# GHSA-CG26-4C6J-2X8H: OHTTP codecs retain consumed outer cumulation buffers after completed exchanges

- Database: GitHub
- Severity: CVSS 5.3 (Medium)
- Published: 2026-09-01
- Modified: 2026-09-01

## Tracked as

- [CVE-2026-69133](https://cve.now/CVE-2026-69133.md): netty-incubator-codec-ohttp unauthenticated memory exhaustion in OHTTP codecs

## Affected versions

- io.netty.incubator:netty-incubator-codec-ohttp (Maven): from 0.0.20.Final, through 0.0.26.Final

## Description

## Summary:

OHttpClientCodec and OHttpServerCodec retain the encrypted outer HttpContent in a handler-level cumulationBuffer. After a valid OHTTP exchange completes, the OHTTP request/response context is destroyed, but the outer cumulationBuffer is not released or reset. It remains referenced by the codec until a later inbound content frame replaces it or the handler is removed/channel is closed.

This can pin attacker-influenced pooled/direct buffer capacity on idle keep-alive channels. Across many long-lived idle connections, this can create availability pressure.

Affected components:

* codec-ohttp/src/main/java/io/netty/incubator/codec/ohttp/OHttpClientCodec.java
* codec-ohttp/src/main/java/io/netty/incubator/codec/ohttp/OHttpServerCodec.java
* codec-ohttp/src/main/java/io/netty/incubator/codec/ohttp/OHttpRequestResponseContext.java

## Root cause:

OHttpClientCodec retains inbound content into a handler-level cumulationBuffer. On LastHttpContent, it destroys only the OHttpRequestResponseContextHolder. The cumulationBuffer is released in handlerRemoved(), not immediately after successful exchange completion.

OHttpServerCodec follows the same pattern: it retains inbound content into a handler-level cumulationBuffer, and on completed request/response it destroys the OHttpRequestResponseContext. The outer cumulationBuffer remains owned by the codec and is released only when the handler is removed.

OHttpRequestResponseContext.parse() consumes bytes from the caller-provided outer cumulation buffer, but its destroy() path releases only the inner binary HTTP cumulation, not the codec-level outer cumulationBuffer.

## PoC:

I added a local test only:

```
codec-ohttp/src/test/java/io/netty/incubator/codec/ohttp/OHttpCumulationBufferRetentionTest.java
```

The test reuses existing OHTTP test helpers and BouncyCastle crypto provider to perform valid completed OHTTP exchanges over EmbeddedChannel pairs. It sends both a small and a large request/response exchange, confirms completion, then reflects the private cumulationBuffer field from both OHttpClientCodec and OHttpServerCodec.

Command:

```
./mvnw -pl codec-ohttp -Dtest=OHttpCumulationBufferRetentionTest -DfailIfNoTests=false test
```

Observed output:

```
small client after completion: refCnt=1 capacity=1064 readerIndex=1064 writerIndex=1064 readableBytes=0
small server after completion: refCnt=1 capacity=1110 readerIndex=1110 writerIndex=1110 readableBytes=0
large client after completion: refCnt=1 capacity=262144 readerIndex=131114 writerIndex=131114 readableBytes=0
large server after completion: refCnt=1 capacity=262144 readerIndex=131160 writerIndex=131160 readableBytes=0
small after close: clientOldRefCnt=0 serverOldRefCnt=0
large after close: clientOldRefCnt=0 serverOldRefCnt=0
Tests run: 1, Failures: 0, Errors: 0, Skipped: 0
BUILD SUCCESS
```

Interpretation:

After completion, both the client and server cumulationBuffer objects have:

* refCnt=1
* readableBytes=0
* large retained capacity for the large exchange
* channel still open/active
* OHTTP context completed/destroyed

After channel close/handler removal, the previously retained buffers are released and their old refCnt values become 0.

## Impact:

A peer can complete a large valid OHTTP request/response and keep the connection open. The codec retains the high-water outer cumulation buffer for the lifetime of the idle keep-alive channel, or until a later inbound content frame replaces it. Across many idle keep-alive channels, this can pin pooled/direct memory and create availability pressure.

This is not unbounded growth per exchange on a single channel. The more precise impact is connection-lifetime retention of the current consumed high-water outer cumulation buffer.

## Suggested fix:

After successful LastHttpContent processing, release and reset the codec-level outer cumulationBuffer once parsing has consumed the complete message and no unread bytes remain.

Care is needed not to release the buffer if unread bytes remain for pipelined data. A safe fix would release/reset the outer cumulationBuffer in the completed-exchange path only when the parse consumed the full outer body.

## Test environment:

* WSL Ubuntu
* Java 11
* Current main branch / 0.0.24.Final-SNAPSHOT
* PoC test used BouncyCastleOHttpCryptoProvider and existing OHttpCodecsTest helpers

Affected version testing:

I reproduced the issue on the following release tags:

* netty-incubator-codec-parent-ohttp-0.0.20.Final
* netty-incubator-codec-parent-ohttp-0.0.21.Final
* netty-incubator-codec-parent-ohttp-0.0.22.Final
* netty-incubator-codec-parent-ohttp-0.0.23.Final

I also reproduced it on current main / 0.0.24.Final-SNAPSHOT.

For 0.0.20.Final through 0.0.23.Final, the PoC output consistently showed:

* large client after completion: refCnt=1 capacity=262144 ... readableBytes=0
* large server after completion: refCnt=1 capacity=262144 ... readableBytes=0
* large after close: clientOldRefCnt=0 serverOldRefCnt=0
* BUILD SUCCESS

This confirms that completed exchanges leave the consumed outer cumulation buffer retained until channel/handler close in these versions.

## References

- https://github.com/netty/netty-incubator-codec-ohttp/security/advisories/GHSA-cg26-4c6j-2x8h

Page: https://cve.now/GHSA-CG26-4C6J-2X8H
