Skip to content

[fix][sec][branch-4.2] Upgrade Netty to 4.1.137 to address several CVEs and bugs - #26301

Merged
lhotari merged 1 commit into
apache:branch-4.2from
lhotari:lh-branch-4.2-netty-4.1.137
Aug 10, 2026
Merged

[fix][sec][branch-4.2] Upgrade Netty to 4.1.137 to address several CVEs and bugs#26301
lhotari merged 1 commit into
apache:branch-4.2from
lhotari:lh-branch-4.2-netty-4.1.137

Conversation

@lhotari

@lhotari lhotari commented Aug 10, 2026

Copy link
Copy Markdown
Member

Motivation

Netty 4.1.137.Final is a bug-fix and security
release, and upstream "strongly recommends" upgrading. It is the 4.1 counterpart of 4.2.17.Final,
which is being upgraded on master in #26300.

Security fixes. It resolves the same seven advisories as 4.2.17.Final — every one of them lists
<= 4.1.136.Final as a vulnerable range, so branch-4.2 is affected by all of them. Five still have
their CVE id pending assignment upstream, so the release notes show CVE-2026-XXXXX:

Advisory Severity Netty artifact Summary
GHSA-p85m-gvr3-788c High netty-handler Hostname verification disabled on the OpenSSL client path when trust-manager wrapping is unavailable (Java 25+)
GHSA-c4c3-7fpv-j4q5 High netty-handler SNI routing bypass via fragmented TLS ClientHello causing fallback to the default SslContext
GHSA-fccg-mwvh-qqg4 Medium netty-handler Fragmented ClientHello records trigger quadratic pre-handshake reassembly in default SNI parsing
GHSA-cc6x-ffm5-83wf Medium netty-codec-socks SOCKS4/5 proxy encoder NUL byte and credential injection
CVE-2026-59903 Medium netty-codec-http Cache poisoning and information disclosure via CORS Vary header overwrite
CVE-2026-59902 High netty-transport-sctp Memory exhaustion in SctpMessageCompletionHandler
GHSA-43fm-7cxg-hf3j Low netty-codec-mqtt MQTT topic name and client id validation bypass

netty-transport-sctp and netty-codec-mqtt are not bundled in Pulsar's distributions (they don't
appear in either LICENSE.bin.txt), so the last two are listed only for completeness.
netty-handler, netty-codec-socks and netty-codec-http are all shipped.

GHSA-p85m-gvr3-788c is the one worth calling out, with a branch-4.2 caveat. Netty's earlier
CVE-2026-50010 fix added hostname verification to a plain X509TrustManager by wrapping it, and on
the SslProvider.OPENSSL path that wrapping is Unsafe-based reflection. Unsafe is not available
by default on Java 25+, so the wrapper degrades to a no-op and an OpenSSL client configured with
a plain (non-extended) X509TrustManager ends up doing no hostname verification.

The OpenSSL half of that precondition is met by default on branch-4.2: tlsProvider is unset in
conf/broker.conf, DefaultPulsarSslFactory then passes null to SslContextBuilder, and Netty
falls back to SslContext.defaultClientProvider(), which is OPENSSL whenever the native engine is
available — and it is, since Pulsar bundles netty-tcnative-boringssl-static. The Java 25+ half is
not met by default here: branch-4.2 builds and ships on JDK 21 (IMAGE_JDK_MAJOR_VERSION=21, and
the CI default is 21), unlike master which has moved to JDK 25 (#26070). So this only reaches
deployments that both run a 4.2 broker/client on a Java 25+ runtime and supply a plain rather than
extended trust manager via a custom PulsarTlsFactory. Lower exposure than on master, but the
upgrade removes the question either way.

Non-security fixes. 4.1.137.Final also carries several fixes in areas Pulsar leans on heavily:

  • Fix buddy cache evicting chunks with live buffers (#17176)
  • Adaptive allocator backports (#17206)
  • Reject negative maxOrder in PooledByteBufAllocator (#17095)
  • Fix AdaptiveByteBuf._setLongLE calling checked setLongLE (#17102)
  • SslHandler: fix possible buffer leak when an OOME is thrown during allocation (#17078)
  • Weakly reference engines from the OpenSSL engine map (#17205)
  • Avoid classloader leak via GlobalEventExecutor terminationFuture failure (#17189)
  • HttpObjectEncoder / DefaultHttp2FrameWriter: fix buffer leak when a Throwable is thrown during header encoding (#17178)
  • HttpServerCodec: do not consume the method queue for 1xx interim responses (#17203)
  • Do not write the WebSocket handshake response to the tail of the pipeline (#17200)
  • AsciiString.cached(String) sanitization and the follow-up performance regression fix (#17075, #17080, #17083)
  • Compression decoder hardening: guard the Snappy decoder against invalid chunk lengths (#17110), use the safe decompressor in Lz4FrameDecoder (#17121), fix maxAllocation for brotli-encoded content in HttpContentDecompressor (#17124) and prevent duplicate BrotliEncoder close scheduling (#17193)

netty-tcnative. Bumped 2.0.78.Final → 2.0.81.Final. Pulsar's pom.xml does not pin tcnative
separately — it is managed by the imported netty-bom, which declares
<tcnative.version>2.0.81.Final</tcnative.version> in 4.1.137.Final (it was 2.0.78.Final in
4.1.136.Final), so the bump comes along with the netty.version change.
The changes across those three tcnative releases
are:

  • Clear the certificate chain in setKeyMaterial (netty-tcnative#987)
  • Drop checks on SSL_CREDENTIAL support for older BoringSSL (netty-tcnative#980)
  • Add SSL.getGroupName(...), returning the named group used by the most recently completed handshake (netty-tcnative#991) — the Netty-side OpenSslSession accessor built on it (netty#17058) landed on the 4.2 branch only, so on 4.1.137 this is just the tcnative bump (netty#17122)
  • Link the linux aarch_64 and x86_64 openssl-dynamic artifacts against OpenSSL 3.x (netty-tcnative#986, netty-tcnative#989) — not applicable to Pulsar, which bundles netty-tcnative-boringssl-static

Modifications

  • pom.xml: netty.version 4.1.136.Final → 4.1.137.Final. netty-tcnative moves 2.0.78.Final →
    2.0.81.Final transitively via the netty-bom import; there is no separate version property to
    update on this branch.
  • Update the bundled jar lists in distribution/server/src/assemble/LICENSE.bin.txt and
    distribution/shell/src/assemble/LICENSE.bin.txt to the new versions

Verifying this change

  • Make sure that the change passes the CI checks.

This change is already covered by existing tests. In addition to CI, src/check-binary-license.sh
was run locally against both distribution/server/target/apache-pulsar-4.2.5-SNAPSHOT-bin.tar.gz
and distribution/shell/target/apache-pulsar-shell-4.2.5-SNAPSHOT-bin.tar.gz (both exit 0) to
confirm the LICENSE.bin.txt entries match the jars that are actually bundled.

Does this pull request potentially affect one of the following parts:

If the box was checked, please highlight the changes

  • Dependencies (add or upgrade a dependency)
  • The public API
  • The schema
  • The default values of configurations
  • The threading model
  • The binary protocol
  • The REST endpoints
  • The admin CLI options
  • The metrics
  • Anything that affects deployment

Netty is upgraded from 4.1.136.Final to 4.1.137.Final and netty-tcnative from 2.0.78.Final to
2.0.81.Final.

@lhotari
lhotari merged commit fede2cc into apache:branch-4.2 Aug 10, 2026
53 checks passed
lhotari added a commit that referenced this pull request Aug 10, 2026
nodece pushed a commit to ascentstream/pulsar that referenced this pull request Aug 28, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants