// Copyright (c) 2013-2017 The btcsuite developers // Use of this source code is governed by an ISC // license that can be found in the LICENSE file. package btcec import ( "bytes" "crypto/ecdsa" "crypto/elliptic" "crypto/hmac" "crypto/sha256" "errors" "fmt" "hash" "math/big" ) // Errors returned by canonicalPadding. var ( errNegativeValue = errors.New("value may be interpreted as negative") errExcessivelyPaddedValue = errors.New("value is excessively padded") ) // Signature is a type representing an ecdsa signature. type Signature struct { R *big.Int S *big.Int } var ( // Used in RFC6979 implementation when testing the nonce for correctness one = big.NewInt(1) // oneInitializer is used to fill a byte slice with byte 0x01. It is provided // here to avoid the need to create it multiple times. oneInitializer = []byte{0x01} ) // Serialize returns the ECDSA signature in the more strict DER format. Note // that the serialized bytes returned do not include the appended hash type // used in Bitcoin signature scripts. // // encoding/asn1 is broken so we hand roll this output: // // 0x30 0x02 r 0x02 s func (sig *Signature) Serialize() []byte { // low 'S' malleability breaker sigS := sig.S if sigS.Cmp(S256().halfOrder) == 1 { sigS = new(big.Int).Sub(S256().N, sigS) } // Ensure the encoded bytes for the r and s values are canonical and // thus suitable for DER encoding. rb := canonicalizeInt(sig.R) sb := canonicalizeInt(sigS) // total length of returned signature is 1 byte for each magic and // length (6 total), plus lengths of r and s length := 6 + len(rb) + len(sb) b := make([]byte, length) b[0] = 0x30 b[1] = byte(length - 2) b[2] = 0x02 b[3] = byte(len(rb)) offset := copy(b[4:], rb) + 4 b[offset] = 0x02 b[offset+1] = byte(len(sb)) copy(b[offset+2:], sb) return b } // Verify calls ecdsa.Verify to verify the signature of hash using the public // key. It returns true if the signature is valid, false otherwise. func (sig *Signature) Verify(hash []byte, pubKey *PublicKey) bool { return ecdsa.Verify(pubKey.ToECDSA(), hash, sig.R, sig.S) } // IsEqual compares this Signature instance to the one passed, returning true // if both Signatures are equivalent. A signature is equivalent to another, if // they both have the same scalar value for R and S. func (sig *Signature) IsEqual(otherSig *Signature) bool { return sig.R.Cmp(otherSig.R) == 0 && sig.S.Cmp(otherSig.S) == 0 } func parseSig(sigStr []byte, curve elliptic.Curve, der bool) (*Signature, error) { // Originally this code used encoding/asn1 in order to parse the // signature, but a number of problems were found with this approach. // Despite the fact that signatures are stored as DER, the difference // between go's idea of a bignum (and that they have sign) doesn't agree // with the openssl one (where they do not). The above is true as of // Go 1.1. In the end it was simpler to rewrite the code to explicitly // understand the format which is this: // 0x30 <0x02> 0x2 // . signature := &Signature{} // minimal message is when both numbers are 1 bytes. adding up to: // 0x30 + len + 0x02 + 0x01 + + 0x2 + 0x01 + if len(sigStr) < 8 { return nil, errors.New("malformed signature: too short") } // 0x30 index := 0 if sigStr[index] != 0x30 { return nil, errors.New("malformed signature: no header magic") } index++ // length of remaining message siglen := sigStr[index] index++ if int(siglen+2) > len(sigStr) { return nil, errors.New("malformed signature: bad length") } // trim the slice we're working on so we only look at what matters. sigStr = sigStr[:siglen+2] // 0x02 if sigStr[index] != 0x02 { return nil, errors.New("malformed signature: no 1st int marker") } index++ // Length of signature R. rLen := int(sigStr[index]) // must be positive, must be able to fit in another 0x2, // hence the -3. We assume that the length must be at least one byte. index++ if rLen <= 0 || rLen > len(sigStr)-index-3 { return nil, errors.New("malformed signature: bogus R length") } // Then R itself. rBytes := sigStr[index : index+rLen] if der { switch err := canonicalPadding(rBytes); err { case errNegativeValue: return nil, errors.New("signature R is negative") case errExcessivelyPaddedValue: return nil, errors.New("signature R is excessively padded") } } signature.R = new(big.Int).SetBytes(rBytes) index += rLen // 0x02. length already checked in previous if. if sigStr[index] != 0x02 { return nil, errors.New("malformed signature: no 2nd int marker") } index++ // Length of signature S. sLen := int(sigStr[index]) index++ // S should be the rest of the string. if sLen <= 0 || sLen > len(sigStr)-index { return nil, errors.New("malformed signature: bogus S length") } // Then S itself. sBytes := sigStr[index : index+sLen] if der { switch err := canonicalPadding(sBytes); err { case errNegativeValue: return nil, errors.New("signature S is negative") case errExcessivelyPaddedValue: return nil, errors.New("signature S is excessively padded") } } signature.S = new(big.Int).SetBytes(sBytes) index += sLen // sanity check length parsing if index != len(sigStr) { return nil, fmt.Errorf("malformed signature: bad final length %v != %v", index, len(sigStr)) } // Verify also checks this, but we can be more sure that we parsed // correctly if we verify here too. // FWIW the ecdsa spec states that R and S must be | 1, N - 1 | // but crypto/ecdsa only checks for Sign != 0. Mirror that. if signature.R.Sign() != 1 { return nil, errors.New("signature R isn't 1 or more") } if signature.S.Sign() != 1 { return nil, errors.New("signature S isn't 1 or more") } if signature.R.Cmp(curve.Params().N) >= 0 { return nil, errors.New("signature R is >= curve.N") } if signature.S.Cmp(curve.Params().N) >= 0 { return nil, errors.New("signature S is >= curve.N") } return signature, nil } // ParseSignature parses a signature in BER format for the curve type `curve' // into a Signature type, perfoming some basic sanity checks. If parsing // according to the more strict DER format is needed, use ParseDERSignature. func ParseSignature(sigStr []byte, curve elliptic.Curve) (*Signature, error) { return parseSig(sigStr, curve, false) } // ParseDERSignature parses a signature in DER format for the curve type // `curve` into a Signature type. If parsing according to the less strict // BER format is needed, use ParseSignature. func ParseDERSignature(sigStr []byte, curve elliptic.Curve) (*Signature, error) { return parseSig(sigStr, curve, true) } // canonicalizeInt returns the bytes for the passed big integer adjusted as // necessary to ensure that a big-endian encoded integer can't possibly be // misinterpreted as a negative number. This can happen when the most // significant bit is set, so it is padded by a leading zero byte in this case. // Also, the returned bytes will have at least a single byte when the passed // value is 0. This is required for DER encoding. func canonicalizeInt(val *big.Int) []byte { b := val.Bytes() if len(b) == 0 { b = []byte{0x00} } if b[0]&0x80 != 0 { paddedBytes := make([]byte, len(b)+1) copy(paddedBytes[1:], b) b = paddedBytes } return b } // canonicalPadding checks whether a big-endian encoded integer could // possibly be misinterpreted as a negative number (even though OpenSSL // treats all numbers as unsigned), or if there is any unnecessary // leading zero padding. func canonicalPadding(b []byte) error { switch { case b[0]&0x80 == 0x80: return errNegativeValue case len(b) > 1 && b[0] == 0x00 && b[1]&0x80 != 0x80: return errExcessivelyPaddedValue default: return nil } } // hashToInt converts a hash value to an integer. There is some disagreement // about how this is done. [NSA] suggests that this is done in the obvious // manner, but [SECG] truncates the hash to the bit-length of the curve order // first. We follow [SECG] because that's what OpenSSL does. Additionally, // OpenSSL right shifts excess bits from the number if the hash is too large // and we mirror that too. // This is borrowed from crypto/ecdsa. func hashToInt(hash []byte, c elliptic.Curve) *big.Int { orderBits := c.Params().N.BitLen() orderBytes := (orderBits + 7) / 8 if len(hash) > orderBytes { hash = hash[:orderBytes] } ret := new(big.Int).SetBytes(hash) excess := len(hash)*8 - orderBits if excess > 0 { ret.Rsh(ret, uint(excess)) } return ret } // recoverKeyFromSignature recoves a public key from the signature "sig" on the // given message hash "msg". Based on the algorithm found in section 5.1.5 of // SEC 1 Ver 2.0, page 47-48 (53 and 54 in the pdf). This performs the details // in the inner loop in Step 1. The counter provided is actually the j parameter // of the loop * 2 - on the first iteration of j we do the R case, else the -R // case in step 1.6. This counter is used in the bitcoin compressed signature // format and thus we match bitcoind's behaviour here. func recoverKeyFromSignature(curve *KoblitzCurve, sig *Signature, msg []byte, iter int, doChecks bool) (*PublicKey, error) { // 1.1 x = (n * i) + r Rx := new(big.Int).Mul(curve.Params().N, new(big.Int).SetInt64(int64(iter/2))) Rx.Add(Rx, sig.R) if Rx.Cmp(curve.Params().P) != -1 { return nil, errors.New("calculated Rx is larger than curve P") } // convert 02 to point R. (step 1.2 and 1.3). If we are on an odd // iteration then 1.6 will be done with -R, so we calculate the other // term when uncompressing the point. Ry, err := decompressPoint(curve, Rx, iter%2 == 1) if err != nil { return nil, err } // 1.4 Check n*R is point at infinity if doChecks { nRx, nRy := curve.ScalarMult(Rx, Ry, curve.Params().N.Bytes()) if nRx.Sign() != 0 || nRy.Sign() != 0 { return nil, errors.New("n*R does not equal the point at infinity") } } // 1.5 calculate e from message using the same algorithm as ecdsa // signature calculation. e := hashToInt(msg, curve) // Step 1.6.1: // We calculate the two terms sR and eG separately multiplied by the // inverse of r (from the signature). We then add them to calculate // Q = r^-1(sR-eG) invr := new(big.Int).ModInverse(sig.R, curve.Params().N) // first term. invrS := new(big.Int).Mul(invr, sig.S) invrS.Mod(invrS, curve.Params().N) sRx, sRy := curve.ScalarMult(Rx, Ry, invrS.Bytes()) // second term. e.Neg(e) e.Mod(e, curve.Params().N) e.Mul(e, invr) e.Mod(e, curve.Params().N) minuseGx, minuseGy := curve.ScalarBaseMult(e.Bytes()) // TODO: this would be faster if we did a mult and add in one // step to prevent the jacobian conversion back and forth. Qx, Qy := curve.Add(sRx, sRy, minuseGx, minuseGy) return &PublicKey{ Curve: curve, X: Qx, Y: Qy, }, nil } // SignCompact produces a compact signature of the data in hash with the given // private key on the given koblitz curve. The isCompressed parameter should // be used to detail if the given signature should reference a compressed // public key or not. If successful the bytes of the compact signature will be // returned in the format: // <(byte of 27+public key solution)+4 if compressed >< padded bytes for signature R> // where the R and S parameters are padde up to the bitlengh of the curve. func SignCompact(curve *KoblitzCurve, key *PrivateKey, hash []byte, isCompressedKey bool) ([]byte, error) { sig, err := key.Sign(hash) if err != nil { return nil, err } // bitcoind checks the bit length of R and S here. The ecdsa signature // algorithm returns R and S mod N therefore they will be the bitsize of // the curve, and thus correctly sized. for i := 0; i < (curve.H+1)*2; i++ { pk, err := recoverKeyFromSignature(curve, sig, hash, i, true) if err == nil && pk.X.Cmp(key.X) == 0 && pk.Y.Cmp(key.Y) == 0 { result := make([]byte, 1, 2*curve.byteSize+1) result[0] = 27 + byte(i) if isCompressedKey { result[0] += 4 } // Not sure this needs rounding but safer to do so. curvelen := (curve.BitSize + 7) / 8 // Pad R and S to curvelen if needed. bytelen := (sig.R.BitLen() + 7) / 8 if bytelen < curvelen { result = append(result, make([]byte, curvelen-bytelen)...) } result = append(result, sig.R.Bytes()...) bytelen = (sig.S.BitLen() + 7) / 8 if bytelen < curvelen { result = append(result, make([]byte, curvelen-bytelen)...) } result = append(result, sig.S.Bytes()...) return result, nil } } return nil, errors.New("no valid solution for pubkey found") } // RecoverCompact verifies the compact signature "signature" of "hash" for the // Koblitz curve in "curve". If the signature matches then the recovered public // key will be returned as well as a boolen if the original key was compressed // or not, else an error will be returned. func RecoverCompact(curve *KoblitzCurve, signature, hash []byte) (*PublicKey, bool, error) { bitlen := (curve.BitSize + 7) / 8 if len(signature) != 1+bitlen*2 { return nil, false, errors.New("invalid compact signature size") } iteration := int((signature[0] - 27) & ^byte(4)) // format is
sig := &Signature{ R: new(big.Int).SetBytes(signature[1 : bitlen+1]), S: new(big.Int).SetBytes(signature[bitlen+1:]), } // The iteration used here was encoded key, err := recoverKeyFromSignature(curve, sig, hash, iteration, false) if err != nil { return nil, false, err } return key, ((signature[0] - 27) & 4) == 4, nil } // signRFC6979 generates a deterministic ECDSA signature according to RFC 6979 and BIP 62. func signRFC6979(privateKey *PrivateKey, hash []byte) (*Signature, error) { privkey := privateKey.ToECDSA() N := S256().N halfOrder := S256().halfOrder k := nonceRFC6979(privkey.D, hash) inv := new(big.Int).ModInverse(k, N) r, _ := privkey.Curve.ScalarBaseMult(k.Bytes()) if r.Cmp(N) == 1 { r.Sub(r, N) } if r.Sign() == 0 { return nil, errors.New("calculated R is zero") } e := hashToInt(hash, privkey.Curve) s := new(big.Int).Mul(privkey.D, r) s.Add(s, e) s.Mul(s, inv) s.Mod(s, N) if s.Cmp(halfOrder) == 1 { s.Sub(N, s) } if s.Sign() == 0 { return nil, errors.New("calculated S is zero") } return &Signature{R: r, S: s}, nil } // nonceRFC6979 generates an ECDSA nonce (`k`) deterministically according to RFC 6979. // It takes a 32-byte hash as an input and returns 32-byte nonce to be used in ECDSA algorithm. func nonceRFC6979(privkey *big.Int, hash []byte) *big.Int { curve := S256() q := curve.Params().N x := privkey alg := sha256.New qlen := q.BitLen() holen := alg().Size() rolen := (qlen + 7) >> 3 bx := append(int2octets(x, rolen), bits2octets(hash, curve, rolen)...) // Step B v := bytes.Repeat(oneInitializer, holen) // Step C (Go zeroes the all allocated memory) k := make([]byte, holen) // Step D k = mac(alg, k, append(append(v, 0x00), bx...)) // Step E v = mac(alg, k, v) // Step F k = mac(alg, k, append(append(v, 0x01), bx...)) // Step G v = mac(alg, k, v) // Step H for { // Step H1 var t []byte // Step H2 for len(t)*8 < qlen { v = mac(alg, k, v) t = append(t, v...) } // Step H3 secret := hashToInt(t, curve) if secret.Cmp(one) >= 0 && secret.Cmp(q) < 0 { return secret } k = mac(alg, k, append(v, 0x00)) v = mac(alg, k, v) } } // mac returns an HMAC of the given key and message. func mac(alg func() hash.Hash, k, m []byte) []byte { h := hmac.New(alg, k) h.Write(m) return h.Sum(nil) } // https://tools.ietf.org/html/rfc6979#section-2.3.3 func int2octets(v *big.Int, rolen int) []byte { out := v.Bytes() // left pad with zeros if it's too short if len(out) < rolen { out2 := make([]byte, rolen) copy(out2[rolen-len(out):], out) return out2 } // drop most significant bytes if it's too long if len(out) > rolen { out2 := make([]byte, rolen) copy(out2, out[len(out)-rolen:]) return out2 } return out } // https://tools.ietf.org/html/rfc6979#section-2.3.4 func bits2octets(in []byte, curve elliptic.Curve, rolen int) []byte { z1 := hashToInt(in, curve) z2 := new(big.Int).Sub(z1, curve.Params().N) if z2.Sign() < 0 { return int2octets(z1, rolen) } return int2octets(z2, rolen) }