass vectors. <https://github.com/w3c/webappsec-trusted-types/issues/47>

5.4. Best practices for policy design

Trusted Types limit the scope of the code that can introduce DOM XSS vulnerabilities to the implementation of policies. In this design, insecure policies can still enable XSS. Special emphasis needs to be taken by use policies that are either secure for all possible inputs, or limit the access to insecure policies, such that they are only called with non-attacker controlled inputs.

As policies are custom JavaScript code, they may be written in a way that heavily depends on a global state. We advise against this. The policies should be self-contained as much as possible. All objects that may alter security decisions a policy makes effectively become the policy, and should be guarded & reviewed together.

Refer to the external document on secure policy design.

6. Implementation Considerations

6.1. Vendor-specific Extensions and Addons

Restriction imposed by Trusted Types SHOULD NOT interfere with the operation of user-agent features like addons, extensions, or bookmarklets. These kinds of features generally advance the user’s priority over page authors, as espoused in [html-design-principles]. Specifically, extensions SHOULD be able to pass strings to the DOM XSS injection sinks without triggering default policy execution, violation generation, or the rejection of the value.