xmldom: XML injection via unsafe CDATA serialization allows attacker-controlled markup insertion

## Summary `@xmldom/xmldom` allows attacker-controlled strings containing the CDATA terminator `]]>` to be inserted into a `CDATASection` node. During serialization, `XMLSerializer` emitted the CDATA content verbatim without rejecting or safely splitting the terminator. As a result, data intended to remain text-only became **active XML markup** in the serialized output, enabling XML structure injection and downstream business-logic manipulation. The sequence `]]>` is not allowed inside CDATA content and must be rejected or safely handled during serialization. ([MDN Web Docs](https://developer.mozilla.org/)) ### Attack surface `Document.createCDATASection(data)` is the most direct entry point, but it is not the only one. The WHATWG DOM spec intentionally does not validate `]]>` in mutation methods — only `createCDATASection` carries that guard. The following paths therefore also allow `]]>` to enter a CDATASection node and reach the serializer: - `CharacterData.appendData()` - `CharacterData.replaceData()` - `CharacterData.insertData()` - Direct assignment to `.data` - Direct assignment to `.textContent` (Note: assigning to `.nodeValue` does **not** update `.data` in this implementation — the serializer reads `.data` directly — so `.nodeValue` is not an exploitable path.) ### Parse path Parsing XML that contains a CDATA section is **not** affected. The SAX parser's non-greedy `CDSect` regex stops at the first `]]>`, so parsed CDATA data never contains the terminator. --- ## Impact If an application uses `xmldom` to generate "trusted" XML documents that embed **untrusted user input** inside CDATA (a common pattern in exports, feeds, SOAP/XML integrations, etc.), an attacker can inject additional XML elements/attributes into the generated document. This can lead to: - Integrity violation of generated XML documents. - Business-logic injection in downstream consumers (e.g., injecting `<approved>true</approved>`, `<role>admin</role>`, workflow flags, or other security-relevant elements). - Unexpected privilege/workflow decisions if downstream logic assumes injected nodes cannot appear. This issue does **not** require malformed parsers or browser behavior; it is caused by serialization producing attacker-influenced XML markup. --- ## Root Cause (with file + line numbers) **File:** `lib/dom.js` ### 1. No validation in `createCDATASection` `createCDATASection: function (data)` accepts any string and appends it directly. - **Lines 2216–2221** (0.9.8) ### 2. Unsafe CDATA serialization Serializer prints CDATA sections as: ``` <![CDATA[ + node.data + ]]> ``` without handling `]]>` in the data. - **Lines 2919–2920** (0.9.8) Because CDATA content is emitted verbatim, an embedded `]]>` closes the CDATA section early and the remainder of the attacker-controlled payload is interpreted as markup in the serialized XML. --- ## Proof of Concept — Fix A: `createCDATASection` now throws On patched versions, passing `]]>` directly to `createCDATASection` throws `InvalidCharacterError` instead of silently accepting the payload: ```js const { DOMImplementation } = require('./lib'); const doc = new DOMImplementation().createDocument(null, 'root', null); try { doc.createCDATASection('SAFE]]><injected attr="pwn"/>'); console.log('VULNERABLE — no error thrown'); } catch (e) { console.log('FIXED — threw:', e.name); // InvalidCharacterError } ``` Expected output on patched versions: ``` FIXED — threw: InvalidCharacterError ``` --- ## Proof of Concept — Fix B: mutation vector now safe On patched versions, injecting `]]>` via a mutation method (`appendData`, `replaceData`, `.data =`, `.textContent =`) no longer produces injectable output. The serializer splits the terminator so the result round-trips as safe text: ```js const { DOMImplementation, XMLSerializer } = require('./lib'); const { DOMParser } = require('./lib'); const doc = new DOMImplementation().createDocument(null, 'root', null); // Start with safe data, then mutate to include the terminator const cdata = doc.createCDATASection('safe'); doc.documentElement.appendChild(cdata); cdata.appendData(']]><injected attr="pwn"/><more>TEXT</more><![CDATA['); const out = new XMLSerializer().serializeToString(doc); console.log('Serialized:', out); const reparsed = new DOMParser().parseFromString(out, 'text/xml'); const injected = reparsed.getElementsByTagName('injected').length > 0; console.log('Injected element found in reparsed doc:', injected); // VULNERABLE: true | FIXED: false ``` Expected output on patched versions: ``` Serialized: <root><![CDATA[safe]]]]><![CDATA[><injected attr="pwn"/><more>TEXT</more><![CDATA[]]></root> Injected element found in reparsed doc: false ``` --- ## Fix Applied Both mitigations were implemented: ### Option A — Strict/spec-aligned: reject `]]>` in `createCDATASection()` `Document.createCDATASection(data)` now throws `InvalidCharacterError` (per the [WHATWG DOM spec](https://dom.spec.whatwg.org/#dom-document-createcdatasection)) when `data` contains `]]>`. This closes the direct entry point. Code that previously passed a string containing `]]>` to `createCDATASection` and relied on the silent/unsafe behaviour will now receive `InvalidCharacterError`. Use a mutation method such as `appendData` if you intentionally need `]]>` in a CDATASection node's data (the serializer split in Option B will keep the output safe). ### Option B — Defensive serialization: split the terminator during serialization `XMLSerializer` now replaces every occurrence of `]]>` in CDATA section data with the split sequence `]]]]><![CDATA[>` before emitting. This closes all mutation-vector paths that Option A alone cannot guard, and means the serialized output is always well-formed XML regardless of how `]]>` entered the node.