Brotli2Shell: Anatomy of a Critical Compression Library Vulnerability (CVE-2026-111497)

A critical out-of-bounds write in Google's Brotli decoder enables remote code execution. Covers root cause, affected software across Node.js, Python, .NET, Envoy, and Istio, and practical detection and remediation strategies.

Terminal output showing a Docker container starting with the mattrayner/lamp:latest-1804 image
Feature image: Alexandru.munteanu98 / CC BY-SA 4.0 / Wikimedia Commons.

CVE-2026-111497 is a critical out-of-bounds write in Google's Brotli decoder that enables remote code execution across Node.js, Python, .NET, Envoy, and many other products. The root cause is an unchecked return value in block-switch processing — when a format error is silently discarded, the decoder continues with corrupted Huffman-tree state and writes past buffer boundaries.

Why is Brotli2Shell (CVE-2026-111497) critical for engineering teams?

Because most networks run affected Node.js versions, the vulnerability is reachable via HTTP request bodies and responses, and exploitation has been confirmed in the wild. Teams need to audit their Brotli dependencies, verify running versions in containers, and prioritize patching internet-facing services that decompress untrusted compressed data.

Sources: Node.js Security Advisory · Google Brotli v1.2.1 Release

CVE-2026-111497 — Out-of-Bounds Write in Google's Brotli Decoder

Published: November 2026
Severity: Critical (CVSSv4.0: 9.2)
Known Exploited: Yes — added to CISA KEV on October 27, 2026


Overview

CVE-2026-111497 is a critical out-of-bounds write vulnerability in Google's Brotli compression library, affecting the decoder implementation in c/dec/decode.c. The flaw has circulated under several names — Brotli2Shell, Node2Shell, HTTP2Shell, and Broccoli2Shell — reflecting its impact across multiple downstream ecosystems.

Google scored it 9.2 under CVSSv4.0 with the vector:

CVSS:4.0/AV:N/AC:L/AT:P/PR:N/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N

The attack vector is network-based, complexity is low, no privileges or user interaction are required, and the impact on confidentiality, integrity, and availability is high. Scope is unchanged (S:C). Microsoft rates the flaw 8.1 under CVSSv3.1 for its .NET implementation.


The Vulnerability: What Went Wrong

Root Cause: Unchecked Return Value (CWE-252)

The vulnerability sits in DecodeBlockTypeAndLength(), a function in the Brotli C decoder that reads block type and length information from the compressed stream during meta-block processing.

Brotli divides each meta-block into three categories — literals, insert/copy commands, and distances. Each category is Huffman-encoded and may be split into multiple blocks, each with its own Huffman tree. When a new block begins, the decoder reads a block switch from the stream: the next block type and its length.

The bug occurs when a category declares only one block type (max_block_type <= 1). In that case, the RFC specifies that no block switches are valid — the stream should carry exactly 16,777,216 blocks of that single type. If the stream forces a switch before that count is exhausted, it is invalid.

In Brotli versions 1.0.3 through 1.2.0, the fast-path wrappers for DecodeLiteralBlockSwitch, DecodeCommandBlockSwitch, and DecodeDistanceBlockSwitch were declared as static void. They called DecodeBlockTypeAndLength(), which returned a boolean indicating success or failure — but the result was discarded. The decoder continued processing with corrupted state.

On the slow (safe) path, the boolean return was read and treated as "needs more input" rather than "format error." This meant even the safe path could be misled by crafted input — it would simply keep appending bytes to an internal eight-byte buffer, eventually writing past its bounds.

The Exploit Path

  1. An attacker crafts a Brotli-compressed stream containing an invalid block switch in a single-type category
  2. The decoder reaches DecodeBlockTypeAndLength(), which returns failure (false) because the stream is malformed
  3. On the fast path, the void wrapper discards this result; on the safe path, it's misinterpreted as NEEDS_MORE_INPUT
  4. The decoder continues with stale Huffman-tree state — block_type_rb, block_length[tree_type], htree_command, and dist_context_map_slice are never updated
  5. Subsequent decoding operations write to the ring buffer using incorrect offset calculations, causing an out-of-bounds write into adjacent decoder state memory

What Successful Exploitation Enables

  • Remote code execution — demonstrated against unpatched Node.js environments
  • Process crash — denial of service via memory corruption
  • Information disclosure — process memory contents can be read
  • State corruption — the decoder's internal buffers and Huffman trees are overwritten, potentially affecting subsequent decompression operations

Affected Software

The risk applies wherever an application decompresses Brotli data that an attacker can control. This includes both inbound (request bodies) and outbound (fetched responses) HTTP traffic, as well as compressed files.

Core Library

Product Affected Versions Fixed Versions
Google Brotli (libbrotli) 1.0.3 – 1.0.9, 1.1.0, 1.2.0 1.0.10, 1.1.1, 1.2.1

Language Runtimes

Product Affected Versions Fixed Versions
Node.js < 22.23.4, < 24.22.1, < 26.12.1 22.23.4, 24.22.1, 26.12.1
Bun >= 1.0.21, < 1.4.4 1.4.4
Python Brotli <= 1.2.0 1.2.1
Python brotlicffi <= 1.2.0.2 1.2.1.0
Java brotli4j < 1.24.0 1.24.0
Ruby brotli gem < 0.21.1 0.21.1

Infrastructure and Frameworks

Product Affected Versions Fixed Versions
Envoy Proxy <= 1.36.11, 1.37.7, 1.38.5, 1.39.2 1.36.12, 1.37.8, 1.38.6, 1.39.3, 1.40.0
Istio < 1.29.9, < 1.30.6, < 1.31.2 1.29.9, 1.30.6, 1.31.2
Microsoft .NET 8 < 8.0.33 8.0.33
Microsoft .NET 9 < 9.0.22 9.0.22
Microsoft .NET 10 < 10.0.14 10.0.14
PHP for Windows Bundled Brotli 1.0.3–1.2.0 8.2.35-1, 8.3.36-1, 8.4.27-1, 8.5.12-1

Other Affected Projects

  • FFmpeg — reachable via libjxl Brotli-compressed metadata in JPEG XL files (JFrog demonstrated instruction-pointer control)
  • Apache Arrow — infinite-loop denial of service with a crafted 534-byte input (fix merged, no release as of October 26)
  • ClickHouse — crashable with a single malformed Brotli-compressed HTTP request (fix merged)
  • curl/libcurl — vulnerable when built with --compressed or CURLOPT_ACCEPT_ENCODING set

End-of-Life Lines

Node.js 20.x, 23.x, and 25.x receive no fix. Akamai telemetry showed that a majority of networks are running affected Node.js versions.


Exposure Paths

The vulnerability is reachable through multiple attack vectors:

1. Inbound Request Bodies (Server-Side)

Applications that decompress incoming Content-Encoding: br request bodies are exposed. Express 5 with default JSON body parsing (express.json()) does this automatically. Any API endpoint that accepts Brotli-encoded POST data is a potential entry point.

2. Outbound HTTP Responses (Client-Side)

Any application that fetches URLs from untrusted sources and decompresses the response is at risk. Node.js fetch() with Content-Encoding: br on the response, or Python HTTP clients (urllib3[brotli], requests, httpx, aiohttp[speedups]) when the Brotli package is installed, will automatically decompress Brotli responses — even if the client did not advertise br in Accept-Encoding, because a hostile server can send Brotli-encoded data unilaterally.

3. Compressed Files

Brotli-compressed files received from untrusted sources (e.g., uploaded documents with Brotli metadata, webhooks containing compressed payloads) are also in scope.


The Fix

Google committed the upstream fix to the Brotli source repository on February 27, 2026 — eight months before public disclosure. Stable fixes were released on October 21, 2026.

What Changed

The fix touches two files: c/dec/decode.c and c/include/brotli/decode.h. The key changes are:

  1. Block switch functions now return error codes. The three Decode*BlockSwitch wrappers changed from static void to returning BrotliDecoderErrorCode, passing the result through a new BROTLI_SAFE_WITH_STATUS macro on both paths.
  2. A new error code. BROTLI_DECODER_ERROR_FORMAT_BLOCK_SWITCH (-17) was introduced as a formal error code. Until this fix, -17 was a reserved value.
  3. Invalid streams are rejected at the switch point. When a single-type category runs out and forces an invalid block switch, the decoder now stops immediately with BROTLI_DECODER_ERROR_FORMAT_BLOCK_SWITCH instead of continuing in corrupted state.

Why mbrotli Is Not Affected

The Rust crate mbrotli is immune to this specific vulnerability because it is built with #![cfg_attr(not(test), forbid(unsafe_code))]. A logic error of the same kind could make its decoder accept a stream it should reject, but it cannot write past buffer boundaries — the Rust borrow checker and bounds checks prevent the out-of-bounds write that turns the format bug into memory corruption.


Detection and Scanning

brotli2scan

The community tool brotli2scan (by shouldipanic.ch) detects vulnerable servers by sending two small POST requests: one with a valid Brotli body (the 6 bytes encoding {}) and one with an unknown Content-Encoding. A server that accepts the Brotli request but rejects the unknown encoding is decoding Brotli it receives — and is potentially vulnerable.

Manual Checks

Node.js:

node -v
# Any version below 22.23.4, 24.22.1, or 26.12.1 is affected

Python:

pip show brotli brotlicffi
# Below 1.2.1 / 1.2.1.0 needs updating

Bun:

bun --version
# Below 1.4.4 is affected

Envoy:

envoy --version
# Below 1.40.0 (or the latest in your supported branch)

Container Images

A running process keeps the old code until restarted. A container image must be rebuilt and redeployed to pick up the fix. Updating the host's system libbrotli library only helps processes that link against it dynamically — applications that bundle their own copy (Node.js, Bun) remain vulnerable until the application itself is updated.


Mitigation Strategies

Immediate (Before Patching)

  1. Reject Content-Encoding: br at the proxy or load balancer. This blocks inbound Brotli request bodies from reaching the vulnerable decoder. Note: this does not protect against outbound responses — if your code fetches URLs and decompresses the response, that path remains exposed.
  2. Disable Brotli decoding in HTTP clients. For Node.js, this means removing br from Accept-Encoding and disabling response decompression. For Python, uninstalling the Brotli and brotlicffi packages forces HTTP clients to fall back to gzip and deflate.
  3. Remove Brotli from Express 5 middleware. Set express.json({ inflate: false }) — note this also disables gzip, so test your clients.
  4. Disable the Brotli decompressor filter in Envoy/Istio. A disabled filter forwards compressed bytes as-is, which leaves any downstream service that decodes Brotli at risk.
  5. Run Node.js as an unprivileged user and keep secrets out of its environment to reduce the impact of a successful exploit.

Long-Term

  1. Upgrade to patched versions. See the version tables above. Restart every process, rebuild every container image.
  2. Move off end-of-life Node.js lines. Node.js 20.x gets no fix. Plan migration to 22.x or 24.x.
  3. Verify the running version, not just the installed one. A patched package on disk doesn't help if the running process hasn't been restarted. Check what actually runs inside containers and workloads.

Audit transitive dependencies. urllib3[brotli], httpx[brotli], and aiohttp[speedups] all install Brotli without anyone explicitly importing it. Search your lockfiles:

grep -RniE "brotli|speedups" --include="requirements*.txt" --include="poetry.lock" --include="uv.lock" .

Timeline

Date Event
February 27, 2026 Upstream fix committed to google/brotli repository
October 21, 2026 Google releases fixed Brotli versions; Node.js discloses vulnerability in security advisory
October 22, 2026 NVD publishes CVE record; Python packages reach PyPI
October 25, 2026 Microsoft releases .NET security updates
October 26, 2026 Canadian Centre for Cyber Security releases advisory AL26-027; ELLIO reports exploitation attempt against Node.js 24.22.0 on x86-64
October 27, 2026 CISA adds CVE-2026-111497 to Known Exploited Vulnerabilities (KEV) catalog; remediation due date November 10 for FCEB agencies

Conclusion

CVE-2026-111497 is a textbook example of how an unchecked return value in error handling can cascade into remote code execution. The fix was committed eight months before disclosure, but the window between fix and public disclosure was critical — public PoCs are available, and exploitation attempts have been observed.

For engineering teams: audit your dependency tree for Brotli, verify running versions in every container and workload, and prioritize patching internet-facing services that decompress untrusted compressed data. The networks still running affected Node.js versions represent a significant attack surface.


References

Feature image: Alexandru.munteanu98 / CC BY-SA 4.0 / Wikimedia Commons.