Removal of SSLv3
Hi all,
This PR proposes to remove support for SSLv3 entirely. It has been disabled by default since 2016 (1.1.0) - but this PR would remove the code entirely.
https://github.com/openssl/openssl/pull/28044
What is the BAC's opinion on this?
Thanks
Matt
Per the BAC meeting in Brno earlier this year, IMHO, all this obsolete code should be purged from the mainline product (à la LibreSSL). Perhaps move it to a supplemental library for backward compatibility or reference. So per @Dmitry Belyavsky, remove if it doesn't break the 3.6 ABI. 😽
From my perspective, and for builds for the HPE NonStop community, SSLv3 is not being used - nor has been. It is not being configured into any packages, so removal should have no impact.
The sooner it is completely removed the better IMO. Users have had nine years to ready themselves for this. That's generous.
If we absolutely must keep it around, I like the supplemental library idea.
So far I got 2 answers in the academics community, with diverging opinions on wether this is a 3.6 or 4.0 change, but general support for removing SSLv3 support as soon as it is proper.
https://openssl-communities.org/d/VF7No4lz/removal-of-sslv3
BAC representatives: When to remove SSLv3 support?
dot vote by Nicola Tuveri Closed Tue 12 Aug 2025 9:00PM
What is this poll about?
I have been tasked to collect the advice from each Foundation BAC representative on the opinion of their community about scheduling this for v3.6 or v4.0.
Please collect feedback in your communities and report here the results as percentages, to represent the consensus within each of your communities.
Why is this important?
At the July BAC meeting we noticed there was no objection so far about removal, but that within each community there were different opinions on removing support for SSLv3 ASAP (i.e., v3.6, planned for October 2025) or if it should be scheduled for the next major release (i.e., v4.0, planned for April 2026).
Let's use this poll to summarize the feedback from each community.
What are you asking people to do?
In this poll you have a budget of 20 dots to allocate to the options your communities prefer. You can place all dots on one option or spread them out: each dot represents 5% of your collected feedback.
Results
| Results | Option | % of points | Points | Mean | Voters | ||
|---|---|---|---|---|---|---|---|
| Remove in v3.6 | 65 | 65 | 13 | 5 | |||
| Remove in v4.0 | 35 | 35 | 7 | 5 | |||
| Do not remove | 0 | 0 | 0 | 5 | |||
| Not voted | 0 | 0 | 1 |
@Dmitry Belyavsky So by presuming no ABI break I suppose it was meant in a strict sense that no public API function is removed but instead just returns error if it is SSLv3 specific?
| 20 - Remove in v3.6 | |
| 0 - Remove in v4.0 | |
| 0 - Do not remove | |
All insecure protocols and corresponding obsolete code should be removed from the codebase ASAP. SSLv3 has been dead and buried since at least 2014. I would even question backwards compatibility. Break stuff now while you have a chance. https://blog.mozilla.org/security/2014/10/14/the-poodle-attack-and-the-end-of-ssl-3-0/
| 20 - Remove in v3.6 | |
| 0 - Remove in v4.0 | |
| 0 - Do not remove | |
I have not been including SSLv3 for at least 5 years in any builds.
| 10 - Remove in v3.6 | |
| 10 - Remove in v4.0 | |
| 0 - Do not remove | |
Committer poll has closed. Proportional representation follows here.
| 20 - Remove in v4.0 | |
| 0 - Remove in v3.6 | |
| 0 - Do not remove | |
https://openssl-communities.org/p/fwQTAuZ2/removal-of-sslv3-v3-6-or-v4-0-
Compared to the early comments discussing the opportunity of completely removing the SSLv3 code, which indicated some opinions in favor for removal as soon as possible, the final poll shows unanimous preference in the Academic community for removal at the next major release (v4.0, planned for Apr 2026).
The final poll recorded 5 votes, all in favor of v4.0 as the target.
Yes, no public functions are removed
No public functions are removed in the proposed PR (https://github.com/openssl/openssl/pull/28044). But functions like `SSLv3_method()`, `SSLv3_server_method` and `SSLv3_client_method` are modified to always return `NULL`. There are also some options removed from the command line apps (which I suppose could in theory break some scripts)
Hmmm... Although I just noticed that the 28044 PR as it currently stands removes the `enable-ssl3` Configure option. So it is not longer possible to build a version of OpenSSL where the public functions `SSLv3_method()`, `SSLv3_server_method` and `SSLv3_client_method` exist, i.e. if you were previously using this option, then you would no longer be able to do so. So that is a technical ABI break - you could not just drop in the 3.6 binaries as a replacement for 3.5.
@Tim Chevalier could you please update the poll with your results ASAP? The Foundation asked for the BAC's feedback by today.
Here is an easy to parse summary of the consultation outcomes: https://gist.github.com/romen/4e5437f380e369f93b3d7f06e6364819
I plan to amend the gist as I get Tim's answer. If you have any edit request, please ping me either here or directly on the gist.
Sorry, I missed this thread (I've been out) and I have not polled the community. Based on the July BAC discussion, the conservative approach would be to fully remove in the 4.0 release.
It definitely should be done for 4.0 but I think, if we don't break ABI, we can already do it in 3.6