Disabling support of explicit curves in OpenSSL via compilation options
Using explicit Elliptic curves is discouraged by security reasons and is disabled in RHEL-based distributions via patch.
I think it would be better to have a configure option for this purpose
https://github.com/openssl/openssl/pull/29639
is an implementation based on
https://github.com/beldmit/openssl/commit/6a2b78bca595435fcbf72d7b2c8bec004d555016
Add a configure option to disable support of explicit elliptic curves
poll by Dmitry Belyavsky Closed Mon 19 Jan 2026 7:00AM
What is this poll about?
Why is this important?
What are you asking people to do?
Choose the option(s) you favor.
Results
| Results | Option | % of points | Voters | |||
|---|---|---|---|---|---|---|
| Yes, in 4.0 | 93 | 14 | ||||
| Yes, after 4.0 | 7 | 1 | ||||
| No | 0 | 0 | ||||
| Not voted | 13 |
Being able to remove it would be good for security purposes. Whether it makes it to 4.0 or not it probably should be done.
I think there is still some time to get things into 4.0 and this is not particularly complicated change. The question is whether explicit curve support should be disabled by default or not. If it makes into 4.0, it could be disabled by default actually.
I'm not opposed to adding such a configuration feature, but I think its a little late in the 4.0 development cycle to be adding it
At least this as an option, but would prefer it on by default (i.e. explicit curves disabled by default)
How does the German government deal with this? They use their own curve from memory and it needs to be specified via explicit parameters.
It's an option, not mandatory
@ppzgs1 they use brainpool curves, those are well-defined (have both OIDs and TLS codepoints)
I seem to remember that it was left in specifically for that reason (as was noted by @mspncp), and @romen mentioned it being used for researchers?. That doesnt seem to exclude it from becoming a configurable option. The decision then is it on by default or not.
I was more interested in how people in Germany use RHEL for government work.
If the OS doesn't support explicit curves, that hurts some people. How do they cope?
@Paul Dale We have disabled it 4 years ago and I recall only one request during this time. in Fedora.
@Paul Dale I think the German government uses Brainpool curves (I got requests to enable those in protocols that have no specification for them yet), not arbitrary parameters ones.
@Shane Lontis It would still be usable for research purposes
What specific APIs are being proposed for removal as part of this?
@Matt Caswell compile-time definition doesn't need any API modification, we just get errors in case when we have an explicit curve if its support is disabled
@Dmitry Belyavsky Then I'm confused as to what exactly you are proposing. At the moment I can call (for example) `EC_GROUP_new_curve_GFp` and create an explicit curve. Does this function exist but just fail if explicit curve support is disabled?
I propose dropping support of explicit curves in X.509/PKCS#8/etc
https://github.com/beldmit/openssl/commit/6a2b78bca595435fcbf72d7b2c8bec004d555016
@Matt Caswell ^^
@Dmitry Belyavsky so, if I understand correctly, the proposed patch prevents EC Keys and Parameters using an explicit curve from being decoded, and prevents providers from creating keys based on such curves - but does not block applications from using the legacy low level functions to create them directly via `EC_GROUP_new_curve_GFp` and `EC_GROUP_new_curve_GF2m`
@Matt Caswell Correct
@beldmit wait... wasn't our code allowing for use of explicit parameters if and only if the parameters specified one of the well-known curves?
@Alicja Kario
The check
EC_GROUP_check_named_curve(group, 0, NULL) == NID_undef
stands explicitly for this
As said elsewhere I think we get the real benefit (for users) if we disable by default.
If not, then the benefit of the option allows distributions to have one less patch to maintain so that is also useful (even if disabled).
@Neil Horman I have just pushed the changes
https://github.com/openssl/openssl/pull/29639