⏱️ Reading time: 7 min
OpenSSL is the cryptography and TLS library used by browsers, web servers and much of the software that encrypts traffic across the internet. On November 10, 2026, pull request #33652 was opened in its repository to fix CVE-2026-121481, a buffer overflow in OpenSSL that affects the EVP_EncodeBlock() function, which converts binary data into Base64 text. The patch’s author, contributor nmouha, documented a concrete way to exploit it: passing an enormous password to the openssl s_client proxy with the GOST cryptographic provider enabled.
📑 En este artículo
- What overflowed: OpenSSL’s Base64 buffer overflow
- Why there’s still no fixed version
- What we tested: cloning the repo, unable to build it in the sandbox
- OpenSSL compared to other cryptography projects on GitHub
- What this means for developers who use OpenSSL
- Limitations: what we still don’t know
- References
What overflowed: OpenSSL’s Base64 buffer overflow
The problem lies in how the function calculated how much space it needed for the encoded output. EVP_EncodeBlock() did not reject NULL buffers, negative lengths, or lengths whose Base64-encoded size (which grows by nearly 4/3 relative to the original) would not fit in a 32-bit integer. If the input is large enough, that calculation silently overflows and the function ends up writing outside the buffer the caller allocated. The proposed patch makes EVP_EncodeBlock() return -1 in any of those three cases, instead of attempting to encode anyway.
The same author proposes a similar guard for EVP_DecodeBlock(): reject NULL buffers and negative lengths, although the overflow check isn’t needed there because the decoded length always fits in an int. The PR also fixes an incomplete length check in the base64encode() function in http_client.c, which until now only rejected len > INT_MAX and let the rest of the dangerous cases through.
The exploitation path: s_client, GOST and a giant password
To trigger the overflow from the command line, nmouha confirmed that the GOST provider accepts extremely long passwords without filtering them first. But there’s a practical limit: a password that long doesn’t fit directly as a shell argument, so the attack only works if it’s passed through an environment variable using the env:VAR syntax instead of writing it on the command line.
Why there’s still no fixed version
There is no official security advisory. The PR thread itself clarifies that OpenSSL’s security response team (SRT) asked for the fix to be discussed in public, but the advisory is still pending. That means that, for now, there’s no confirmed list of affected versions and no CVSS score published by OpenSSL. In the Hacker News thread discussing the case, the title assigns it a CVSS of 9.8, but that number comes from the community, not from an NVD or OpenSSL advisory.
The openssl/openssl repository has about 31,500 stars and 11,600 forks as of 2026-11-20, is written in C and is published under the Apache 2.0 license. Its latest stable release, OpenSSL 4.1.0, came out on October 23, 2026, almost three weeks before this PR was opened. No released version to date includes this fix yet.
The PR is also still open. nhorman approved it on November 11, but the next day two reviewers, jogme and t8m, requested changes. jogme pointed out that, by turning EVP_EncodeBlock() into a function that can fail, there are callers inside OpenSSL that still don’t check that new return value, such as PEM_SignFinal() and NETSCAPE_SPKI_b64_encode(). t8m agreed the problem is real but suggested fixing it in a separate commit. Until that point is resolved, the patch remains under review.
The Hacker News discussion isn’t limited to reporting the CVE. User firesteelrain raises a process objection: if a project already runs linting, static analysis, dynamic analysis, DAST testing and fuzzing on its code and a bug like this still slips through, does it still make sense to say the code ‘wasn’t properly reviewed,’ or is the real question what standard counts as sufficient review? It’s an argument about process, not a technical detail of the bug, but it explains why a CVE like this keeps showing up in a library with more than two decades of audits behind it.
What we tested: cloning the repo, unable to build it in the sandbox
For this story we tested cloning the repository in an isolated environment, with no external network access, with git 2.39.5 and Python 3.12.16 installed. The clone finished in 4,921 seconds, downloaded 188,8 MB and used a peak of 90,4 MB of memory:
$ git -c protocol.allow=never -c protocol.https.allow=always -c core.hooksPath=/dev/null clone --depth 1 --single-branch --no-tags -- https://github.com/openssl/openssl /tmp/lab-runs/lab-dqoqerm9/w/repo
Cloning into '/tmp/lab-runs/lab-dqoqerm9/w/repo'...
Our sandbox couldn’t go any further: the runner reported no supported stack detected, meaning it didn’t identify an automatable build or test flow for a C project that compiles with Perl and Configure instead of a standard package manager. We weren’t able to build OpenSSL or reproduce the overflow ourselves; what we describe about the bug comes from the PR and its code review, not from our own test.
OpenSSL compared to other cryptography projects on GitHub
openssl/openssl shares GitHub’s ‘cryptography’ tag with other large projects, although none of them is a real substitute for a TLS library embedded in browsers and servers:
| Project | What it does | Stars (as of 2026-11-20) | Relationship to OpenSSL |
|---|---|---|---|
| openssl/openssl | General-purpose TLS and cryptography in C | ~31,500 | Subject of this story |
| bitcoin/bitcoin | Bitcoin node and reference implementation | ~90,700 | Shares the ‘cryptography’ topic on GitHub |
| TheAlgorithms/JavaScript | Educational catalog of algorithms and data structures | ~34,300 | Shares the ‘cryptography’ topic on GitHub |
| anoma/anoma | Reference implementation of the Anoma protocol | ~33,500 | Shares the ‘cryptography’ topic on GitHub |
Bitcoin Core doesn’t implement TLS for its P2P network and doesn’t replace OpenSSL: it uses its own elliptic-curve cryptography (libsecp256k1) for signatures. TheAlgorithms/JavaScript is educational material meant for learning algorithms, not for production. anoma/anoma is the reference implementation of a different privacy protocol, built on its own primitives. The comparison shows that ‘cryptography’ on GitHub groups together projects with very different goals: none of them competes with OpenSSL as an embedded TLS library, but they do compete for the attention of the same developer community.
What this means for developers who use OpenSSL
This directly affects two kinds of developers. If you maintain C code (or bindings) that calls EVP_EncodeBlock() or EVP_DecodeBlock() with lengths that come from external input (a network payload, a user-uploaded file, a configuration parameter), add your own length validation right now. Don’t assume the PR will be merged soon: it has had two pending review rounds since November 12.
If instead you operate infrastructure that invokes openssl s_client with the GOST provider and passes secrets through -proxy_pass, for example in automation scripts or CI pipelines that test GOST TLS connections, avoid long unvalidated strings even when you pass them via env:VAR. That’s exactly the path the patch’s own author confirmed.
Limitations: what we still don’t know
There are several things this case still lacks. There’s no official OpenSSL advisory, so there’s no confirmed list of affected versions: everything known comes from the PR itself. The PR hasn’t been merged and has unresolved change requests from two reviewers, so the function’s final behavior (and that of its callers) may change before it reaches a release. The CVSS of 9.8 circulating comes from a Hacker News headline, not from NVD or OpenSSL, and our own attempt to reproduce the bug in a sandbox didn’t get far enough to build the project, so we didn’t independently verify the overflow. Treat this story as following an ongoing PR, not as the announcement of an already-available patch.
References
- Pull request #33652: description of the patch, the cases it covers and the review thread with the requested changes.
- openssl/openssl repository: project stats, language and license.
- OpenSSL 4.1.0 release: latest stable version published, prior to this patch.
- Hacker News thread on CVE-2026-121481: community discussion and the CVSS figure circulating.
Featured image: Foto de Kedibone Isaac Makhumisane en Unsplash
How each article is made (sources, testing and editorial criteria): How we work.
Did it work for you? Got a different error? Say so below: questions get answered and help the next reader.
Leave a comment
0 Comments