dompurify 3.4.6 → 3.4.8

package/dist/purify.js

@@ -1,4 +1,4 @@
-/*! @license DOMPurify 3.4.6 | (c) Cure53 and other contributors | Released under the Apache license 2.0 and Mozilla Public License 2.0 | github.com/cure53/DOMPurify/blob/3.4.6/LICENSE */
+/*! @license DOMPurify 3.4.8 | (c) Cure53 and other contributors | Released under the Apache license 2.0 and Mozilla Public License 2.0 | github.com/cure53/DOMPurify/blob/3.4.8/LICENSE */
 
 (function (global, factory) {
   typeof exports === 'object' && typeof module !== 'undefined' ? module.exports = factory() :
@@ -409,7 +409,7 @@
   function createDOMPurify() {
     let window = arguments.length > 0 && arguments[0] !== undefined ? arguments[0] : getGlobal();
     const DOMPurify = root => createDOMPurify(root);
-    DOMPurify.version = '3.4.6';
+    DOMPurify.version = '3.4.8';
     DOMPurify.removed = [];
     if (!window || !window.document || window.document.nodeType !== NODE_TYPE.document || !window.Element) {
       // Not running in a browser, provide a factory function
@@ -454,6 +454,23 @@
     }
     let trustedTypesPolicy;
     let emptyHTML = '';
+    // Tracks whether we are already inside a call to the configured Trusted Types
+    // policy's `createHTML`. If the supplied `TRUSTED_TYPES_POLICY.createHTML`
+    // itself calls `DOMPurify.sanitize` (the cause of #1422), `sanitize` would
+    // re-enter the policy and recurse until the stack overflows. We detect that
+    // re-entry and throw a clear, actionable error instead.
+    let IN_POLICY_CREATE_HTML = 0;
+    const _createTrustedHTML = function _createTrustedHTML(html) {
+      if (IN_POLICY_CREATE_HTML > 0) {
+        throw typeErrorCreate('The configured TRUSTED_TYPES_POLICY.createHTML must not call ' + 'DOMPurify.sanitize, as that causes infinite recursion. Do not pass ' + 'a policy whose createHTML wraps DOMPurify as TRUSTED_TYPES_POLICY; ' + 'see the "DOMPurify and Trusted Types" section of the README.');
+      }
+      IN_POLICY_CREATE_HTML++;
+      try {
+        return trustedTypesPolicy.createHTML(html);
+      } finally {
+        IN_POLICY_CREATE_HTML--;
+      }
+    };
     const _document = document,
       implementation = _document.implementation,
       createNodeIterator = _document.createNodeIterator,
@@ -781,19 +798,47 @@
           throw typeErrorCreate('TRUSTED_TYPES_POLICY configuration option must provide a "createScriptURL" hook.');
         }
         // Overwrite existing TrustedTypes policy.
+        const previousTrustedTypesPolicy = trustedTypesPolicy;
         trustedTypesPolicy = cfg.TRUSTED_TYPES_POLICY;
-        // Sign local variables required by `sanitize`.
-        emptyHTML = trustedTypesPolicy.createHTML('');
+        // Sign local variables required by `sanitize`. If the supplied policy's
+        // `createHTML` is circular (i.e. it calls `DOMPurify.sanitize`), this
+        // throws via the re-entrancy guard. Restore the previous policy first so
+        // the instance is not left in a poisoned state. See #1422.
+        try {
+          emptyHTML = _createTrustedHTML('');
+        } catch (error) {
+          trustedTypesPolicy = previousTrustedTypesPolicy;
+          throw error;
+        }
       } else {
         // Uninitialized policy, attempt to initialize the internal dompurify policy.
-        if (trustedTypesPolicy === undefined) {
+        if (trustedTypesPolicy === undefined && cfg.TRUSTED_TYPES_POLICY !== null) {
           trustedTypesPolicy = _createTrustedTypesPolicy(trustedTypes, currentScript);
         }
         // If creating the internal policy succeeded sign internal variables.
-        if (trustedTypesPolicy !== null && typeof emptyHTML === 'string') {
-          emptyHTML = trustedTypesPolicy.createHTML('');
+        // Note: a falsy `trustedTypesPolicy` (null when policy creation failed or
+        // was skipped via `TRUSTED_TYPES_POLICY: null`, or undefined when no
+        // policy has been initialized yet) must be excluded here, otherwise we
+        // would call `.createHTML` on a non-policy and throw. See #1422.
+        if (trustedTypesPolicy && typeof emptyHTML === 'string') {
+          emptyHTML = _createTrustedHTML('');
         }
       }
+      /*
+       * Mirror the clone-before-mutate pattern already applied above for
+       * cfg.ADD_TAGS / cfg.ADD_ATTR: if any uponSanitize* hook is
+       * registered AND the set still points at the default constant,
+       * clone it. The hook then mutates the clone (in-call widening
+       * still works exactly as documented) and the next default-cfg
+       * call rebinds to the untouched original via the reassignment at
+       * the top of this function.
+       */
+      if ((hooks.uponSanitizeElement.length > 0 || hooks.uponSanitizeAttribute.length > 0) && ALLOWED_TAGS === DEFAULT_ALLOWED_TAGS) {
+        ALLOWED_TAGS = clone(ALLOWED_TAGS);
+      }
+      if (hooks.uponSanitizeAttribute.length > 0 && ALLOWED_ATTR === DEFAULT_ALLOWED_ATTR) {
+        ALLOWED_ATTR = clone(ALLOWED_ATTR);
+      }
       // Prevent further manipulation of configuration.
       // Not available in IE8, Safari 5, etc.
       if (freeze) {
@@ -953,7 +998,7 @@
         // Root of XHTML doc must contain xmlns declaration (see https://www.w3.org/TR/xhtml1/normative.html#strict)
         dirty = '<html xmlns="http://www.w3.org/1999/xhtml"><head></head><body>' + dirty + '</body></html>';
       }
-      const dirtyPayload = trustedTypesPolicy ? trustedTypesPolicy.createHTML(dirty) : dirty;
+      const dirtyPayload = trustedTypesPolicy ? _createTrustedHTML(dirty) : dirty;
       /*
        * Use the DOMParser API by default, fallback later if needs be
        * DOMParser not work for svg when has multiple root element.
@@ -1012,7 +1057,8 @@
      *
      * @param node The root element whose character data should be scrubbed.
      */
-    const _scrubTemplateExpressions = function _scrubTemplateExpressions(node) {
+    const _scrubTemplateExpressions2 = function _scrubTemplateExpressions(node) {
+      var _node$querySelectorAl, _node$querySelectorAl2;
       node.normalize();
       const walker = createNodeIterator.call(node.ownerDocument || node, node,
       // eslint-disable-next-line no-bitwise
@@ -1026,6 +1072,15 @@
         currentNode.data = data;
         currentNode = walker.nextNode();
       }
+      // NodeIterator does not descend into <template>.content per the DOM spec,
+      // so we must explicitly recurse into each template's content fragment,
+      // mirroring the approach used by _sanitizeShadowDOM.
+      const templates = (_node$querySelectorAl = (_node$querySelectorAl2 = node.querySelectorAll) === null || _node$querySelectorAl2 === void 0 ? void 0 : _node$querySelectorAl2.call(node, 'template')) !== null && _node$querySelectorAl !== void 0 ? _node$querySelectorAl : [];
+      arrayForEach(Array.from(templates), tmpl => {
+        if (_isDocumentFragment(tmpl.content)) {
+          _scrubTemplateExpressions2(tmpl.content);
+        }
+      });
     };
     /**
      * _isClobbered
@@ -1036,32 +1091,6 @@
      * on direct reads. We use this check at the IN_PLACE entry-point and
      * during attribute sanitization to refuse clobbered forms.
      *
-     * Realm safety (GHSA-hpcv-96wg-7vj8): every check in this function must
-     * work for foreign-realm forms — e.g. a <form> created inside a same-
-     * origin iframe and then handed to a parent-realm DOMPurify instance
-     * with IN_PLACE: true. The original implementation used
-     * `element instanceof HTMLFormElement` and `element.attributes
-     * instanceof NamedNodeMap`, both of which are realm-bound: a foreign-
-     * realm form is an instance of the *foreign* realm's HTMLFormElement,
-     * not the parent realm's. The instanceof short-circuited to false and
-     * the function returned false (= not clobbered) regardless of how
-     * thoroughly the form was clobbered. Sanitize then walked a clobbered
-     * .attributes and missed every attribute on the form root, leaving
-     * onmouseover / onclick / formaction / etc. intact.
-     *
-     * The realm-independent replacements:
-     *   - HTMLFormElement detection — read the tag name through the cached
-     *     Node.prototype.nodeName getter. WebIDL getters operate on internal
-     *     slots that exist on every real Node regardless of which realm
-     *     minted the JS wrapper, so getNodeName(foreignForm) === "FORM".
-     *   - NamedNodeMap detection — compare the direct .attributes read
-     *     against the cached Element.prototype.attributes getter. Same
-     *     equality-probe pattern we use for .childNodes: if a clobbering
-     *     child shadows the named property, the two reads diverge; if not,
-     *     both return the same NamedNodeMap (same-realm OR foreign-realm —
-     *     doesn't matter, both are the canonical attributes object for the
-     *     node).
-     *
      * @param element element to check for clobbering attacks
      * @return true if clobbered, false if safe
      */
@@ -1083,6 +1112,14 @@
       // (same-realm OR foreign-realm) has both reads pointing at the same
       // canonical NamedNodeMap.
       element.attributes !== getAttributes(element) || typeof element.removeAttribute !== 'function' || typeof element.setAttribute !== 'function' || typeof element.namespaceURI !== 'string' || typeof element.insertBefore !== 'function' || typeof element.hasChildNodes !== 'function' ||
+      // NodeType clobbering probe. Cached Node.prototype.nodeType getter
+      // returns the integer 1 for any Element regardless of realm; direct
+      // read on a clobbered form (e.g. <input name="nodeType">) returns
+      // the named child element. Cheap addition — nodeType is read from
+      // an internal slot, no serialization cost — and removes a residual
+      // clobbering surface used by several mXSS / PI / comment branches
+      // in _sanitizeElements that compare currentNode.nodeType directly.
+      element.nodeType !== getNodeType(element) ||
       // HTMLFormElement has [LegacyOverrideBuiltIns]: a descendant named
       // "childNodes" shadows the prototype getter. Direct reads of
       // form.childNodes from a clobbered form return the named child
@@ -1099,14 +1136,6 @@
     /**
      * Checks whether the given value is a DocumentFragment from any realm.
      *
-     * Realm safety (GHSA-hpcv-96wg-7vj8): the original sites used
-     * `value instanceof DocumentFragment`, which is realm-bound — a fragment
-     * from a foreign realm (template content or shadow root from an iframe
-     * document) is an instance of the foreign realm's DocumentFragment, not
-     * the parent realm's, so the check returned false and the template-
-     * content / shadow-root recursion was silently skipped. The attacker
-     * payload inside survived untouched.
-     *
      * The realm-independent replacement reads `nodeType` through the cached
      * Node.prototype getter and compares to the DOCUMENT_FRAGMENT_NODE
      * constant (11). nodeType is a numeric value resolved from the node's
@@ -1133,12 +1162,6 @@
      * sanitize() to silently stringify them and reset IN_PLACE to false,
      * returning the original node unsanitized. See GHSA-4w3q-35jp-p934.
      *
-     * Implementation: call the cached `nodeType` getter from Node.prototype
-     * directly on the value. This bypasses any clobbered instance property
-     * (e.g. a child element named "nodeType") and works across realms
-     * because the WebIDL `nodeType` getter reads an internal slot that
-     * every real Node has, regardless of which window minted it.
-     *
      * @param value object to check whether it's a DOM node
      * @return true if value is a DOM node from any realm
      */
@@ -1176,7 +1199,7 @@
         return true;
       }
       /* Now let's check the element's type and name */
-      const tagName = transformCaseFunc(currentNode.nodeName);
+      const tagName = transformCaseFunc(getNodeName ? getNodeName(currentNode) : currentNode.nodeName);
       /* Execute a hook if present */
       _executeHooks(hooks.uponSanitizeElement, currentNode, {
         tagName,
@@ -1213,10 +1236,17 @@
             return false;
           }
         }
-        /* Keep content except for bad-listed elements */
+        /* Keep content except for bad-listed elements.
+           Use the cached prototype getters exclusively — the previous code
+           had `|| currentNode.parentNode` / `|| currentNode.childNodes`
+           fallbacks, but the cached getters always return the canonical
+           value (or null for a real parent-less node), so the fallback
+           path was dead in safe cases and a clobbering surface in unsafe
+           ones. Falsy cached results stay falsy; the `if (childNodes &&
+           parentNode)` check already gates correctly. */
         if (KEEP_CONTENT && !FORBID_CONTENTS[tagName]) {
-          const parentNode = getParentNode(currentNode) || currentNode.parentNode;
-          const childNodes = getChildNodes(currentNode) || currentNode.childNodes;
+          const parentNode = getParentNode(currentNode);
+          const childNodes = getChildNodes(currentNode);
           if (childNodes && parentNode) {
             const childCount = childNodes.length;
             for (let i = childCount - 1; i >= 0; --i) {
@@ -1412,7 +1442,7 @@
             switch (trustedTypes.getAttributeType(lcTag, lcName)) {
               case 'TrustedHTML':
                 {
-                  value = trustedTypesPolicy.createHTML(value);
+                  value = _createTrustedHTML(value);
                   break;
                 }
               case 'TrustedScriptURL':
@@ -1469,6 +1499,24 @@
         if (_isDocumentFragment(shadowNode.content)) {
           _sanitizeShadowDOM2(shadowNode.content);
         }
+        /* An element iterated here may itself host an attached
+           shadow root. The default NodeIterator does not enter shadow
+           trees, so a shadow root nested inside template.content was
+           previously reached by no walk at all (the pre-pass at
+           _sanitizeAttachedShadowRoots descends via childNodes, which
+           doesn't enter template.content; the template-content recursion
+           above iterates the content but never inspected shadowRoot).
+           Walk it explicitly. The nodeType guard avoids reading
+           shadowRoot off text / comment / CDATA / PI nodes that the
+           iterator also surfaces. */
+        const shadowNodeType = getNodeType ? getNodeType(shadowNode) : shadowNode.nodeType;
+        if (shadowNodeType === NODE_TYPE.element) {
+          const innerSr = getShadowRoot ? getShadowRoot(shadowNode) : shadowNode.shadowRoot;
+          if (_isDocumentFragment(innerSr)) {
+            _sanitizeAttachedShadowRoots2(innerSr);
+            _sanitizeShadowDOM2(innerSr);
+          }
+        }
       }
       /* Execute a hook if present */
       _executeHooks(hooks.afterSanitizeShadowDOM, fragment, null);
@@ -1490,17 +1538,6 @@
      * existing _sanitizeShadowDOM template-content recursion) stay
      * untouched — string-input paths are not affected.
      *
-     * DOM-Clobbering hardening: HTMLFormElement carries the WebIDL
-     * [LegacyOverrideBuiltIns] extended attribute, so a descendant element
-     * named `nodeType`, `shadowRoot`, or `childNodes` shadows the matching
-     * prototype getter on the form. Reading those properties directly off
-     * the node would let an attacker steer this walk past shadow hosts
-     * (e.g. <input name="childNodes"> collapses the form's child list to
-     * the input itself, so descent stops dead and any shadow root deeper
-     * in the subtree is never sanitized). Every property access here is
-     * therefore routed through the cached prototype getter; the form's
-     * named-property getter cannot intercept those reads.
-     *
      * @param root the subtree root to walk for attached shadow roots
      */
     const _sanitizeAttachedShadowRoots2 = function _sanitizeAttachedShadowRoots(root) {
@@ -1536,6 +1573,16 @@
       for (const child of snapshot) {
         _sanitizeAttachedShadowRoots2(child);
       }
+      /* When the root is a <template>, also descend into root.content */
+      if (nodeType === NODE_TYPE.element) {
+        const rootName = getNodeName ? getNodeName(root) : null;
+        if (typeof rootName === 'string' && transformCaseFunc(rootName) === 'template') {
+          const content = root.content;
+          if (_isDocumentFragment(content)) {
+            _sanitizeAttachedShadowRoots2(content);
+          }
+        }
+      }
     };
     // eslint-disable-next-line complexity
     DOMPurify.sanitize = function (dirty) {
@@ -1626,7 +1673,7 @@
         if (!RETURN_DOM && !SAFE_FOR_TEMPLATES && !WHOLE_DOCUMENT &&
         // eslint-disable-next-line unicorn/prefer-includes
         dirty.indexOf('<') === -1) {
-          return trustedTypesPolicy && RETURN_TRUSTED_TYPE ? trustedTypesPolicy.createHTML(dirty) : dirty;
+          return trustedTypesPolicy && RETURN_TRUSTED_TYPE ? _createTrustedHTML(dirty) : dirty;
         }
         /* Initialize the document to work on */
         body = _initDocument(dirty);
@@ -1658,14 +1705,14 @@
       /* If we sanitized `dirty` in-place, return it. */
       if (IN_PLACE) {
         if (SAFE_FOR_TEMPLATES) {
-          _scrubTemplateExpressions(dirty);
+          _scrubTemplateExpressions2(dirty);
         }
         return dirty;
       }
       /* Return sanitized string or DOM */
       if (RETURN_DOM) {
         if (SAFE_FOR_TEMPLATES) {
-          _scrubTemplateExpressions(body);
+          _scrubTemplateExpressions2(body);
         }
         if (RETURN_DOM_FRAGMENT) {
           returnNode = createDocumentFragment.call(body.ownerDocument);
@@ -1699,7 +1746,7 @@
           serializedHTML = stringReplace(serializedHTML, expr, ' ');
         });
       }
-      return trustedTypesPolicy && RETURN_TRUSTED_TYPE ? trustedTypesPolicy.createHTML(serializedHTML) : serializedHTML;
+      return trustedTypesPolicy && RETURN_TRUSTED_TYPE ? _createTrustedHTML(serializedHTML) : serializedHTML;
     };
     DOMPurify.setConfig = function () {
       let cfg = arguments.length > 0 && arguments[0] !== undefined ? arguments[0] : {};