Follow @Openwall on Twitter for new release announcements and other news
[<prev] [day] [month] [year] [list]
Message-ID: <9FC5D21F-35FF-4FFC-8A07-0E8071D6CFAF.1@smtp-inbound1.duck.com>
Date: Sun, 22 Nov 2026 03:22:36 -0500
From: theknell@...k.com
To: oss-security@...ts.openwall.com
Subject: aiohttp: half a fix — the CVE-2025-69224 header hardening covers the request parser, not the response parser (live-proven, low)

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

Half a fix.

In January 2026 aiohttp fixed CVE-2025-69224, rated low: Unicode
processing of header values could cause parsing discrepancies. The
fix rejected non-ASCII characters in some headers. It was applied
to the parser that receives requests. The parser that receives
responses — the same check, one class further down the same file —
was not touched, and ten months later it still has not been.

Tested against aio-libs/aiohttp at commit
fe099491d275ef2c21bf683e024e34d51ac05d43 (2026-11-13), full history
reviewed:

1. The hardening (commit 5affd64f86d28a16a8f8e6fea2d217c99bf7831f,
2026-01-02, "Reject non-ascii characters in some headers") added
an .isascii() guard before .lower() in
HttpRequestParser._is_chunked_te (aiohttp/http_parser.py:824-838).
The guard carries the author's own comment: ".lower() transforms
some non-ascii chars, so must check first."

2. HttpResponseParser._is_chunked_te (http_parser.py:921-923) is
the response-side twin — a single line,
te.rsplit(",", maxsplit=1)[-1].strip(" \t").lower() == "chunked",
with no .isascii() guard. A history search confirms it has never
received one; a later regression fix on top of the hardening
(be0278388, 2026-10-03) passed it by again.

3. The Kelvin sign, U+212A, lowercases to the ASCII letter k. So
"chun\u212aed".lower() == "chunked" is True. A response carrying
Transfer-Encoding with the folded value is framed as chunked by
the pure-Python response parser, while the request parser raises
BadHttpMessage on the identical value.

4. The C parser is not affected: llhttp matches "chunked"
byte-exact. The gap is in the pure-Python parser, the active
parser where the extension is absent or disabled —
AIOHTTP_NO_EXTENSIONS=1, PyPy, and no-compiler installs.

Live proof, against our own rig only (pinned source, pure-Python
parser, mock protocol): the control — the request parser — accepts
ASCII "chunked" and raises BadHttpMessage on the Kelvin-folded
value. The candidate — the response parser — accepts both. Control
rejects, sibling accepts, replicated twice. Harness and raw log
accompany the paper.

Scope, stated plainly: low severity, scored that way here,
matching the vendor's own rating of the parent CVE. It is
client-side. The precondition is a malicious or compromised
server — or a shared intermediary — that sends the folded header:
the client then frames the body as chunked where a
byte-exact-matching intermediary does not, a response-framing
desynchronization. Not an unauthenticated RCE, and no such class
is claimed. What is claimed is narrower and fully verifiable: a
security fix was applied to one of two identical checks, and the
unfixed half is still open at HEAD.

What held is credited at the same weight. The request-side
hardening is correct where it was applied, in both the Cython and
pure-Python sources. The C parser was never vulnerable to the
fold. The fix is one guard: the same .isascii() check the request
parser already carries, applied in
HttpResponseParser._is_chunked_te.

The fix was right. It was applied to the parser that receives
requests and not to the one that receives responses. Ten months
on, the response parser is still folding.

Paper: https://archive.org/details/knell-half-a-fix

Key: 133883AB9F6EDE5B9BA6B82E27BEDAF0CE152F25 (knell)

Verify: gpg --import knell-public.asc, then
gpg --verify paper-aiohttp-FINAL-2026-11-13.md.asc

— knell

More, signed and timestamped:
https://github.com/the-knell
-----BEGIN PGP SIGNATURE-----

wnUEARYIAB0FgmsCphgWIQQTOIOrn27eW5umuC4nvtrwzhUvJQAKCRAnvtrwzhUv
JUSfAP9YCyf7I9GqPOGMXDwR2Ag++cY2wlwA4zy/8dCgrYsADwEAl9WlGHxKwTCD
o0ReJIdPGnFwaCV13Zhk+cPw57fhVQU=
=4X+C
-----END PGP SIGNATURE-----

Powered by blists - more mailing lists

Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.