Repository navigation
fix(deps): bump brotli to 1.2.1 for scanners complaining about CVE-2026-111497 - #45743
Conversation
Brotli 1.0.3-1.2.0 has an out-of-bounds write in the decoder (CVE-2026-111497, PYSEC-2026-4322, GHSA-9v8h-qqgm-4m35). Superset pins brotli==1.2.0; HTTP clients (requests, httpx, aiohttp, urllib3) use it to decode `Content-Encoding: br` responses. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## master #45743 +/- ##
=======================================
Coverage 84.31% 84.31%
=======================================
Files 3059 3059
Lines 202225 202225
Branches 47367 47367
=======================================
Hits 170516 170516
Misses 28451 28451
Partials 3258 3258
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
sha174n
left a comment
There was a problem hiding this comment.
LGTM. Verified the bump independently before approving:
- brotli 1.2.1 (published 2026-10-22) ships manylinux and musllinux wheels for cp310 through cp315 on both x86_64 and aarch64, plus an sdist, so nothing in the support matrix falls back to a source build.
base.inand both lock files are consistent. Moving the floor rather than only the lock is what keeps the nextpip-compilefrom reverting it.- No other brotli pin in the repo. The only other reference is
superset-frontend/webpack.proxy-config.js, which is dev proxy tooling, and there is nobrotlicffi.
On the reachability note in the description: agreed, and the inbound direction is worth stating explicitly for anyone triaging this downstream. brotli arrives via flask-compress, which compresses responses, and Werkzeug does not decompress request bodies, so there is no Content-Encoding: br request path into the decoder. The exposure is outbound: urllib3 2.x adds br to its default Accept-Encoding and decodes Brotli responses whenever the brotli package is importable, with no application code involved.
Two non-blocking notes:
-
base.inline 27 now reads# Security: decompression bomb fix (required by aiohttp 3.13.3); CVE-2026-111497, but aiohttp is not inbase.txtat this head. The onlyviaentries underbrotli==1.2.1are-r requirements/base.inandflask-compress, so the aiohttp half of that comment looks stale. -
+1 on the pyarrow point, and apache/arrow#55166 is the right tracker. Separate change, and no reason to gate this one on it.
…26-111497 (#45743) Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit 86d54d9)
|
Pyarrow vendors its own brotli and they released a new version - #46018 bump for pyarrow |
SUMMARY
Bump brotli to avoid scanners complaining about CVEs.
Brotli 1.0.3–1.2.0 has an out-of-bounds write in its decoder: CVE-2026-111497 (GHSA-9v8h-qqgm-4m35, PYSEC-2026-4322).
It is fixed in 1.2.1.
Superset pins
brotli==1.2.0. Bumps the floor.Exploitation looks very difficult to do, you need to make Superset make HTTP request to a server that sends poisoned response - exploitation not known in the wild any python stack yet.
Testing
python -c "import brotli; print(brotli.__version__)"→1.2.1.BTW
brotlifrom v1.2.0 to v1.2.1 ClickHouse/ClickHouse#127419 clickhouse seen DoS like this