Making Chinese Cryptography Accessible for All

Digital Signature – Usage of Chinese Cryptography Standards

Contributed by Pierre Gaulon, Senior Cloud Infrastructure & Security Engineer at Thunes

I have recently studied the Chinese cryptography standards published by the Chinese Commercial Cryptography Administration Office: SM2, SM3, and SM4 (SM stands for ShangMi).

  • SM2 provides signature and verification based on Elliptic Curve Cryptography (ECC) and was the interesting point of research,
  • SM3 is a hashing algorithm
  • SM4 is a set of encryption/decryption block-cipher algorithms

As a lot of the SM2, SM3, and SM4 documentation and code usage is written in Mandarin, this presents a significant barrier to entry for beginners. Furthermore, it is time-consuming to develop a solution. I have therefore penned this article to guide developers who have to adhere to such cryptography standards in their work.

As the article is long, you can jump directly to the section you are interested in:

  1. Elliptic Curve Cryptography (ECC)
    1.1 Public-key Cryptography
    1.2 High-level Functions
    1.3 Theory Summary
    1.3.1 Key Generation
    1.3.2 Digital Signature
  2. Back to SM2/SM3/SM4
    2.1 How to Get Keys
    2.2 How to Sign with SM2
    2.3 Modifications to tjfoc Implementation
    2.4 How to reverse engineer Java functions
  3. Key takeaways

Elliptic Curve Cryptography (ECC)

ECC is one of the approaches to public-key cryptography.

Public-key Cryptography

Public-key cryptography relies on the generation of two keys:

  • one private key which must remain private
  • one public key which can be shared with the world

It is impossible to know a private key from a public key (it takes more than centuries to compute). It is possible to prove the possession of a private key without disclosing it. This proof can be verified by using its corresponding public key. This proof is called a digital signature.

High-level Functions

ECC can perform signature and verification of messages (authenticity). ECC can also perform encryption and decryption (confidentiality), however, not directly. For encryption/decryption, it needs the help of a shared secret, namely a key.

ECC delivers the same protection as RSA (Rivest-Shamir-Adleman) with a smaller key size. Smaller keys mean fewer operations and less storage. For instance, an ECC 256 bits key should provide the same level of security as an RSA key of 3072 bits.

Theory Summary

Following the very well explained guides from Hans Knutson on Hacker Noon and CryptoBook from Svetlin Nakov, this section aims to provide a sense of different parameters found in SM2 libraries and what they correspond to.

Key Generation

As a comparison, RSA is based on prime number factorization, and its private key is composed of two long prime numbers (called p and q). The modulus m is the product pq=m which constitutes the public key. The size of m in bits is the key size of RSA. From the knowledge of m, it is hard to decompose it back into the two prime numbers, p and q.

ECC is based on the discrete logarithm of elliptic curve elements. An elliptic curve consists of all the points of coordinates (x,y) verifying y² = x³+ax+b. For instance, Bitcoin uses the curve called secp256k1, which verifies y² = x³+7. It is possible to add two points of that curve together:

Or to add a point to itself:

Moreover, to make sure that points stay within reasonable coordinates, the curve y² = x³+ax+b is wrapped around itself using a modulus p: y² mod p = (x³ + ax + b) mod p

The ECC private key is given by choosing a base point P (also called Generator point G) on the wrapped curve and adding itself x times (which defines the operation • as x•P = P + … + P, x times), with x a random 256 bits integer. The resulting point from x•P will be called X and is fast to compute thanks to exponentiation by squaring. The coordinates of the base point P are known for a type of curve, and the coordinates of X are the public key. The private key is x which by definition has a size of 256 bits. From the endpoint X, it is hard to find the number of iterations x.

Digital Signature

The signature algorithm for ECC is called Elliptic Curve Digital Signature Algorithm, or ECDSA. Without going into the details, signing a message m is done by first hashing it. A random integer k is chosen and used to multiply the base point P. The result of that signature consists of two elements: r and s.

  • r is the horizontal coordinate of k•P
  • s is computed from the hash of the message m, r, and the private key x

An important note is that since there is a random number k in the signature, two signatures of the same message m will not look the same.

Back to SM2/SM3/SM4

One of the main open-source implementations of SM2/SM3/SM4 algorithms is GmSSL (Gm stands for Guomi). Other implementations exist, such as gmsm in Golang, gmssl in Python, or Chinese Financial Certification Authority (CFCA) SADK in Java.

My goal was to port Java code to Golang: reverse-engineering the usage of CFCA SADK in this use case and adapt gmsm to it. The hashing algorithm SM3 and the encryption algorithm SM4 were used as-is and could be ported from one language to another using the equivalent functions.

From a classic REST API POST, with several parameters, a few additional security operations are taking place:

  • The original parameters are concatenated in alphabetical order, then concatenated to an API key, and hashed using SM3. The resulting hash is later added as an additional POST parameter
  • The original parameters are concatenated in alphabetical order and signed using SM2. That signature using a PKCS7-like format is attached to the request as an additional POST parameter
  • The response body is encrypted using SM4 with a key derived from the API key
  • The response body also contains both an SM3 hash and SM2 signature for verification

The second step was the most interesting as the Golang library was not implementing the PKCS7 formatting of the signature: only American standards were supported.

How to Get Keys

The private key used for SM2 signing was provided to us, along with a passphrase for testing purposes. Of course, in production systems, the private key is generated and kept private. The file extension is .sm2; the first step was to make use of it.

It can be parsed with:

$ openssl asn1parse -in file.sm2

    0:d=0  hl=4 l= 802 cons: SEQUENCE
    4:d=1  hl=2 l=   1 prim: INTEGER           :01
    7:d=1  hl=2 l=  71 cons: SEQUENCE
    9:d=2  hl=2 l=  10 prim: OBJECT            :
   21:d=2  hl=2 l=   7 prim: OBJECT            :
   30:d=2  hl=2 l=  48 prim: OCTET STRING      [HEX DUMP]:8[redacted]7
   80:d=1  hl=4 l= 722 cons: SEQUENCE
   84:d=2  hl=2 l=  10 prim: OBJECT            :
   96:d=2  hl=4 l= 706 prim: OCTET STRING      [HEX DUMP]:308[redacted]249

The OID means SM4 Block Cipher. The OID simply means data.

.sm2 files are an ASN.1 structure encoded in DER and base64-ed. The ASN.1 structure contains (int, seq1, seq2). Seq1 contains the SM4-encrypted SM2 private key x. Seq2 contains the x509 cert of the corresponding SM2 public key (ECC coordinates (x,y) of the point X). From the private key x, it is also possible to get X=x•P.

The x509 certificate is signed by CFCA, and the signature algorithm means SM2 Signing with SM3.

$ openssl x509 -inform der -text -noout -in publickey-test.cer
        Version: 3 (0x2)
        Serial Number: 275466457874 (0x4023149312)
        Signature Algorithm:
        Issuer: C = CN, O = China Financial Certification Authority, CN = CFCA ACS TEST SM2 OCA31
            Not Before: Mar 25 08:39:36 2020 GMT
            Not After : Mar 25 08:39:36 2025 GMT
        Subject: C = CN, O = OCA31SM2, OU = [redacted], OU = Organizational-1, CN = [redacted]
        Subject Public Key Info:
            Public Key Algorithm: id-ecPublicKey
                Public-Key: (256 bit)
                ASN1 OID: SM2
        X509v3 extensions:
            Authority Information Access:
                OCSP - URI:

            X509v3 Authority Key Identifier:

            X509v3 Basic Constraints: critical
            X509v3 CRL Distribution Points:

                Full Name:

            X509v3 Key Usage: critical
                Digital Signature, Non Repudiation
            X509v3 Subject Key Identifier:
            X509v3 Extended Key Usage:
                TLS Web Client Authentication, E-mail Protection
    Signature Algorithm:

The equivalent Java code to read the private key x from a .sm2 file is:

import cfca.sadk.util.KeyUtil;
import cfca.sadk.algorithm.sm2.SM2PrivateKey;


SM2PrivateKey privKey = KeyUtil.getPrivateKeyFromSM2("file.sm2", "passphrase");

A utility to generate a random private key is also provided and can be used for production key generation. In there the SM2 elliptic curve parameters y² mod p = (x³ + ax + b) mod p can be found:

  • b = 28E9FA9E 9D9F5E34 4D5A9E4B CF6509A7 F39789F5 15AB8F92 DDBCBD41 4D940E93
  • the base point x coordinate Gx = 32C4AE2C 1F198119 5F990446 6A39C994 8FE30BBF F2660BE1 715A4589 334C74C7
  • the base point y coordinate Gy = BC3736A2 F4F6779C 59BDCEE3 6B692153 D0A9877C C62A4740 02DF32E5 2139F0A0

How to Sign with SM2

Now that the private key x is known, it is possible to use it to sign the concatenation of parameters and return the PKCS7 format expected.

As a reminder, ECC Digital Signature Algorithm takes a random number k. This is why it is important to add a random generator to the signing function. It is also difficult to troubleshoot: signing the same message twice will provide different outputs.

The signature will return two integers, r and s, as defined previously.

The format returned is PKCS7, which is structured with ASN.1. The asn1js tool is perfect for reading and comparing ASN.1 structures. For maximum privacy, it should be cloned and used locally.

The ASN.1 structure of the signature will follow:

  • The algorithm used as hash, namely (sm3Hash)
  • The data that is signed, with OID (data)
  • A sequence of the x509 certificates corresponding to the private keys used to sign (we can sign with multiple keys)
  • A set of the digital signatures for all the keys/certificates signing. Each signature is a sequence of the corresponding certificate information (countryName, organizationName, commonName) and finally the two integer r and s, in hexadecimal representation (the last two elements in the figure below)
Screenshot from asn1js

To generate such signature, the Golang equivalent is:

import (
	"" // modified PKCS7


	PRIVATE, _ := hex.DecodeString("somehexhere")
	PUBLICX, _ := hex.DecodeString("6de24a97f67c0c8424d993f42854f9003bde6997ed8726335f8d300c34be8321")
	PUBLICY, _ := hex.DecodeString("b177aeb12930141f02aed9f97b70b5a7c82a63d294787a15a6944b591ae74469")

	priv := new(sm2.PrivateKey)
	priv.D = new(big.Int).SetBytes(PRIVATE)
	priv.PublicKey.X = new(big.Int).SetBytes(PUBLICX)
	priv.PublicKey.Y = new(big.Int).SetBytes(PUBLICY)
	priv.PublicKey.Curve = sm2.P256Sm2()

	cert := getCertFromSM2(sm2CertPath) // utility to provision a x509 object from the .sm2 file data
	sign, _ := priv.Sign(rand.Reader, []byte(toSign), nil)
	signedData, _ := x509.NewSignedData([]byte(toSign))
	signerInfoConf := x509.SignerInfoConfig{}
	signedData.AddSigner(cert, priv, signerInfoConf, sign)
	pkcs7SignedBytes, _ := signedData.Finish()
	return base64.StdEncoding.EncodeToString(pkcs7SignedBytes)

Modifications to tjfoc Implementation

tjfoc PKCS7 utility was missing a few things to fit the expected format. As such, a fork has been created to accommodate it.

Disclaimer time. The resulting code is very specific and by no means perfect nor reusable: it just works and doesn’t do anything more than that.

  • The OIDs were changed to use SM2/SM3 OIDs
  • An extra parameter to AddSigner was created to append the signature (r and s) to the x509 signer info. This also removes extra attributes which mangle the shape of the resulting PKCS7
  • Extra fields were removed, and using SM3 hash with SM2 signature was allowed. From there, Chinese standard is accepted, along with the US ones

How to Reverse Engineer Java Functions

To debug Java, it is better to have a clean environment to start with:

  • a Vagrant config, to run a VM with vagrant up
Vagrant.configure("2") do |config| = "ubuntu/bionic64"
  config.vm.synced_folder ".", "/vagrant_data"
  • a java class with the high-level code to inspect
import java.nio.file.Files;
import java.nio.file.Paths;
import com.[redacted].sdk.service.impl.OpenApiSecurityService;

class HelloWorld {
    public static void main(String[] args) throws Exception {
        try {
            byte[] sm2Cert = Files.readAllBytes(Paths.get("certs-test/test.sm2"));
            OpenApiSecurityService securityService = new OpenApiSecurityService();
            String cfcaSign = securityService.cfcaSignature(signResource, "somepassowrd", sm2Cert);
        catch(IOException e) {
  • a compilation script to compile any change from the code

javac -Xlint:deprecation -cp "JAR/*"
java -cp "JAR/*:." HelloWorld
  • all the .jar to run the code
  • decompiled .java sources from the .jar files (only the ones to inspect are needed)
  • the .sm2 file provided

Once the Java code is running in the VM, jdb is used:

vagrant@ubuntu-bionic:/vagrant_data$ jdb -classpath "JAR/*:." -sourcepath sources/ HelloWorld
Initializing jdb ...
> stop in com.[redacted].commons.cfca.SignVerUtils.signature
Deferring breakpoint com.[redacted].commons.cfca.SignVerUtils.signature.
It will be set after the class is loaded.
> run
run HelloWorld
Set uncaught java.lang.Throwable
Set deferred uncaught java.lang.Throwable
VM Started: Set deferred breakpoint com.[redacted].commons.cfca.SignVerUtils.signature

Breakpoint hit: "thread=main", com.[redacted].commons.cfca.SignVerUtils.signature(), line=28 bci=0
28    /* 28 */     X509Cert x509Cert = SM2CertUtils.getX509CertFromSm2(sm2CertData);

main[1] where
  [1] com.[redacted].commons.cfca.SignVerUtils.signature (
  [2] com.[redacted].sdk.service.impl.OpenApiSecurityService.cfcaSignature (
  [3] HelloWorld.main (
main[1] list
24    /*    */
25    /*    */
26    /*    */
27    /*    */   public static byte[] signature(byte[] sm2CertData, String sm2FilePass, byte[] sourceData, Session session) throws PKIException, UnsupportedEncodingException {
28 => /* 28 */     X509Cert x509Cert = SM2CertUtils.getX509CertFromSm2(sm2CertData);
29    /*    */
30    /* 30 */     PrivateKey privateKey = SM2CertUtils.getPrivateKeyFromSm2(sm2CertData, sm2FilePass);
31    /*    */
32    /* 32 */     Signature signKit = new Signature();
33    /* 33 */     String signAlg = "sm3WithSM2Encryption";
main[1] step
Step completed: "thread=main", com.[redacted].commons.cfca.SM2CertUtils.getX509CertFromSm2(), line=67 bci=0
67    /* 67 */     return CertUtil.getCertFromSM2(certBytes);

The commands where, list, step or next, can be repeated until the code is understood. Keeping jd-gui open beside it to follow classes is also very helpful.

Screenshot from jd-gui on the signature method

Key takeaways

  • Chinese standard SM2 uses Elliptic Curve Cryptography
  • ECC public key is a point X reached by adding x times the base point P to itself. x is the private key
  • ECC signatures use a random number as input, thus varying even with the same message. A signature is composed of two numbers, r and s
  • To troubleshoot ASN.1, use asn1js
  • To troubleshoot Java, use jdb and jd-gui
  • Cryptography is hard!