npm · @xmldom/xmldom
xmldom: HTML raw-text closing-tag case mismatch causes output amplification
In HTML mode (text/html), a raw-text element (script, style, textarea, title) whose closing
tag differs in case from its opening tag (e.g. </ScRiPt> for <script>) is mishandled by the
parser, producing quadratic (O(n²)) output growth — a small crafted document parses and serializes
into output orders of magnitude larger, exhausting CPU and memory. A modest input of tens of KB can
therefore cause a denial of service in any service that parses untrusted HTML with xmldom. Only HTML
mode is affected.
The parser calls parseHtmlSpecialContent for each raw-text element in HTML mode, matched via
isHTMLRawTextElement / isHTMLEscapableRawTextElement (so all four types — script, style,
textarea, title — are in scope). It searches for the element's closing tag with
source.indexOf('</' + tagName + '>', elStartEnd), a byte-for-byte case-sensitive match. A
mixed-case closing tag never matches, so the search returns -1, and the following
source.substring(elStartEnd + 1, -1) extracts text backwards from the start of the document
instead of the element's content. The function then returns -1 to the parse loop, which cannot
advance normally and falls back to character-by-character reprocessing. Every raw-text element
re-captures all source text preceding it, so output grows as O(n²) in the number of such elements.
lib/sax.js:549): source.indexOf('</' + tagName + '>', elStartEnd) does not fold case, contrary to the WHATWG HTML RAWTEXT end-tag-name rule.-1 (lib/sax.js:550): source.substring(elStartEnd + 1, elEndStart) runs even
when elEndStart === -1, extracting text backwards from position 0.lib/sax.js:556): the function returns elEndStart (-1), driving
repeated character-by-character fallback in the parse loop.Only the 0.9.x line is affected — the amplification was introduced in 0.9.0-beta.1 when
parseHtmlSpecialContent was refactored, and remains through 0.9.11. The 0.8.x line is not
affected: its older parseHtmlSpecialContent does not amplify, despite sharing the same
case-sensitive indexOf.
const { DOMParser, XMLSerializer } = require('@xmldom/xmldom');
const n = 1000;
const payload = '<html><body>' + '<script>x</ScRiPt>'.repeat(n) + '</body></html>';
const doc = new DOMParser().parseFromString(payload, 'text/html');
const out = new XMLSerializer().serializeToString(doc);
console.log(payload.length, out.length, (out.length / payload.length).toFixed(1) + 'x');
// 18026 9037063 501.3x — an 18 KB input yields ~9 MB of output
Output size grows quadratically with the number of case-mismatched raw-text elements:
Repeats | Input len | Output len | Ratio
1 | 44 | 109 | 2.5x
100 | 1826 | 93763 | 51.3x
500 | 9026 | 2268563 | 251.3x
1000 | 18026 | 9037063 | 501.3x
2000 | 36026 | 36074063 | 1001.3x
Proof of Concept from @KarimTantawey (tested with script); the same amplification occurs for style,
textarea, and title.
Small attacker payloads can force disproportionate CPU and memory usage in services that parse and serialize untrusted HTML via xmldom. The quadratic growth means a modest-sized input (tens of kilobytes) can produce output in the tens or hundreds of megabytes, potentially exhausting memory or causing timeouts.
The attack only requires HTML mode (text/html MIME type) and mixed-case closing tags for
any of the four raw-text element types. No special configuration or error handler setup is needed.
The CVSS 4.0 vector scores availability only (VA:H, with VC:N/VI:N): the flaw neither discloses
nor corrupts data, but a small untrusted HTML input (tens of KB) can force output and memory in the
tens to hundreds of MB, enough to exhaust a service's heap or stall its event loop. It is reachable
with no authentication, configuration, or error-handler setup — only that the application parses
untrusted text/html and serializes the result.
The raw-text closing tag is now matched case-insensitively in HTML raw-text mode (per the WHATWG HTML RAWTEXT end-tag rule), and a missing closing tag is handled explicitly, removing the quadratic output amplification. Output for well-formed input is unchanged. Non-breaking; 0.9.x-only.
Is your project exposed to this? Stateward checks every dependency on every pull request and flags it only if your code actually reaches it.
Check my repoSources: CISA KEV (public domain), OSV.dev & GitHub Advisory Database (CC-BY-4.0), FIRST EPSS, NVD/CWE (public domain). Served live from the Stateward advisory database.