Skip to content

JS: HTML escaping with replaceAll("<", "&lt;") is not recognized as a sanitizer by js/xss and js/reflected-xss #23179

Backlog

Description

@wildoranges

Description of the false positive

Escaping HTML with a chain of String.prototype.replaceAll calls that take string patterns, such as s.replaceAll("&", "&amp;").replaceAll("<", "&lt;").replaceAll(">", "&gt;"), is not recognized as a sanitizer, so js/reflected-xss and js/xss report the escaped value. Unlike replace with a string pattern, replaceAll replaces every occurrence, so this escapes the input exactly like the same chain written with global regular expressions (replace(/</g, "&lt;")), which is not reported. replaceAll(/</g, "&lt;") is not reported either. The string form is a natural way to write this since replaceAll became available (ES2021, Node.js 15).

As far as I can tell, this is because Shared::MetacharEscapeSanitizer only inspects this.getRegExp().getRoot(), which has no result when the first argument is a string. StringReplaceCall already handles this case: it includes replaceAll, treats it as global in isGlobal() and maybeGlobal(), and getAReplacedString() returns a string first argument. The sanitizer is shared by the DOM-based, reflected and stored XSS queries (DomBasedXssCustomizations.qll#L322-L328).

Code samples or links to source code

server.js:

const express = require("express");
const app = express();

function escapeHtml(s) {
  return s.replaceAll("&", "&amp;").replaceAll("<", "&lt;").replaceAll(">", "&gt;")
          .replaceAll('"', "&quot;").replaceAll("'", "&#39;");
}

function escapeHtmlRegExp(s) {
  return s.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;")
          .replace(/"/g, "&quot;").replace(/'/g, "&#39;");
}

app.get("/a", (req, res) => {
  res.send("<p>" + escapeHtml(String(req.query.q)) + "</p>"); // line 15: reported
});

app.get("/b", (req, res) => {
  res.send("<p>" + escapeHtmlRegExp(String(req.query.q)) + "</p>"); // line 19: not reported
});

app.get("/c", (req, res) => {
  const q = String(req.query.q);
  res.send("<p>" + q.replaceAll(/</g, "&lt;") + "</p>"); // line 24: not reported
});

app.get("/d", (req, res) => {
  res.send("<p>" + String(req.query.q) + "</p>"); // line 28: reported (control)
});

app.listen(process.env.PORT || 3000);

page.js:

function show() {
  const v = decodeURIComponent(location.hash.slice(1));
  document.getElementById("a").innerHTML =
    v.replaceAll("&", "&amp;").replaceAll("<", "&lt;").replaceAll(">", "&gt;"); // line 4: reported
  document.getElementById("b").innerHTML =
    v.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;");       // line 6: not reported
}

With CodeQL 2.27.5, js/reflected-xss reports server.js lines 15 and 28, and js/xss reports page.js line 4. Running the server with Node.js v24.18.0 and Express 4.22.3, /a and /b return byte-identical responses when q is <script>alert(1)</script>"'& (sent URL-encoded):

GET /a -> <p>&lt;script&gt;alert(1)&lt;/script&gt;&quot;&#39;&amp;</p>
GET /b -> <p>&lt;script&gt;alert(1)&lt;/script&gt;&quot;&#39;&amp;</p>
GET /d -> <p><script>alert(1)</script>"'&</p>

The pattern occurs in real code, for example (in each repository, scanned as a whole with the default javascript-code-scanning.qls suite on 2.27.5, this is the only XSS alert):

  • voicetreelab/voicetree: esc() escapes &, <, >, " and ' with replaceAll. The run value from location.hash (L17-L20) is rendered as <h2>${esc(ts)}</h2> (L80), and js/xss reports the innerHTML assignment at L170.
  • lbwnb666-ai/DeepSeekHarnessRemoteGateway: escapeHtml() escapes &, <, > and " with replaceAll. The next query parameter (L201) is rendered as <input type="hidden" name="next" value="${escapeHtml(next)}"> (L133), and js/reflected-xss reports the response at L18.

Would it be reasonable for MetacharEscapeSanitizer to also accept such replaceAll calls?

Possible change

One possible change is a third case in Shared::MetacharEscapeSanitizer for replaceAll calls whose first argument is the constant string <, ' or " (against current main):

--- a/javascript/ql/lib/semmle/javascript/security/dataflow/Xss.qll
+++ b/javascript/ql/lib/semmle/javascript/security/dataflow/Xss.qll
@@ -42,6 +42,10 @@ module Shared {
         or
         // or it's like a wild-card.
         RegExp::isWildcardLike(this.getRegExp().getRoot())
+        or
+        // `replaceAll` with a string pattern replaces every occurrence of that string
+        this.getMethodName() = "replaceAll" and
+        this.getArgument(0).getStringValue() = ["<", "'", "\""]
       )
     }
   }

replace with a string pattern is left out on purpose, since it only replaces the first occurrence. With this change, server.js line 15 and page.js line 4 are no longer reported, line 28 still is, and the tests in javascript/ql/test/query-tests/Security/CWE-079/ still pass with unchanged .expected files.

This does not add a new kind of false negative. The string form is then treated exactly like the regexp form, so every call it starts accepting has a regexp equivalent (replace(/</g, …), replaceAll(/'/g, …)) that is already accepted today. What it stops reporting falls under the existing limitations of this sanitizer: it does not consider where the escaped value ends up (for example a javascript: URL in an href attribute, or an unquoted inline script), and a call that only escapes quotes counts as a sanitizer even in element content, as replace(/'/g, "''") already does. In a sample of 278 real files that call replaceAll("<", …) and use request data, the change removed 21 alerts and added none. In 19 of them the escaped value ends up in element content or a double-quoted attribute; the other 2 also have a path into an href attribute or a string in an inline script, where HTML escaping is not enough.

Environment: CodeQL CLI 2.27.5 (codeql/javascript-queries 2.5.1); links point to CodeQL main f7a381fd5a8eee235dda7f83dd1f81e0a109e3f0.

Activity

  1. jketema commented on Nov 20, 2026

    @jketema
    Contributor

    Hi @wildoranges,

    Thank you for this false positive report. Resolving this issue is not a current product priority, but we acknowledge the report and will track it internally for future consideration, or if we observe repeated instances of the same problem.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions