ML-KEM/ML-DSA part 2: param builder

[swenson]

Splitting up #2405 into a few parts as suggest by @alex.

This adds the param-builder.

Original commit message:

Add internal module to simplify working with OSSL_PARAM structure We discussed that this API is not well suitable for the end users but still, it required for several operations in OpenSSL 3.* so instead of calling to FFI for every use of this API, this introduces simple wrappers that allow building of the params and their usage.

[alex]

If you can put the openssl-sys changes into their own PR, it makes it considerably easier to review. thanks.

[swenson]

No problem.

[swenson]

@alex this is just the param builder and the related Argon changes now, I believe.

[alex]

Will look tomorrow, thanks.

…

On Fri, Aug 29, 2025 at 12:16 AM Christopher Swenson < ***@***.***> wrote: *swenson* left a comment (sfackler/rust-openssl#2451) <#2451 (comment)> @alex <https://github.com/alex> this is just the param builder and the related Argon changes now, I believe. — Reply to this email directly, view it on GitHub <#2451 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAAGBB6QUDQV46UYFDOOL33P7H2JAVCNFSM6AAAAACFDKWLTGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTEMZVGY3DAOJQGM> . You are receiving this because you were mentioned.Message ID: ***@***.***>

-- All that is necessary for evil to succeed is for good people to do nothing.

[alex]

Merge conflicts, sorry :-(

[swenson]

@alex no worries, rebased

[swenson]

@alex anything else you wanted to address with this PR?

[alex]

Sigh, sorry there was a review comment that apparently never got submitted (and is caught in limbo). Unfortunately the lifetime changes aren't enough:

use std::ffi::CStr;


pub struct OsslParamBuilderRef;
impl OsslParamBuilderRef {
    pub(crate) fn add_octet_string<'a>(
        &'a mut self,
        _key: &'a CStr,
        _buf: &'a [u8],
    ) -> Result<(), ()> {
        Ok(())
    }
}

pub fn f(x: &mut OsslParamBuilderRef) -> Result<(), ()> {
    let key = std::ffi::CString::new(b"my str").unwrap();
    x.add_octet_string(&key, &[])
}

this compiles, and has a UAF.

[swenson]

@alex Absolutely right. For some reason I thought adding &'a to the mut self would be sufficient, but that only ties it to the lifetime of the borrow (?) and not of the actual builder struct.

I think this latest commit correctly adds the lifetime bound, e.g., I ran this test based on yours:

#[cfg(test)]
mod test {
    #[test]
    fn test_osssl_param_builder() {
        let mut p = crate::ossl_param::OsslParamBuilder::new().unwrap();
        {
            let key = std::ffi::CString::new(b"my str").unwrap();
            p.add_octet_string(&key, &[]).unwrap();
        }
        p.to_param().unwrap();
    }
}

Before, it compiled (incorrectly), but now it fails to compile as expected with:

error[E0597]: `key` does not live long enough
   --> openssl/src/ossl_param.rs:162:32
    |
161 |             let key = std::ffi::CString::new(b"my str").unwrap();
    |                 --- binding `key` declared here
162 |             p.add_octet_string(&key, &[]).unwrap();
    |                                ^^^^ borrowed value does not live long enough
163 |         }
    |         - `key` dropped here while still borrowed
164 |         p.to_param().unwrap();
    |         - borrow later used here

[alex]

Great, will review later this afternoon.

[alex]

thanks for your patience here

[swenson]

Thanks!