326f77ca2b
Rebuild
7751 lines
268 KiB
JavaScript
7751 lines
268 KiB
JavaScript
'use strict';
|
|
|
|
Object.defineProperty(exports, '__esModule', { value: true });
|
|
|
|
var binFileUtils = require('@iden3/binfileutils');
|
|
var ffjavascript = require('ffjavascript');
|
|
var Blake2b = require('blake2b-wasm');
|
|
var readline = require('readline');
|
|
var crypto = require('crypto');
|
|
var fastFile = require('fastfile');
|
|
var circom_runtime = require('circom_runtime');
|
|
var r1csfile = require('r1csfile');
|
|
var ejs = require('ejs');
|
|
var jsSha3 = require('js-sha3');
|
|
|
|
function _interopDefaultLegacy (e) { return e && typeof e === 'object' && 'default' in e ? e : { 'default': e }; }
|
|
|
|
function _interopNamespace(e) {
|
|
if (e && e.__esModule) return e;
|
|
var n = Object.create(null);
|
|
if (e) {
|
|
Object.keys(e).forEach(function (k) {
|
|
if (k !== 'default') {
|
|
var d = Object.getOwnPropertyDescriptor(e, k);
|
|
Object.defineProperty(n, k, d.get ? d : {
|
|
enumerable: true,
|
|
get: function () { return e[k]; }
|
|
});
|
|
}
|
|
});
|
|
}
|
|
n["default"] = e;
|
|
return Object.freeze(n);
|
|
}
|
|
|
|
var binFileUtils__namespace = /*#__PURE__*/_interopNamespace(binFileUtils);
|
|
var Blake2b__default = /*#__PURE__*/_interopDefaultLegacy(Blake2b);
|
|
var readline__default = /*#__PURE__*/_interopDefaultLegacy(readline);
|
|
var crypto__default = /*#__PURE__*/_interopDefaultLegacy(crypto);
|
|
var fastFile__namespace = /*#__PURE__*/_interopNamespace(fastFile);
|
|
var ejs__default = /*#__PURE__*/_interopDefaultLegacy(ejs);
|
|
var jsSha3__default = /*#__PURE__*/_interopDefaultLegacy(jsSha3);
|
|
|
|
ffjavascript.Scalar.e("73eda753299d7d483339d80809a1d80553bda402fffe5bfeffffffff00000001", 16);
|
|
ffjavascript.Scalar.e("21888242871839275222246405745257275088548364400416034343698204186575808495617");
|
|
|
|
const bls12381q = ffjavascript.Scalar.e("1a0111ea397fe69a4b1ba7b6434bacd764774b84f38512bf6730d2a0f6b0f6241eabfffeb153ffffb9feffffffffaaab", 16);
|
|
const bn128q = ffjavascript.Scalar.e("21888242871839275222246405745257275088696311157297823662689037894645226208583");
|
|
|
|
async function getCurveFromQ(q) {
|
|
let curve;
|
|
if (ffjavascript.Scalar.eq(q, bn128q)) {
|
|
curve = await ffjavascript.buildBn128();
|
|
} else if (ffjavascript.Scalar.eq(q, bls12381q)) {
|
|
curve = await ffjavascript.buildBls12381();
|
|
} else {
|
|
throw new Error(`Curve not supported: ${ffjavascript.Scalar.toString(q)}`);
|
|
}
|
|
return curve;
|
|
}
|
|
|
|
async function getCurveFromName(name) {
|
|
let curve;
|
|
const normName = normalizeName(name);
|
|
if (["BN128", "BN254", "ALTBN128"].indexOf(normName) >= 0) {
|
|
curve = await ffjavascript.buildBn128();
|
|
} else if (["BLS12381"].indexOf(normName) >= 0) {
|
|
curve = await ffjavascript.buildBls12381();
|
|
} else {
|
|
throw new Error(`Curve not supported: ${name}`);
|
|
}
|
|
return curve;
|
|
|
|
function normalizeName(n) {
|
|
return n.toUpperCase().match(/[A-Za-z0-9]+/g).join("");
|
|
}
|
|
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
const _revTable = [];
|
|
for (let i=0; i<256; i++) {
|
|
_revTable[i] = _revSlow(i, 8);
|
|
}
|
|
|
|
function _revSlow(idx, bits) {
|
|
let res =0;
|
|
let a = idx;
|
|
for (let i=0; i<bits; i++) {
|
|
res <<= 1;
|
|
res = res | (a &1);
|
|
a >>=1;
|
|
}
|
|
return res;
|
|
}
|
|
|
|
|
|
function log2( V )
|
|
{
|
|
return( ( ( V & 0xFFFF0000 ) !== 0 ? ( V &= 0xFFFF0000, 16 ) : 0 ) | ( ( V & 0xFF00FF00 ) !== 0 ? ( V &= 0xFF00FF00, 8 ) : 0 ) | ( ( V & 0xF0F0F0F0 ) !== 0 ? ( V &= 0xF0F0F0F0, 4 ) : 0 ) | ( ( V & 0xCCCCCCCC ) !== 0 ? ( V &= 0xCCCCCCCC, 2 ) : 0 ) | ( ( V & 0xAAAAAAAA ) !== 0 ) );
|
|
}
|
|
|
|
|
|
function formatHash(b, title) {
|
|
const a = new DataView(b.buffer, b.byteOffset, b.byteLength);
|
|
let S = "";
|
|
for (let i=0; i<4; i++) {
|
|
if (i>0) S += "\n";
|
|
S += "\t\t";
|
|
for (let j=0; j<4; j++) {
|
|
if (j>0) S += " ";
|
|
S += a.getUint32(i*16+j*4).toString(16).padStart(8, "0");
|
|
}
|
|
}
|
|
if (title) S = title + "\n" + S;
|
|
return S;
|
|
}
|
|
|
|
function hashIsEqual(h1, h2) {
|
|
if (h1.byteLength != h2.byteLength) return false;
|
|
var dv1 = new Int8Array(h1);
|
|
var dv2 = new Int8Array(h2);
|
|
for (var i = 0 ; i != h1.byteLength ; i++)
|
|
{
|
|
if (dv1[i] != dv2[i]) return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
function cloneHasher(h) {
|
|
const ph = h.getPartialHash();
|
|
const res = Blake2b__default["default"](64);
|
|
res.setPartialHash(ph);
|
|
return res;
|
|
}
|
|
|
|
async function sameRatio$2(curve, g1s, g1sx, g2s, g2sx) {
|
|
if (curve.G1.isZero(g1s)) return false;
|
|
if (curve.G1.isZero(g1sx)) return false;
|
|
if (curve.G2.isZero(g2s)) return false;
|
|
if (curve.G2.isZero(g2sx)) return false;
|
|
// return curve.F12.eq(curve.pairing(g1s, g2sx), curve.pairing(g1sx, g2s));
|
|
const res = await curve.pairingEq(g1s, g2sx, curve.G1.neg(g1sx), g2s);
|
|
return res;
|
|
}
|
|
|
|
|
|
function askEntropy() {
|
|
if (process.browser) {
|
|
return window.prompt("Enter a random text. (Entropy): ", "");
|
|
} else {
|
|
const rl = readline__default["default"].createInterface({
|
|
input: process.stdin,
|
|
output: process.stdout
|
|
});
|
|
|
|
return new Promise((resolve) => {
|
|
rl.question("Enter a random text. (Entropy): ", (input) => resolve(input) );
|
|
});
|
|
}
|
|
}
|
|
|
|
async function getRandomRng(entropy) {
|
|
// Generate a random Rng
|
|
while (!entropy) {
|
|
entropy = await askEntropy();
|
|
}
|
|
const hasher = Blake2b__default["default"](64);
|
|
hasher.update(crypto__default["default"].randomBytes(64));
|
|
const enc = new TextEncoder(); // always utf-8
|
|
hasher.update(enc.encode(entropy));
|
|
const hash = Buffer.from(hasher.digest());
|
|
|
|
const seed = [];
|
|
for (let i=0;i<8;i++) {
|
|
seed[i] = hash.readUInt32BE(i*4);
|
|
}
|
|
const rng = new ffjavascript.ChaCha(seed);
|
|
return rng;
|
|
}
|
|
|
|
function rngFromBeaconParams(beaconHash, numIterationsExp) {
|
|
let nIterationsInner;
|
|
let nIterationsOuter;
|
|
if (numIterationsExp<32) {
|
|
nIterationsInner = (1 << numIterationsExp) >>> 0;
|
|
nIterationsOuter = 1;
|
|
} else {
|
|
nIterationsInner = 0x100000000;
|
|
nIterationsOuter = (1 << (numIterationsExp-32)) >>> 0;
|
|
}
|
|
|
|
let curHash = beaconHash;
|
|
for (let i=0; i<nIterationsOuter; i++) {
|
|
for (let j=0; j<nIterationsInner; j++) {
|
|
curHash = crypto__default["default"].createHash("sha256").update(curHash).digest();
|
|
}
|
|
}
|
|
|
|
const curHashV = new DataView(curHash.buffer, curHash.byteOffset, curHash.byteLength);
|
|
const seed = [];
|
|
for (let i=0; i<8; i++) {
|
|
seed[i] = curHashV.getUint32(i*4, false);
|
|
}
|
|
|
|
const rng = new ffjavascript.ChaCha(seed);
|
|
|
|
return rng;
|
|
}
|
|
|
|
function hex2ByteArray(s) {
|
|
if (s instanceof Uint8Array) return s;
|
|
if (s.slice(0,2) == "0x") s= s.slice(2);
|
|
return new Uint8Array(s.match(/[\da-f]{2}/gi).map(function (h) {
|
|
return parseInt(h, 16);
|
|
}));
|
|
}
|
|
|
|
function byteArray2hex(byteArray) {
|
|
return Array.prototype.map.call(byteArray, function(byte) {
|
|
return ("0" + (byte & 0xFF).toString(16)).slice(-2);
|
|
}).join("");
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function writeHeader(fd, zkey) {
|
|
|
|
// Write the header
|
|
///////////
|
|
await binFileUtils__namespace.startWriteSection(fd, 1);
|
|
await fd.writeULE32(1); // Groth
|
|
await binFileUtils__namespace.endWriteSection(fd);
|
|
|
|
// Write the Groth header section
|
|
///////////
|
|
|
|
const curve = await getCurveFromQ(zkey.q);
|
|
|
|
await binFileUtils__namespace.startWriteSection(fd, 2);
|
|
const primeQ = curve.q;
|
|
const n8q = (Math.floor( (ffjavascript.Scalar.bitLength(primeQ) - 1) / 64) +1)*8;
|
|
|
|
const primeR = curve.r;
|
|
const n8r = (Math.floor( (ffjavascript.Scalar.bitLength(primeR) - 1) / 64) +1)*8;
|
|
|
|
await fd.writeULE32(n8q);
|
|
await binFileUtils__namespace.writeBigInt(fd, primeQ, n8q);
|
|
await fd.writeULE32(n8r);
|
|
await binFileUtils__namespace.writeBigInt(fd, primeR, n8r);
|
|
await fd.writeULE32(zkey.nVars); // Total number of bars
|
|
await fd.writeULE32(zkey.nPublic); // Total number of public vars (not including ONE)
|
|
await fd.writeULE32(zkey.domainSize); // domainSize
|
|
await writeG1(fd, curve, zkey.vk_alpha_1);
|
|
await writeG1(fd, curve, zkey.vk_beta_1);
|
|
await writeG2(fd, curve, zkey.vk_beta_2);
|
|
await writeG2(fd, curve, zkey.vk_gamma_2);
|
|
await writeG1(fd, curve, zkey.vk_delta_1);
|
|
await writeG2(fd, curve, zkey.vk_delta_2);
|
|
|
|
await binFileUtils__namespace.endWriteSection(fd);
|
|
|
|
|
|
}
|
|
|
|
async function writeG1(fd, curve, p) {
|
|
const buff = new Uint8Array(curve.G1.F.n8*2);
|
|
curve.G1.toRprLEM(buff, 0, p);
|
|
await fd.write(buff);
|
|
}
|
|
|
|
async function writeG2(fd, curve, p) {
|
|
const buff = new Uint8Array(curve.G2.F.n8*2);
|
|
curve.G2.toRprLEM(buff, 0, p);
|
|
await fd.write(buff);
|
|
}
|
|
|
|
async function readG1(fd, curve, toObject) {
|
|
const buff = await fd.read(curve.G1.F.n8*2);
|
|
const res = curve.G1.fromRprLEM(buff, 0);
|
|
return toObject ? curve.G1.toObject(res) : res;
|
|
}
|
|
|
|
async function readG2(fd, curve, toObject) {
|
|
const buff = await fd.read(curve.G2.F.n8*2);
|
|
const res = curve.G2.fromRprLEM(buff, 0);
|
|
return toObject ? curve.G2.toObject(res) : res;
|
|
}
|
|
|
|
|
|
async function readHeader$1(fd, sections, toObject) {
|
|
// Read Header
|
|
/////////////////////
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 1);
|
|
const protocolId = await fd.readULE32();
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
if (protocolId == 1) {
|
|
return await readHeaderGroth16(fd, sections, toObject);
|
|
} else if (protocolId == 2) {
|
|
return await readHeaderPlonk(fd, sections);
|
|
} else {
|
|
throw new Error("Protocol not supported: ");
|
|
}
|
|
}
|
|
|
|
|
|
|
|
|
|
async function readHeaderGroth16(fd, sections, toObject) {
|
|
const zkey = {};
|
|
|
|
zkey.protocol = "groth16";
|
|
|
|
// Read Groth Header
|
|
/////////////////////
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 2);
|
|
const n8q = await fd.readULE32();
|
|
zkey.n8q = n8q;
|
|
zkey.q = await binFileUtils__namespace.readBigInt(fd, n8q);
|
|
|
|
const n8r = await fd.readULE32();
|
|
zkey.n8r = n8r;
|
|
zkey.r = await binFileUtils__namespace.readBigInt(fd, n8r);
|
|
|
|
let curve = await getCurveFromQ(zkey.q);
|
|
|
|
zkey.nVars = await fd.readULE32();
|
|
zkey.nPublic = await fd.readULE32();
|
|
zkey.domainSize = await fd.readULE32();
|
|
zkey.power = log2(zkey.domainSize);
|
|
zkey.vk_alpha_1 = await readG1(fd, curve, toObject);
|
|
zkey.vk_beta_1 = await readG1(fd, curve, toObject);
|
|
zkey.vk_beta_2 = await readG2(fd, curve, toObject);
|
|
zkey.vk_gamma_2 = await readG2(fd, curve, toObject);
|
|
zkey.vk_delta_1 = await readG1(fd, curve, toObject);
|
|
zkey.vk_delta_2 = await readG2(fd, curve, toObject);
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
return zkey;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
async function readHeaderPlonk(fd, sections, protocol, toObject) {
|
|
const zkey = {};
|
|
|
|
zkey.protocol = "plonk";
|
|
|
|
// Read Plonk Header
|
|
/////////////////////
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 2);
|
|
const n8q = await fd.readULE32();
|
|
zkey.n8q = n8q;
|
|
zkey.q = await binFileUtils__namespace.readBigInt(fd, n8q);
|
|
|
|
const n8r = await fd.readULE32();
|
|
zkey.n8r = n8r;
|
|
zkey.r = await binFileUtils__namespace.readBigInt(fd, n8r);
|
|
|
|
let curve = await getCurveFromQ(zkey.q);
|
|
|
|
zkey.nVars = await fd.readULE32();
|
|
zkey.nPublic = await fd.readULE32();
|
|
zkey.domainSize = await fd.readULE32();
|
|
zkey.power = log2(zkey.domainSize);
|
|
zkey.nAdditions = await fd.readULE32();
|
|
zkey.nConstrains = await fd.readULE32();
|
|
zkey.k1 = await fd.read(n8r);
|
|
zkey.k2 = await fd.read(n8r);
|
|
|
|
zkey.Qm = await readG1(fd, curve, toObject);
|
|
zkey.Ql = await readG1(fd, curve, toObject);
|
|
zkey.Qr = await readG1(fd, curve, toObject);
|
|
zkey.Qo = await readG1(fd, curve, toObject);
|
|
zkey.Qc = await readG1(fd, curve, toObject);
|
|
zkey.S1 = await readG1(fd, curve, toObject);
|
|
zkey.S2 = await readG1(fd, curve, toObject);
|
|
zkey.S3 = await readG1(fd, curve, toObject);
|
|
zkey.X_2 = await readG2(fd, curve, toObject);
|
|
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
return zkey;
|
|
}
|
|
|
|
async function readZKey(fileName, toObject) {
|
|
const {fd, sections} = await binFileUtils__namespace.readBinFile(fileName, "zkey", 1);
|
|
|
|
const zkey = await readHeader$1(fd, sections, "groth16");
|
|
|
|
const Fr = new ffjavascript.F1Field(zkey.r);
|
|
const Rr = ffjavascript.Scalar.mod(ffjavascript.Scalar.shl(1, zkey.n8r*8), zkey.r);
|
|
const Rri = Fr.inv(Rr);
|
|
const Rri2 = Fr.mul(Rri, Rri);
|
|
|
|
let curve = await getCurveFromQ(zkey.q);
|
|
|
|
// Read IC Section
|
|
///////////
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 3);
|
|
zkey.IC = [];
|
|
for (let i=0; i<= zkey.nPublic; i++) {
|
|
const P = await readG1(fd, curve, toObject);
|
|
zkey.IC.push(P);
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
|
|
// Read Coefs
|
|
///////////
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 4);
|
|
const nCCoefs = await fd.readULE32();
|
|
zkey.ccoefs = [];
|
|
for (let i=0; i<nCCoefs; i++) {
|
|
const m = await fd.readULE32();
|
|
const c = await fd.readULE32();
|
|
const s = await fd.readULE32();
|
|
const v = await readFr2();
|
|
zkey.ccoefs.push({
|
|
matrix: m,
|
|
constraint: c,
|
|
signal: s,
|
|
value: v
|
|
});
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
// Read A points
|
|
///////////
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 5);
|
|
zkey.A = [];
|
|
for (let i=0; i<zkey.nVars; i++) {
|
|
const A = await readG1(fd, curve, toObject);
|
|
zkey.A[i] = A;
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
|
|
// Read B1
|
|
///////////
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 6);
|
|
zkey.B1 = [];
|
|
for (let i=0; i<zkey.nVars; i++) {
|
|
const B1 = await readG1(fd, curve, toObject);
|
|
|
|
zkey.B1[i] = B1;
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
|
|
// Read B2 points
|
|
///////////
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 7);
|
|
zkey.B2 = [];
|
|
for (let i=0; i<zkey.nVars; i++) {
|
|
const B2 = await readG2(fd, curve, toObject);
|
|
zkey.B2[i] = B2;
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
|
|
// Read C points
|
|
///////////
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 8);
|
|
zkey.C = [];
|
|
for (let i=zkey.nPublic+1; i<zkey.nVars; i++) {
|
|
const C = await readG1(fd, curve, toObject);
|
|
|
|
zkey.C[i] = C;
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
|
|
// Read H points
|
|
///////////
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 9);
|
|
zkey.hExps = [];
|
|
for (let i=0; i<zkey.domainSize; i++) {
|
|
const H = await readG1(fd, curve, toObject);
|
|
zkey.hExps.push(H);
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
await fd.close();
|
|
|
|
return zkey;
|
|
|
|
async function readFr2(/* toObject */) {
|
|
const n = await binFileUtils__namespace.readBigInt(fd, zkey.n8r);
|
|
return Fr.mul(n, Rri2);
|
|
}
|
|
|
|
}
|
|
|
|
|
|
async function readContribution$1(fd, curve, toObject) {
|
|
const c = {delta:{}};
|
|
c.deltaAfter = await readG1(fd, curve, toObject);
|
|
c.delta.g1_s = await readG1(fd, curve, toObject);
|
|
c.delta.g1_sx = await readG1(fd, curve, toObject);
|
|
c.delta.g2_spx = await readG2(fd, curve, toObject);
|
|
c.transcript = await fd.read(64);
|
|
c.type = await fd.readULE32();
|
|
|
|
const paramLength = await fd.readULE32();
|
|
const curPos = fd.pos;
|
|
let lastType =0;
|
|
while (fd.pos-curPos < paramLength) {
|
|
const buffType = await fd.read(1);
|
|
if (buffType[0]<= lastType) throw new Error("Parameters in the contribution must be sorted");
|
|
lastType = buffType[0];
|
|
if (buffType[0]==1) { // Name
|
|
const buffLen = await fd.read(1);
|
|
const buffStr = await fd.read(buffLen[0]);
|
|
c.name = new TextDecoder().decode(buffStr);
|
|
} else if (buffType[0]==2) {
|
|
const buffExp = await fd.read(1);
|
|
c.numIterationsExp = buffExp[0];
|
|
} else if (buffType[0]==3) {
|
|
const buffLen = await fd.read(1);
|
|
c.beaconHash = await fd.read(buffLen[0]);
|
|
} else {
|
|
throw new Error("Parameter not recognized");
|
|
}
|
|
}
|
|
if (fd.pos != curPos + paramLength) {
|
|
throw new Error("Parametes do not match");
|
|
}
|
|
|
|
return c;
|
|
}
|
|
|
|
|
|
async function readMPCParams(fd, curve, sections) {
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 10);
|
|
const res = { contributions: []};
|
|
res.csHash = await fd.read(64);
|
|
const n = await fd.readULE32();
|
|
for (let i=0; i<n; i++) {
|
|
const c = await readContribution$1(fd, curve);
|
|
res.contributions.push(c);
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
return res;
|
|
}
|
|
|
|
async function writeContribution$1(fd, curve, c) {
|
|
await writeG1(fd, curve, c.deltaAfter);
|
|
await writeG1(fd, curve, c.delta.g1_s);
|
|
await writeG1(fd, curve, c.delta.g1_sx);
|
|
await writeG2(fd, curve, c.delta.g2_spx);
|
|
await fd.write(c.transcript);
|
|
await fd.writeULE32(c.type || 0);
|
|
|
|
const params = [];
|
|
if (c.name) {
|
|
params.push(1); // Param Name
|
|
const nameData = new TextEncoder("utf-8").encode(c.name.substring(0,64));
|
|
params.push(nameData.byteLength);
|
|
for (let i=0; i<nameData.byteLength; i++) params.push(nameData[i]);
|
|
}
|
|
if (c.type == 1) {
|
|
params.push(2); // Param numIterationsExp
|
|
params.push(c.numIterationsExp);
|
|
|
|
params.push(3); // Beacon Hash
|
|
params.push(c.beaconHash.byteLength);
|
|
for (let i=0; i<c.beaconHash.byteLength; i++) params.push(c.beaconHash[i]);
|
|
}
|
|
if (params.length>0) {
|
|
const paramsBuff = new Uint8Array(params);
|
|
await fd.writeULE32(paramsBuff.byteLength);
|
|
await fd.write(paramsBuff);
|
|
} else {
|
|
await fd.writeULE32(0);
|
|
}
|
|
|
|
}
|
|
|
|
async function writeMPCParams(fd, curve, mpcParams) {
|
|
await binFileUtils__namespace.startWriteSection(fd, 10);
|
|
await fd.write(mpcParams.csHash);
|
|
await fd.writeULE32(mpcParams.contributions.length);
|
|
for (let i=0; i<mpcParams.contributions.length; i++) {
|
|
await writeContribution$1(fd, curve,mpcParams.contributions[i]);
|
|
}
|
|
await binFileUtils__namespace.endWriteSection(fd);
|
|
}
|
|
|
|
function hashG1(hasher, curve, p) {
|
|
const buff = new Uint8Array(curve.G1.F.n8*2);
|
|
curve.G1.toRprUncompressed(buff, 0, p);
|
|
hasher.update(buff);
|
|
}
|
|
|
|
function hashG2(hasher,curve, p) {
|
|
const buff = new Uint8Array(curve.G2.F.n8*2);
|
|
curve.G2.toRprUncompressed(buff, 0, p);
|
|
hasher.update(buff);
|
|
}
|
|
|
|
function hashPubKey(hasher, curve, c) {
|
|
hashG1(hasher, curve, c.deltaAfter);
|
|
hashG1(hasher, curve, c.delta.g1_s);
|
|
hashG1(hasher, curve, c.delta.g1_sx);
|
|
hashG2(hasher, curve, c.delta.g2_spx);
|
|
hasher.update(c.transcript);
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
|
|
async function write(fd, witness, prime) {
|
|
|
|
await binFileUtils__namespace.startWriteSection(fd, 1);
|
|
const n8 = (Math.floor( (ffjavascript.Scalar.bitLength(prime) - 1) / 64) +1)*8;
|
|
await fd.writeULE32(n8);
|
|
await binFileUtils__namespace.writeBigInt(fd, prime, n8);
|
|
await fd.writeULE32(witness.length);
|
|
await binFileUtils__namespace.endWriteSection(fd);
|
|
|
|
await binFileUtils__namespace.startWriteSection(fd, 2);
|
|
for (let i=0; i<witness.length; i++) {
|
|
await binFileUtils__namespace.writeBigInt(fd, witness[i], n8);
|
|
}
|
|
await binFileUtils__namespace.endWriteSection(fd, 2);
|
|
|
|
|
|
}
|
|
|
|
async function writeBin(fd, witnessBin, prime) {
|
|
|
|
await binFileUtils__namespace.startWriteSection(fd, 1);
|
|
const n8 = (Math.floor( (ffjavascript.Scalar.bitLength(prime) - 1) / 64) +1)*8;
|
|
await fd.writeULE32(n8);
|
|
await binFileUtils__namespace.writeBigInt(fd, prime, n8);
|
|
if (witnessBin.byteLength % n8 != 0) {
|
|
throw new Error("Invalid witness length");
|
|
}
|
|
await fd.writeULE32(witnessBin.byteLength / n8);
|
|
await binFileUtils__namespace.endWriteSection(fd);
|
|
|
|
|
|
await binFileUtils__namespace.startWriteSection(fd, 2);
|
|
await fd.write(witnessBin);
|
|
await binFileUtils__namespace.endWriteSection(fd);
|
|
|
|
}
|
|
|
|
async function readHeader(fd, sections) {
|
|
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 1);
|
|
const n8 = await fd.readULE32();
|
|
const q = await binFileUtils__namespace.readBigInt(fd, n8);
|
|
const nWitness = await fd.readULE32();
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
return {n8, q, nWitness};
|
|
|
|
}
|
|
|
|
async function read(fileName) {
|
|
|
|
const {fd, sections} = await binFileUtils__namespace.readBinFile(fileName, "wtns", 2);
|
|
|
|
const {n8, nWitness} = await readHeader(fd, sections);
|
|
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 2);
|
|
const res = [];
|
|
for (let i=0; i<nWitness; i++) {
|
|
const v = await binFileUtils__namespace.readBigInt(fd, n8);
|
|
res.push(v);
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
await fd.close();
|
|
|
|
return res;
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
const {stringifyBigInts: stringifyBigInts$3} = ffjavascript.utils;
|
|
|
|
async function groth16Prove(zkeyFileName, witnessFileName, logger) {
|
|
const {fd: fdWtns, sections: sectionsWtns} = await binFileUtils__namespace.readBinFile(witnessFileName, "wtns", 2, 1<<25, 1<<23);
|
|
|
|
const wtns = await readHeader(fdWtns, sectionsWtns);
|
|
|
|
const {fd: fdZKey, sections: sectionsZKey} = await binFileUtils__namespace.readBinFile(zkeyFileName, "zkey", 2, 1<<25, 1<<23);
|
|
|
|
const zkey = await readHeader$1(fdZKey, sectionsZKey);
|
|
|
|
if (zkey.protocol != "groth16") {
|
|
throw new Error("zkey file is not groth16");
|
|
}
|
|
|
|
if (!ffjavascript.Scalar.eq(zkey.r, wtns.q)) {
|
|
throw new Error("Curve of the witness does not match the curve of the proving key");
|
|
}
|
|
|
|
if (wtns.nWitness != zkey.nVars) {
|
|
throw new Error(`Invalid witness length. Circuit: ${zkey.nVars}, witness: ${wtns.nWitness}`);
|
|
}
|
|
|
|
const curve = await getCurveFromQ(zkey.q);
|
|
const Fr = curve.Fr;
|
|
const G1 = curve.G1;
|
|
const G2 = curve.G2;
|
|
|
|
const power = log2(zkey.domainSize);
|
|
|
|
if (logger) logger.debug("Reading Wtns");
|
|
const buffWitness = await binFileUtils__namespace.readSection(fdWtns, sectionsWtns, 2);
|
|
if (logger) logger.debug("Reading Coeffs");
|
|
const buffCoeffs = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 4);
|
|
|
|
if (logger) logger.debug("Building ABC");
|
|
const [buffA_T, buffB_T, buffC_T] = await buildABC1(curve, zkey, buffWitness, buffCoeffs, logger);
|
|
|
|
const inc = power == Fr.s ? curve.Fr.shift : curve.Fr.w[power+1];
|
|
|
|
const buffA = await Fr.ifft(buffA_T, "", "", logger, "IFFT_A");
|
|
const buffAodd = await Fr.batchApplyKey(buffA, Fr.e(1), inc);
|
|
const buffAodd_T = await Fr.fft(buffAodd, "", "", logger, "FFT_A");
|
|
|
|
const buffB = await Fr.ifft(buffB_T, "", "", logger, "IFFT_B");
|
|
const buffBodd = await Fr.batchApplyKey(buffB, Fr.e(1), inc);
|
|
const buffBodd_T = await Fr.fft(buffBodd, "", "", logger, "FFT_B");
|
|
|
|
const buffC = await Fr.ifft(buffC_T, "", "", logger, "IFFT_C");
|
|
const buffCodd = await Fr.batchApplyKey(buffC, Fr.e(1), inc);
|
|
const buffCodd_T = await Fr.fft(buffCodd, "", "", logger, "FFT_C");
|
|
|
|
if (logger) logger.debug("Join ABC");
|
|
const buffPodd_T = await joinABC(curve, zkey, buffAodd_T, buffBodd_T, buffCodd_T, logger);
|
|
|
|
let proof = {};
|
|
|
|
if (logger) logger.debug("Reading A Points");
|
|
const buffBasesA = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 5);
|
|
proof.pi_a = await curve.G1.multiExpAffine(buffBasesA, buffWitness, logger, "multiexp A");
|
|
|
|
if (logger) logger.debug("Reading B1 Points");
|
|
const buffBasesB1 = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 6);
|
|
let pib1 = await curve.G1.multiExpAffine(buffBasesB1, buffWitness, logger, "multiexp B1");
|
|
|
|
if (logger) logger.debug("Reading B2 Points");
|
|
const buffBasesB2 = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 7);
|
|
proof.pi_b = await curve.G2.multiExpAffine(buffBasesB2, buffWitness, logger, "multiexp B2");
|
|
|
|
if (logger) logger.debug("Reading C Points");
|
|
const buffBasesC = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 8);
|
|
proof.pi_c = await curve.G1.multiExpAffine(buffBasesC, buffWitness.slice((zkey.nPublic+1)*curve.Fr.n8), logger, "multiexp C");
|
|
|
|
if (logger) logger.debug("Reading H Points");
|
|
const buffBasesH = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 9);
|
|
const resH = await curve.G1.multiExpAffine(buffBasesH, buffPodd_T, logger, "multiexp H");
|
|
|
|
const r = curve.Fr.random();
|
|
const s = curve.Fr.random();
|
|
|
|
proof.pi_a = G1.add( proof.pi_a, zkey.vk_alpha_1 );
|
|
proof.pi_a = G1.add( proof.pi_a, G1.timesFr( zkey.vk_delta_1, r ));
|
|
|
|
proof.pi_b = G2.add( proof.pi_b, zkey.vk_beta_2 );
|
|
proof.pi_b = G2.add( proof.pi_b, G2.timesFr( zkey.vk_delta_2, s ));
|
|
|
|
pib1 = G1.add( pib1, zkey.vk_beta_1 );
|
|
pib1 = G1.add( pib1, G1.timesFr( zkey.vk_delta_1, s ));
|
|
|
|
proof.pi_c = G1.add(proof.pi_c, resH);
|
|
|
|
|
|
proof.pi_c = G1.add( proof.pi_c, G1.timesFr( proof.pi_a, s ));
|
|
proof.pi_c = G1.add( proof.pi_c, G1.timesFr( pib1, r ));
|
|
proof.pi_c = G1.add( proof.pi_c, G1.timesFr( zkey.vk_delta_1, Fr.neg(Fr.mul(r,s) )));
|
|
|
|
|
|
let publicSignals = [];
|
|
|
|
for (let i=1; i<= zkey.nPublic; i++) {
|
|
const b = buffWitness.slice(i*Fr.n8, i*Fr.n8+Fr.n8);
|
|
publicSignals.push(ffjavascript.Scalar.fromRprLE(b));
|
|
}
|
|
|
|
proof.pi_a = G1.toObject(G1.toAffine(proof.pi_a));
|
|
proof.pi_b = G2.toObject(G2.toAffine(proof.pi_b));
|
|
proof.pi_c = G1.toObject(G1.toAffine(proof.pi_c));
|
|
|
|
proof.protocol = "groth16";
|
|
proof.curve = curve.name;
|
|
|
|
await fdZKey.close();
|
|
await fdWtns.close();
|
|
|
|
proof = stringifyBigInts$3(proof);
|
|
publicSignals = stringifyBigInts$3(publicSignals);
|
|
|
|
return {proof, publicSignals};
|
|
}
|
|
|
|
|
|
async function buildABC1(curve, zkey, witness, coeffs, logger) {
|
|
const n8 = curve.Fr.n8;
|
|
const sCoef = 4*3 + zkey.n8r;
|
|
const nCoef = (coeffs.byteLength-4) / sCoef;
|
|
|
|
const outBuffA = new ffjavascript.BigBuffer(zkey.domainSize * n8);
|
|
const outBuffB = new ffjavascript.BigBuffer(zkey.domainSize * n8);
|
|
const outBuffC = new ffjavascript.BigBuffer(zkey.domainSize * n8);
|
|
|
|
const outBuf = [ outBuffA, outBuffB ];
|
|
for (let i=0; i<nCoef; i++) {
|
|
if ((logger)&&(i%1000000 == 0)) logger.debug(`QAP AB: ${i}/${nCoef}`);
|
|
const buffCoef = coeffs.slice(4+i*sCoef, 4+i*sCoef+sCoef);
|
|
const buffCoefV = new DataView(buffCoef.buffer);
|
|
const m= buffCoefV.getUint32(0, true);
|
|
const c= buffCoefV.getUint32(4, true);
|
|
const s= buffCoefV.getUint32(8, true);
|
|
const coef = buffCoef.slice(12, 12+n8);
|
|
outBuf[m].set(
|
|
curve.Fr.add(
|
|
outBuf[m].slice(c*n8, c*n8+n8),
|
|
curve.Fr.mul(coef, witness.slice(s*n8, s*n8+n8))
|
|
),
|
|
c*n8
|
|
);
|
|
}
|
|
|
|
for (let i=0; i<zkey.domainSize; i++) {
|
|
if ((logger)&&(i%1000000 == 0)) logger.debug(`QAP C: ${i}/${zkey.domainSize}`);
|
|
outBuffC.set(
|
|
curve.Fr.mul(
|
|
outBuffA.slice(i*n8, i*n8+n8),
|
|
outBuffB.slice(i*n8, i*n8+n8),
|
|
),
|
|
i*n8
|
|
);
|
|
}
|
|
|
|
return [outBuffA, outBuffB, outBuffC];
|
|
|
|
}
|
|
|
|
/*
|
|
async function buldABC(curve, zkey, witness, coeffs, logger) {
|
|
const concurrency = curve.tm.concurrency;
|
|
const sCoef = 4*3 + zkey.n8r;
|
|
|
|
let getUint32;
|
|
|
|
if (coeffs instanceof BigBuffer) {
|
|
const coeffsDV = [];
|
|
const PAGE_LEN = coeffs.buffers[0].length;
|
|
for (let i=0; i< coeffs.buffers.length; i++) {
|
|
coeffsDV.push(new DataView(coeffs.buffers[i].buffer));
|
|
}
|
|
getUint32 = function (pos) {
|
|
return coeffsDV[Math.floor(pos/PAGE_LEN)].getUint32(pos % PAGE_LEN, true);
|
|
};
|
|
} else {
|
|
const coeffsDV = new DataView(coeffs.buffer, coeffs.byteOffset, coeffs.byteLength);
|
|
getUint32 = function (pos) {
|
|
return coeffsDV.getUint32(pos, true);
|
|
};
|
|
}
|
|
|
|
const elementsPerChunk = Math.floor(zkey.domainSize/concurrency);
|
|
const promises = [];
|
|
|
|
const cutPoints = [];
|
|
for (let i=0; i<concurrency; i++) {
|
|
cutPoints.push( getCutPoint( Math.floor(i*elementsPerChunk) ));
|
|
}
|
|
cutPoints.push(coeffs.byteLength);
|
|
|
|
const chunkSize = 2**26;
|
|
for (let s=0 ; s<zkey.nVars ; s+= chunkSize) {
|
|
if (logger) logger.debug(`QAP ${s}: ${s}/${zkey.nVars}`);
|
|
const ns= Math.min(zkey.nVars-s, chunkSize );
|
|
|
|
for (let i=0; i<concurrency; i++) {
|
|
let n;
|
|
if (i< concurrency-1) {
|
|
n = elementsPerChunk;
|
|
} else {
|
|
n = zkey.domainSize - i*elementsPerChunk;
|
|
}
|
|
if (n==0) continue;
|
|
|
|
const task = [];
|
|
|
|
task.push({cmd: "ALLOCSET", var: 0, buff: coeffs.slice(cutPoints[i], cutPoints[i+1])});
|
|
task.push({cmd: "ALLOCSET", var: 1, buff: witness.slice(s*curve.Fr.n8, (s+ns)*curve.Fr.n8)});
|
|
task.push({cmd: "ALLOC", var: 2, len: n*curve.Fr.n8});
|
|
task.push({cmd: "ALLOC", var: 3, len: n*curve.Fr.n8});
|
|
task.push({cmd: "ALLOC", var: 4, len: n*curve.Fr.n8});
|
|
task.push({cmd: "CALL", fnName: "qap_buildABC", params:[
|
|
{var: 0},
|
|
{val: (cutPoints[i+1] - cutPoints[i])/sCoef},
|
|
{var: 1},
|
|
{var: 2},
|
|
{var: 3},
|
|
{var: 4},
|
|
{val: i*elementsPerChunk},
|
|
{val: n},
|
|
{val: s},
|
|
{val: ns}
|
|
]});
|
|
task.push({cmd: "GET", out: 0, var: 2, len: n*curve.Fr.n8});
|
|
task.push({cmd: "GET", out: 1, var: 3, len: n*curve.Fr.n8});
|
|
task.push({cmd: "GET", out: 2, var: 4, len: n*curve.Fr.n8});
|
|
promises.push(curve.tm.queueAction(task));
|
|
}
|
|
}
|
|
|
|
let result = await Promise.all(promises);
|
|
|
|
const nGroups = result.length / concurrency;
|
|
if (nGroups>1) {
|
|
const promises2 = [];
|
|
for (let i=0; i<concurrency; i++) {
|
|
const task=[];
|
|
task.push({cmd: "ALLOC", var: 0, len: result[i][0].byteLength});
|
|
task.push({cmd: "ALLOC", var: 1, len: result[i][0].byteLength});
|
|
for (let m=0; m<3; m++) {
|
|
task.push({cmd: "SET", var: 0, buff: result[i][m]});
|
|
for (let s=1; s<nGroups; s++) {
|
|
task.push({cmd: "SET", var: 1, buff: result[s*concurrency + i][m]});
|
|
task.push({cmd: "CALL", fnName: "qap_batchAdd", params:[
|
|
{var: 0},
|
|
{var: 1},
|
|
{val: result[i][m].length/curve.Fr.n8},
|
|
{var: 0}
|
|
]});
|
|
}
|
|
task.push({cmd: "GET", out: m, var: 0, len: result[i][m].length});
|
|
}
|
|
promises2.push(curve.tm.queueAction(task));
|
|
}
|
|
result = await Promise.all(promises2);
|
|
}
|
|
|
|
const outBuffA = new BigBuffer(zkey.domainSize * curve.Fr.n8);
|
|
const outBuffB = new BigBuffer(zkey.domainSize * curve.Fr.n8);
|
|
const outBuffC = new BigBuffer(zkey.domainSize * curve.Fr.n8);
|
|
let p=0;
|
|
for (let i=0; i<result.length; i++) {
|
|
outBuffA.set(result[i][0], p);
|
|
outBuffB.set(result[i][1], p);
|
|
outBuffC.set(result[i][2], p);
|
|
p += result[i][0].byteLength;
|
|
}
|
|
|
|
return [outBuffA, outBuffB, outBuffC];
|
|
|
|
function getCutPoint(v) {
|
|
let m = 0;
|
|
let n = getUint32(0);
|
|
while (m < n) {
|
|
var k = Math.floor((n + m) / 2);
|
|
const va = getUint32(4 + k*sCoef + 4);
|
|
if (va > v) {
|
|
n = k - 1;
|
|
} else if (va < v) {
|
|
m = k + 1;
|
|
} else {
|
|
n = k;
|
|
}
|
|
}
|
|
return 4 + m*sCoef;
|
|
}
|
|
}
|
|
*/
|
|
|
|
async function joinABC(curve, zkey, a, b, c, logger) {
|
|
const MAX_CHUNK_SIZE = 1 << 22;
|
|
|
|
const n8 = curve.Fr.n8;
|
|
const nElements = Math.floor(a.byteLength / curve.Fr.n8);
|
|
|
|
const promises = [];
|
|
|
|
for (let i=0; i<nElements; i += MAX_CHUNK_SIZE) {
|
|
if (logger) logger.debug(`JoinABC: ${i}/${nElements}`);
|
|
const n= Math.min(nElements - i, MAX_CHUNK_SIZE);
|
|
|
|
const task = [];
|
|
|
|
const aChunk = a.slice(i*n8, (i + n)*n8 );
|
|
const bChunk = b.slice(i*n8, (i + n)*n8 );
|
|
const cChunk = c.slice(i*n8, (i + n)*n8 );
|
|
|
|
task.push({cmd: "ALLOCSET", var: 0, buff: aChunk});
|
|
task.push({cmd: "ALLOCSET", var: 1, buff: bChunk});
|
|
task.push({cmd: "ALLOCSET", var: 2, buff: cChunk});
|
|
task.push({cmd: "ALLOC", var: 3, len: n*n8});
|
|
task.push({cmd: "CALL", fnName: "qap_joinABC", params:[
|
|
{var: 0},
|
|
{var: 1},
|
|
{var: 2},
|
|
{val: n},
|
|
{var: 3},
|
|
]});
|
|
task.push({cmd: "CALL", fnName: "frm_batchFromMontgomery", params:[
|
|
{var: 3},
|
|
{val: n},
|
|
{var: 3}
|
|
]});
|
|
task.push({cmd: "GET", out: 0, var: 3, len: n*n8});
|
|
promises.push(curve.tm.queueAction(task));
|
|
}
|
|
|
|
const result = await Promise.all(promises);
|
|
|
|
let outBuff;
|
|
if (a instanceof ffjavascript.BigBuffer) {
|
|
outBuff = new ffjavascript.BigBuffer(a.byteLength);
|
|
} else {
|
|
outBuff = new Uint8Array(a.byteLength);
|
|
}
|
|
|
|
let p=0;
|
|
for (let i=0; i<result.length; i++) {
|
|
outBuff.set(result[i][0], p);
|
|
p += result[i][0].byteLength;
|
|
}
|
|
|
|
return outBuff;
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function wtnsCalculate(input, wasmFileName, wtnsFileName, options) {
|
|
|
|
const fdWasm = await fastFile__namespace.readExisting(wasmFileName);
|
|
const wasm = await fdWasm.read(fdWasm.totalSize);
|
|
await fdWasm.close();
|
|
|
|
const wc = await circom_runtime.WitnessCalculatorBuilder(wasm);
|
|
if (wc.circom_version() == 1) {
|
|
const w = await wc.calculateBinWitness(input);
|
|
|
|
const fdWtns = await binFileUtils__namespace.createBinFile(wtnsFileName, "wtns", 2, 2);
|
|
|
|
await writeBin(fdWtns, w, wc.prime);
|
|
await fdWtns.close();
|
|
} else {
|
|
const fdWtns = await fastFile__namespace.createOverride(wtnsFileName);
|
|
|
|
const w = await wc.calculateWTNSBin(input);
|
|
|
|
await fdWtns.write(w);
|
|
await fdWtns.close();
|
|
}
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function groth16FullProve(input, wasmFile, zkeyFileName, logger) {
|
|
const wtns= {
|
|
type: "mem"
|
|
};
|
|
await wtnsCalculate(input, wasmFile, wtns);
|
|
return await groth16Prove(zkeyFileName, wtns, logger);
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0kims association.
|
|
|
|
This file is part of snarkjs.
|
|
|
|
snarkjs is a free software: you can redistribute it and/or
|
|
modify it under the terms of the GNU General Public License as published by the
|
|
Free Software Foundation, either version 3 of the License, or (at your option)
|
|
any later version.
|
|
|
|
snarkjs is distributed in the hope that it will be useful,
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
|
|
more details.
|
|
|
|
You should have received a copy of the GNU General Public License along with
|
|
snarkjs. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
const {unstringifyBigInts: unstringifyBigInts$1} = ffjavascript.utils;
|
|
|
|
async function groth16Verify(vk_verifier, publicSignals, proof, logger) {
|
|
/*
|
|
let cpub = vk_verifier.IC[0];
|
|
for (let s= 0; s< vk_verifier.nPublic; s++) {
|
|
cpub = G1.add( cpub, G1.timesScalar( vk_verifier.IC[s+1], publicSignals[s]));
|
|
}
|
|
*/
|
|
|
|
vk_verifier = unstringifyBigInts$1(vk_verifier);
|
|
proof = unstringifyBigInts$1(proof);
|
|
publicSignals = unstringifyBigInts$1(publicSignals);
|
|
|
|
const curve = await getCurveFromName(vk_verifier.curve);
|
|
|
|
const IC0 = curve.G1.fromObject(vk_verifier.IC[0]);
|
|
const IC = new Uint8Array(curve.G1.F.n8*2 * publicSignals.length);
|
|
const w = new Uint8Array(curve.Fr.n8 * publicSignals.length);
|
|
|
|
for (let i=0; i<publicSignals.length; i++) {
|
|
const buffP = curve.G1.fromObject(vk_verifier.IC[i+1]);
|
|
IC.set(buffP, i*curve.G1.F.n8*2);
|
|
ffjavascript.Scalar.toRprLE(w, curve.Fr.n8*i, publicSignals[i], curve.Fr.n8);
|
|
}
|
|
|
|
let cpub = await curve.G1.multiExpAffine(IC, w);
|
|
cpub = curve.G1.add(cpub, IC0);
|
|
|
|
const pi_a = curve.G1.fromObject(proof.pi_a);
|
|
const pi_b = curve.G2.fromObject(proof.pi_b);
|
|
const pi_c = curve.G1.fromObject(proof.pi_c);
|
|
|
|
const vk_gamma_2 = curve.G2.fromObject(vk_verifier.vk_gamma_2);
|
|
const vk_delta_2 = curve.G2.fromObject(vk_verifier.vk_delta_2);
|
|
const vk_alpha_1 = curve.G1.fromObject(vk_verifier.vk_alpha_1);
|
|
const vk_beta_2 = curve.G2.fromObject(vk_verifier.vk_beta_2);
|
|
|
|
const res = await curve.pairingEq(
|
|
curve.G1.neg(pi_a) , pi_b,
|
|
cpub , vk_gamma_2,
|
|
pi_c , vk_delta_2,
|
|
|
|
vk_alpha_1, vk_beta_2
|
|
);
|
|
|
|
if (! res) {
|
|
if (logger) logger.error("Invalid proof");
|
|
return false;
|
|
}
|
|
|
|
if (logger) logger.info("OK!");
|
|
return true;
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
function p256$1(n) {
|
|
let nstr = n.toString(16);
|
|
while (nstr.length < 64) nstr = "0"+nstr;
|
|
nstr = `"0x${nstr}"`;
|
|
return nstr;
|
|
}
|
|
|
|
async function groth16ExportSolidityCallData(proof, pub) {
|
|
|
|
let inputs = "";
|
|
for (let i=0; i<pub.length; i++) {
|
|
if (inputs != "") inputs = inputs + ",";
|
|
inputs = inputs + p256$1(pub[i]);
|
|
}
|
|
|
|
let S;
|
|
S=`[${p256$1(proof.pi_a[0])}, ${p256$1(proof.pi_a[1])}],` +
|
|
`[[${p256$1(proof.pi_b[0][1])}, ${p256$1(proof.pi_b[0][0])}],[${p256$1(proof.pi_b[1][1])}, ${p256$1(proof.pi_b[1][0])}]],` +
|
|
`[${p256$1(proof.pi_c[0])}, ${p256$1(proof.pi_c[1])}],` +
|
|
`[${inputs}]`;
|
|
|
|
return S;
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
var groth16 = /*#__PURE__*/Object.freeze({
|
|
__proto__: null,
|
|
fullProve: groth16FullProve,
|
|
prove: groth16Prove,
|
|
verify: groth16Verify,
|
|
exportSolidityCallData: groth16ExportSolidityCallData
|
|
});
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
function hashToG2(curve, hash) {
|
|
const hashV = new DataView(hash.buffer, hash.byteOffset, hash.byteLength);
|
|
const seed = [];
|
|
for (let i=0; i<8; i++) {
|
|
seed[i] = hashV.getUint32(i*4);
|
|
}
|
|
|
|
const rng = new ffjavascript.ChaCha(seed);
|
|
|
|
const g2_sp = curve.G2.fromRng(rng);
|
|
|
|
return g2_sp;
|
|
}
|
|
|
|
function getG2sp(curve, persinalization, challenge, g1s, g1sx) {
|
|
|
|
const h = Blake2b__default["default"](64);
|
|
const b1 = new Uint8Array([persinalization]);
|
|
h.update(b1);
|
|
h.update(challenge);
|
|
const b3 = curve.G1.toUncompressed(g1s);
|
|
h.update( b3);
|
|
const b4 = curve.G1.toUncompressed(g1sx);
|
|
h.update( b4);
|
|
const hash =h.digest();
|
|
|
|
return hashToG2(curve, hash);
|
|
}
|
|
|
|
function calculatePubKey(k, curve, personalization, challengeHash, rng ) {
|
|
k.g1_s = curve.G1.toAffine(curve.G1.fromRng(rng));
|
|
k.g1_sx = curve.G1.toAffine(curve.G1.timesFr(k.g1_s, k.prvKey));
|
|
k.g2_sp = curve.G2.toAffine(getG2sp(curve, personalization, challengeHash, k.g1_s, k.g1_sx));
|
|
k.g2_spx = curve.G2.toAffine(curve.G2.timesFr(k.g2_sp, k.prvKey));
|
|
return k;
|
|
}
|
|
|
|
function createPTauKey(curve, challengeHash, rng) {
|
|
const key = {
|
|
tau: {},
|
|
alpha: {},
|
|
beta: {}
|
|
};
|
|
key.tau.prvKey = curve.Fr.fromRng(rng);
|
|
key.alpha.prvKey = curve.Fr.fromRng(rng);
|
|
key.beta.prvKey = curve.Fr.fromRng(rng);
|
|
calculatePubKey(key.tau, curve, 0, challengeHash, rng);
|
|
calculatePubKey(key.alpha, curve, 1, challengeHash, rng);
|
|
calculatePubKey(key.beta, curve, 2, challengeHash, rng);
|
|
return key;
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function writePTauHeader(fd, curve, power, ceremonyPower) {
|
|
// Write the header
|
|
///////////
|
|
|
|
if (! ceremonyPower) ceremonyPower = power;
|
|
await fd.writeULE32(1); // Header type
|
|
const pHeaderSize = fd.pos;
|
|
await fd.writeULE64(0); // Temporally set to 0 length
|
|
|
|
await fd.writeULE32(curve.F1.n64*8);
|
|
|
|
const buff = new Uint8Array(curve.F1.n8);
|
|
ffjavascript.Scalar.toRprLE(buff, 0, curve.q, curve.F1.n8);
|
|
await fd.write(buff);
|
|
await fd.writeULE32(power); // power
|
|
await fd.writeULE32(ceremonyPower); // power
|
|
|
|
const headerSize = fd.pos - pHeaderSize - 8;
|
|
|
|
const oldPos = fd.pos;
|
|
|
|
await fd.writeULE64(headerSize, pHeaderSize);
|
|
|
|
fd.pos = oldPos;
|
|
}
|
|
|
|
async function readPTauHeader(fd, sections) {
|
|
if (!sections[1]) throw new Error(fd.fileName + ": File has no header");
|
|
if (sections[1].length>1) throw new Error(fd.fileName +": File has more than one header");
|
|
|
|
fd.pos = sections[1][0].p;
|
|
const n8 = await fd.readULE32();
|
|
const buff = await fd.read(n8);
|
|
const q = ffjavascript.Scalar.fromRprLE(buff);
|
|
|
|
const curve = await getCurveFromQ(q);
|
|
|
|
if (curve.F1.n64*8 != n8) throw new Error(fd.fileName +": Invalid size");
|
|
|
|
const power = await fd.readULE32();
|
|
const ceremonyPower = await fd.readULE32();
|
|
|
|
if (fd.pos-sections[1][0].p != sections[1][0].size) throw new Error("Invalid PTau header size");
|
|
|
|
return {curve, power, ceremonyPower};
|
|
}
|
|
|
|
|
|
async function readPtauPubKey(fd, curve, montgomery) {
|
|
|
|
const buff = await fd.read(curve.F1.n8*2*6 + curve.F2.n8*2*3);
|
|
|
|
return fromPtauPubKeyRpr(buff, 0, curve, montgomery);
|
|
}
|
|
|
|
function fromPtauPubKeyRpr(buff, pos, curve, montgomery) {
|
|
|
|
const key = {
|
|
tau: {},
|
|
alpha: {},
|
|
beta: {}
|
|
};
|
|
|
|
key.tau.g1_s = readG1();
|
|
key.tau.g1_sx = readG1();
|
|
key.alpha.g1_s = readG1();
|
|
key.alpha.g1_sx = readG1();
|
|
key.beta.g1_s = readG1();
|
|
key.beta.g1_sx = readG1();
|
|
key.tau.g2_spx = readG2();
|
|
key.alpha.g2_spx = readG2();
|
|
key.beta.g2_spx = readG2();
|
|
|
|
return key;
|
|
|
|
function readG1() {
|
|
let p;
|
|
if (montgomery) {
|
|
p = curve.G1.fromRprLEM( buff, pos );
|
|
} else {
|
|
p = curve.G1.fromRprUncompressed( buff, pos );
|
|
}
|
|
pos += curve.G1.F.n8*2;
|
|
return p;
|
|
}
|
|
|
|
function readG2() {
|
|
let p;
|
|
if (montgomery) {
|
|
p = curve.G2.fromRprLEM( buff, pos );
|
|
} else {
|
|
p = curve.G2.fromRprUncompressed( buff, pos );
|
|
}
|
|
pos += curve.G2.F.n8*2;
|
|
return p;
|
|
}
|
|
}
|
|
|
|
function toPtauPubKeyRpr(buff, pos, curve, key, montgomery) {
|
|
|
|
writeG1(key.tau.g1_s);
|
|
writeG1(key.tau.g1_sx);
|
|
writeG1(key.alpha.g1_s);
|
|
writeG1(key.alpha.g1_sx);
|
|
writeG1(key.beta.g1_s);
|
|
writeG1(key.beta.g1_sx);
|
|
writeG2(key.tau.g2_spx);
|
|
writeG2(key.alpha.g2_spx);
|
|
writeG2(key.beta.g2_spx);
|
|
|
|
async function writeG1(p) {
|
|
if (montgomery) {
|
|
curve.G1.toRprLEM(buff, pos, p);
|
|
} else {
|
|
curve.G1.toRprUncompressed(buff, pos, p);
|
|
}
|
|
pos += curve.F1.n8*2;
|
|
}
|
|
|
|
async function writeG2(p) {
|
|
if (montgomery) {
|
|
curve.G2.toRprLEM(buff, pos, p);
|
|
} else {
|
|
curve.G2.toRprUncompressed(buff, pos, p);
|
|
}
|
|
pos += curve.F2.n8*2;
|
|
}
|
|
|
|
return buff;
|
|
}
|
|
|
|
async function writePtauPubKey(fd, curve, key, montgomery) {
|
|
const buff = new Uint8Array(curve.F1.n8*2*6 + curve.F2.n8*2*3);
|
|
toPtauPubKeyRpr(buff, 0, curve, key, montgomery);
|
|
await fd.write(buff);
|
|
}
|
|
|
|
async function readContribution(fd, curve) {
|
|
const c = {};
|
|
|
|
c.tauG1 = await readG1();
|
|
c.tauG2 = await readG2();
|
|
c.alphaG1 = await readG1();
|
|
c.betaG1 = await readG1();
|
|
c.betaG2 = await readG2();
|
|
c.key = await readPtauPubKey(fd, curve, true);
|
|
c.partialHash = await fd.read(216);
|
|
c.nextChallenge = await fd.read(64);
|
|
c.type = await fd.readULE32();
|
|
|
|
const buffV = new Uint8Array(curve.G1.F.n8*2*6+curve.G2.F.n8*2*3);
|
|
toPtauPubKeyRpr(buffV, 0, curve, c.key, false);
|
|
|
|
const responseHasher = Blake2b__default["default"](64);
|
|
responseHasher.setPartialHash(c.partialHash);
|
|
responseHasher.update(buffV);
|
|
c.responseHash = responseHasher.digest();
|
|
|
|
const paramLength = await fd.readULE32();
|
|
const curPos = fd.pos;
|
|
let lastType =0;
|
|
while (fd.pos-curPos < paramLength) {
|
|
const buffType = await readDV(1);
|
|
if (buffType[0]<= lastType) throw new Error("Parameters in the contribution must be sorted");
|
|
lastType = buffType[0];
|
|
if (buffType[0]==1) { // Name
|
|
const buffLen = await readDV(1);
|
|
const buffStr = await readDV(buffLen[0]);
|
|
c.name = new TextDecoder().decode(buffStr);
|
|
} else if (buffType[0]==2) {
|
|
const buffExp = await readDV(1);
|
|
c.numIterationsExp = buffExp[0];
|
|
} else if (buffType[0]==3) {
|
|
const buffLen = await readDV(1);
|
|
c.beaconHash = await readDV(buffLen[0]);
|
|
} else {
|
|
throw new Error("Parameter not recognized");
|
|
}
|
|
}
|
|
if (fd.pos != curPos + paramLength) {
|
|
throw new Error("Parametes do not match");
|
|
}
|
|
|
|
return c;
|
|
|
|
async function readG1() {
|
|
const pBuff = await fd.read(curve.G1.F.n8*2);
|
|
return curve.G1.fromRprLEM( pBuff );
|
|
}
|
|
|
|
async function readG2() {
|
|
const pBuff = await fd.read(curve.G2.F.n8*2);
|
|
return curve.G2.fromRprLEM( pBuff );
|
|
}
|
|
|
|
async function readDV(n) {
|
|
const b = await fd.read(n);
|
|
return new Uint8Array(b);
|
|
}
|
|
}
|
|
|
|
async function readContributions(fd, curve, sections) {
|
|
if (!sections[7]) throw new Error(fd.fileName + ": File has no contributions");
|
|
if (sections[7][0].length>1) throw new Error(fd.fileName +": File has more than one contributions section");
|
|
|
|
fd.pos = sections[7][0].p;
|
|
const nContributions = await fd.readULE32();
|
|
const contributions = [];
|
|
for (let i=0; i<nContributions; i++) {
|
|
const c = await readContribution(fd, curve);
|
|
c.id = i+1;
|
|
contributions.push(c);
|
|
}
|
|
|
|
if (fd.pos-sections[7][0].p != sections[7][0].size) throw new Error("Invalid contribution section size");
|
|
|
|
return contributions;
|
|
}
|
|
|
|
async function writeContribution(fd, curve, contribution) {
|
|
|
|
const buffG1 = new Uint8Array(curve.F1.n8*2);
|
|
const buffG2 = new Uint8Array(curve.F2.n8*2);
|
|
await writeG1(contribution.tauG1);
|
|
await writeG2(contribution.tauG2);
|
|
await writeG1(contribution.alphaG1);
|
|
await writeG1(contribution.betaG1);
|
|
await writeG2(contribution.betaG2);
|
|
await writePtauPubKey(fd, curve, contribution.key, true);
|
|
await fd.write(contribution.partialHash);
|
|
await fd.write(contribution.nextChallenge);
|
|
await fd.writeULE32(contribution.type || 0);
|
|
|
|
const params = [];
|
|
if (contribution.name) {
|
|
params.push(1); // Param Name
|
|
const nameData = new TextEncoder("utf-8").encode(contribution.name.substring(0,64));
|
|
params.push(nameData.byteLength);
|
|
for (let i=0; i<nameData.byteLength; i++) params.push(nameData[i]);
|
|
}
|
|
if (contribution.type == 1) {
|
|
params.push(2); // Param numIterationsExp
|
|
params.push(contribution.numIterationsExp);
|
|
|
|
params.push(3); // Beacon Hash
|
|
params.push(contribution.beaconHash.byteLength);
|
|
for (let i=0; i<contribution.beaconHash.byteLength; i++) params.push(contribution.beaconHash[i]);
|
|
}
|
|
if (params.length>0) {
|
|
const paramsBuff = new Uint8Array(params);
|
|
await fd.writeULE32(paramsBuff.byteLength);
|
|
await fd.write(paramsBuff);
|
|
} else {
|
|
await fd.writeULE32(0);
|
|
}
|
|
|
|
|
|
async function writeG1(p) {
|
|
curve.G1.toRprLEM(buffG1, 0, p);
|
|
await fd.write(buffG1);
|
|
}
|
|
|
|
async function writeG2(p) {
|
|
curve.G2.toRprLEM(buffG2, 0, p);
|
|
await fd.write(buffG2);
|
|
}
|
|
|
|
}
|
|
|
|
async function writeContributions(fd, curve, contributions) {
|
|
|
|
await fd.writeULE32(7); // Header type
|
|
const pContributionsSize = fd.pos;
|
|
await fd.writeULE64(0); // Temporally set to 0 length
|
|
|
|
await fd.writeULE32(contributions.length);
|
|
for (let i=0; i< contributions.length; i++) {
|
|
await writeContribution(fd, curve, contributions[i]);
|
|
}
|
|
const contributionsSize = fd.pos - pContributionsSize - 8;
|
|
|
|
const oldPos = fd.pos;
|
|
|
|
await fd.writeULE64(contributionsSize, pContributionsSize);
|
|
fd.pos = oldPos;
|
|
}
|
|
|
|
function calculateFirstChallengeHash(curve, power, logger) {
|
|
if (logger) logger.debug("Calculating First Challenge Hash");
|
|
|
|
const hasher = new Blake2b__default["default"](64);
|
|
|
|
const vG1 = new Uint8Array(curve.G1.F.n8*2);
|
|
const vG2 = new Uint8Array(curve.G2.F.n8*2);
|
|
curve.G1.toRprUncompressed(vG1, 0, curve.G1.g);
|
|
curve.G2.toRprUncompressed(vG2, 0, curve.G2.g);
|
|
|
|
hasher.update(Blake2b__default["default"](64).digest());
|
|
|
|
let n;
|
|
|
|
n=(2 ** power)*2 -1;
|
|
if (logger) logger.debug("Calculate Initial Hash: tauG1");
|
|
hashBlock(vG1, n);
|
|
n= 2 ** power;
|
|
if (logger) logger.debug("Calculate Initial Hash: tauG2");
|
|
hashBlock(vG2, n);
|
|
if (logger) logger.debug("Calculate Initial Hash: alphaTauG1");
|
|
hashBlock(vG1, n);
|
|
if (logger) logger.debug("Calculate Initial Hash: betaTauG1");
|
|
hashBlock(vG1, n);
|
|
hasher.update(vG2);
|
|
|
|
return hasher.digest();
|
|
|
|
function hashBlock(buff, n) {
|
|
const blockSize = 500000;
|
|
const nBlocks = Math.floor(n / blockSize);
|
|
const rem = n % blockSize;
|
|
const bigBuff = new Uint8Array(blockSize * buff.byteLength);
|
|
for (let i=0; i<blockSize; i++) {
|
|
bigBuff.set(buff, i*buff.byteLength);
|
|
}
|
|
for (let i=0; i<nBlocks; i++) {
|
|
hasher.update(bigBuff);
|
|
if (logger) logger.debug("Initial hash: " +i*blockSize);
|
|
}
|
|
for (let i=0; i<rem; i++) {
|
|
hasher.update(buff);
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
function keyFromBeacon(curve, challengeHash, beaconHash, numIterationsExp) {
|
|
|
|
const rng = rngFromBeaconParams(beaconHash, numIterationsExp);
|
|
|
|
const key = createPTauKey(curve, challengeHash, rng);
|
|
|
|
return key;
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function newAccumulator(curve, power, fileName, logger) {
|
|
|
|
await Blake2b__default["default"].ready();
|
|
|
|
const fd = await binFileUtils__namespace.createBinFile(fileName, "ptau", 1, 7);
|
|
|
|
await writePTauHeader(fd, curve, power, 0);
|
|
|
|
const buffG1 = curve.G1.oneAffine;
|
|
const buffG2 = curve.G2.oneAffine;
|
|
|
|
// Write tauG1
|
|
///////////
|
|
await binFileUtils__namespace.startWriteSection(fd, 2);
|
|
const nTauG1 = (2 ** power) * 2 -1;
|
|
for (let i=0; i< nTauG1; i++) {
|
|
await fd.write(buffG1);
|
|
if ((logger)&&((i%100000) == 0)&&i) logger.log("tauG1: " + i);
|
|
}
|
|
await binFileUtils__namespace.endWriteSection(fd);
|
|
|
|
// Write tauG2
|
|
///////////
|
|
await binFileUtils__namespace.startWriteSection(fd, 3);
|
|
const nTauG2 = (2 ** power);
|
|
for (let i=0; i< nTauG2; i++) {
|
|
await fd.write(buffG2);
|
|
if ((logger)&&((i%100000) == 0)&&i) logger.log("tauG2: " + i);
|
|
}
|
|
await binFileUtils__namespace.endWriteSection(fd);
|
|
|
|
// Write alphaTauG1
|
|
///////////
|
|
await binFileUtils__namespace.startWriteSection(fd, 4);
|
|
const nAlfaTauG1 = (2 ** power);
|
|
for (let i=0; i< nAlfaTauG1; i++) {
|
|
await fd.write(buffG1);
|
|
if ((logger)&&((i%100000) == 0)&&i) logger.log("alphaTauG1: " + i);
|
|
}
|
|
await binFileUtils__namespace.endWriteSection(fd);
|
|
|
|
// Write betaTauG1
|
|
///////////
|
|
await binFileUtils__namespace.startWriteSection(fd, 5);
|
|
const nBetaTauG1 = (2 ** power);
|
|
for (let i=0; i< nBetaTauG1; i++) {
|
|
await fd.write(buffG1);
|
|
if ((logger)&&((i%100000) == 0)&&i) logger.log("betaTauG1: " + i);
|
|
}
|
|
await binFileUtils__namespace.endWriteSection(fd);
|
|
|
|
// Write betaG2
|
|
///////////
|
|
await binFileUtils__namespace.startWriteSection(fd, 6);
|
|
await fd.write(buffG2);
|
|
await binFileUtils__namespace.endWriteSection(fd);
|
|
|
|
// Contributions
|
|
///////////
|
|
await binFileUtils__namespace.startWriteSection(fd, 7);
|
|
await fd.writeULE32(0); // 0 Contributions
|
|
await binFileUtils__namespace.endWriteSection(fd);
|
|
|
|
await fd.close();
|
|
|
|
const firstChallengeHash = calculateFirstChallengeHash(curve, power, logger);
|
|
|
|
if (logger) logger.debug(formatHash(Blake2b__default["default"](64).digest(), "Blank Contribution Hash:"));
|
|
|
|
if (logger) logger.info(formatHash(firstChallengeHash, "First Contribution Hash:"));
|
|
|
|
return firstChallengeHash;
|
|
|
|
}
|
|
|
|
// Format of the outpu
|
|
|
|
async function exportChallenge(pTauFilename, challengeFilename, logger) {
|
|
await Blake2b__default["default"].ready();
|
|
const {fd: fdFrom, sections} = await binFileUtils__namespace.readBinFile(pTauFilename, "ptau", 1);
|
|
|
|
const {curve, power} = await readPTauHeader(fdFrom, sections);
|
|
|
|
const contributions = await readContributions(fdFrom, curve, sections);
|
|
let lastResponseHash, curChallengeHash;
|
|
if (contributions.length == 0) {
|
|
lastResponseHash = Blake2b__default["default"](64).digest();
|
|
curChallengeHash = calculateFirstChallengeHash(curve, power);
|
|
} else {
|
|
lastResponseHash = contributions[contributions.length-1].responseHash;
|
|
curChallengeHash = contributions[contributions.length-1].nextChallenge;
|
|
}
|
|
|
|
if (logger) logger.info(formatHash(lastResponseHash, "Last Response Hash: "));
|
|
|
|
if (logger) logger.info(formatHash(curChallengeHash, "New Challenge Hash: "));
|
|
|
|
|
|
const fdTo = await fastFile__namespace.createOverride(challengeFilename);
|
|
|
|
const toHash = Blake2b__default["default"](64);
|
|
await fdTo.write(lastResponseHash);
|
|
toHash.update(lastResponseHash);
|
|
|
|
await exportSection(2, "G1", (2 ** power) * 2 -1, "tauG1");
|
|
await exportSection(3, "G2", (2 ** power) , "tauG2");
|
|
await exportSection(4, "G1", (2 ** power) , "alphaTauG1");
|
|
await exportSection(5, "G1", (2 ** power) , "betaTauG1");
|
|
await exportSection(6, "G2", 1 , "betaG2");
|
|
|
|
await fdFrom.close();
|
|
await fdTo.close();
|
|
|
|
const calcCurChallengeHash = toHash.digest();
|
|
|
|
if (!hashIsEqual (curChallengeHash, calcCurChallengeHash)) {
|
|
if (logger) logger.info(formatHash(calcCurChallengeHash, "Calc Curret Challenge Hash: "));
|
|
|
|
if (logger) logger.error("PTau file is corrupted. Calculated new challenge hash does not match with the eclared one");
|
|
throw new Error("PTau file is corrupted. Calculated new challenge hash does not match with the eclared one");
|
|
}
|
|
|
|
return curChallengeHash;
|
|
|
|
async function exportSection(sectionId, groupName, nPoints, sectionName) {
|
|
const G = curve[groupName];
|
|
const sG = G.F.n8*2;
|
|
const nPointsChunk = Math.floor((1<<24)/sG);
|
|
|
|
await binFileUtils__namespace.startReadUniqueSection(fdFrom, sections, sectionId);
|
|
for (let i=0; i< nPoints; i+= nPointsChunk) {
|
|
if (logger) logger.debug(`Exporting ${sectionName}: ${i}/${nPoints}`);
|
|
const n = Math.min(nPoints-i, nPointsChunk);
|
|
let buff;
|
|
buff = await fdFrom.read(n*sG);
|
|
buff = await G.batchLEMtoU(buff);
|
|
await fdTo.write(buff);
|
|
toHash.update(buff);
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fdFrom);
|
|
}
|
|
|
|
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function importResponse(oldPtauFilename, contributionFilename, newPTauFilename, name, importPoints, logger) {
|
|
|
|
await Blake2b__default["default"].ready();
|
|
|
|
const noHash = new Uint8Array(64);
|
|
for (let i=0; i<64; i++) noHash[i] = 0xFF;
|
|
|
|
const {fd: fdOld, sections} = await binFileUtils__namespace.readBinFile(oldPtauFilename, "ptau", 1);
|
|
const {curve, power} = await readPTauHeader(fdOld, sections);
|
|
const contributions = await readContributions(fdOld, curve, sections);
|
|
const currentContribution = {};
|
|
|
|
if (name) currentContribution.name = name;
|
|
|
|
const sG1 = curve.F1.n8*2;
|
|
const scG1 = curve.F1.n8; // Compresed size
|
|
const sG2 = curve.F2.n8*2;
|
|
const scG2 = curve.F2.n8; // Compresed size
|
|
|
|
const fdResponse = await fastFile__namespace.readExisting(contributionFilename);
|
|
|
|
if (fdResponse.totalSize !=
|
|
64 + // Old Hash
|
|
((2 ** power)*2-1)*scG1 +
|
|
(2 ** power)*scG2 +
|
|
(2 ** power)*scG1 +
|
|
(2 ** power)*scG1 +
|
|
scG2 +
|
|
sG1*6 + sG2*3)
|
|
throw new Error("Size of the contribution is invalid");
|
|
|
|
let lastChallengeHash;
|
|
|
|
if (contributions.length>0) {
|
|
lastChallengeHash = contributions[contributions.length-1].nextChallenge;
|
|
} else {
|
|
lastChallengeHash = calculateFirstChallengeHash(curve, power, logger);
|
|
}
|
|
|
|
const fdNew = await binFileUtils__namespace.createBinFile(newPTauFilename, "ptau", 1, importPoints ? 7: 2);
|
|
await writePTauHeader(fdNew, curve, power);
|
|
|
|
const contributionPreviousHash = await fdResponse.read(64);
|
|
|
|
if (hashIsEqual(noHash,lastChallengeHash)) {
|
|
lastChallengeHash = contributionPreviousHash;
|
|
contributions[contributions.length-1].nextChallenge = lastChallengeHash;
|
|
}
|
|
|
|
if(!hashIsEqual(contributionPreviousHash,lastChallengeHash))
|
|
throw new Error("Wrong contribution. this contribution is not based on the previus hash");
|
|
|
|
const hasherResponse = new Blake2b__default["default"](64);
|
|
hasherResponse.update(contributionPreviousHash);
|
|
|
|
const startSections = [];
|
|
let res;
|
|
res = await processSection(fdResponse, fdNew, "G1", 2, (2 ** power) * 2 -1, [1], "tauG1");
|
|
currentContribution.tauG1 = res[0];
|
|
res = await processSection(fdResponse, fdNew, "G2", 3, (2 ** power) , [1], "tauG2");
|
|
currentContribution.tauG2 = res[0];
|
|
res = await processSection(fdResponse, fdNew, "G1", 4, (2 ** power) , [0], "alphaG1");
|
|
currentContribution.alphaG1 = res[0];
|
|
res = await processSection(fdResponse, fdNew, "G1", 5, (2 ** power) , [0], "betaG1");
|
|
currentContribution.betaG1 = res[0];
|
|
res = await processSection(fdResponse, fdNew, "G2", 6, 1 , [0], "betaG2");
|
|
currentContribution.betaG2 = res[0];
|
|
|
|
currentContribution.partialHash = hasherResponse.getPartialHash();
|
|
|
|
|
|
const buffKey = await fdResponse.read(curve.F1.n8*2*6+curve.F2.n8*2*3);
|
|
|
|
currentContribution.key = fromPtauPubKeyRpr(buffKey, 0, curve, false);
|
|
|
|
hasherResponse.update(new Uint8Array(buffKey));
|
|
const hashResponse = hasherResponse.digest();
|
|
|
|
if (logger) logger.info(formatHash(hashResponse, "Contribution Response Hash imported: "));
|
|
|
|
if (importPoints) {
|
|
const nextChallengeHasher = new Blake2b__default["default"](64);
|
|
nextChallengeHasher.update(hashResponse);
|
|
|
|
await hashSection(nextChallengeHasher, fdNew, "G1", 2, (2 ** power) * 2 -1, "tauG1", logger);
|
|
await hashSection(nextChallengeHasher, fdNew, "G2", 3, (2 ** power) , "tauG2", logger);
|
|
await hashSection(nextChallengeHasher, fdNew, "G1", 4, (2 ** power) , "alphaTauG1", logger);
|
|
await hashSection(nextChallengeHasher, fdNew, "G1", 5, (2 ** power) , "betaTauG1", logger);
|
|
await hashSection(nextChallengeHasher, fdNew, "G2", 6, 1 , "betaG2", logger);
|
|
|
|
currentContribution.nextChallenge = nextChallengeHasher.digest();
|
|
|
|
if (logger) logger.info(formatHash(currentContribution.nextChallenge, "Next Challenge Hash: "));
|
|
} else {
|
|
currentContribution.nextChallenge = noHash;
|
|
}
|
|
|
|
contributions.push(currentContribution);
|
|
|
|
await writeContributions(fdNew, curve, contributions);
|
|
|
|
await fdResponse.close();
|
|
await fdNew.close();
|
|
await fdOld.close();
|
|
|
|
return currentContribution.nextChallenge;
|
|
|
|
async function processSection(fdFrom, fdTo, groupName, sectionId, nPoints, singularPointIndexes, sectionName) {
|
|
if (importPoints) {
|
|
return await processSectionImportPoints(fdFrom, fdTo, groupName, sectionId, nPoints, singularPointIndexes, sectionName);
|
|
} else {
|
|
return await processSectionNoImportPoints(fdFrom, fdTo, groupName, sectionId, nPoints, singularPointIndexes, sectionName);
|
|
}
|
|
}
|
|
|
|
async function processSectionImportPoints(fdFrom, fdTo, groupName, sectionId, nPoints, singularPointIndexes, sectionName) {
|
|
|
|
const G = curve[groupName];
|
|
const scG = G.F.n8;
|
|
const sG = G.F.n8*2;
|
|
|
|
const singularPoints = [];
|
|
|
|
await binFileUtils__namespace.startWriteSection(fdTo, sectionId);
|
|
const nPointsChunk = Math.floor((1<<24)/sG);
|
|
|
|
startSections[sectionId] = fdTo.pos;
|
|
|
|
for (let i=0; i< nPoints; i += nPointsChunk) {
|
|
if (logger) logger.debug(`Importing ${sectionName}: ${i}/${nPoints}`);
|
|
const n = Math.min(nPoints-i, nPointsChunk);
|
|
|
|
const buffC = await fdFrom.read(n * scG);
|
|
hasherResponse.update(buffC);
|
|
|
|
const buffLEM = await G.batchCtoLEM(buffC);
|
|
|
|
await fdTo.write(buffLEM);
|
|
for (let j=0; j<singularPointIndexes.length; j++) {
|
|
const sp = singularPointIndexes[j];
|
|
if ((sp >=i) && (sp < i+n)) {
|
|
const P = G.fromRprLEM(buffLEM, (sp-i)*sG);
|
|
singularPoints.push(P);
|
|
}
|
|
}
|
|
}
|
|
|
|
await binFileUtils__namespace.endWriteSection(fdTo);
|
|
|
|
return singularPoints;
|
|
}
|
|
|
|
|
|
async function processSectionNoImportPoints(fdFrom, fdTo, groupName, sectionId, nPoints, singularPointIndexes, sectionName) {
|
|
|
|
const G = curve[groupName];
|
|
const scG = G.F.n8;
|
|
|
|
const singularPoints = [];
|
|
|
|
const nPointsChunk = Math.floor((1<<24)/scG);
|
|
|
|
for (let i=0; i< nPoints; i += nPointsChunk) {
|
|
if (logger) logger.debug(`Importing ${sectionName}: ${i}/${nPoints}`);
|
|
const n = Math.min(nPoints-i, nPointsChunk);
|
|
|
|
const buffC = await fdFrom.read(n * scG);
|
|
hasherResponse.update(buffC);
|
|
|
|
for (let j=0; j<singularPointIndexes.length; j++) {
|
|
const sp = singularPointIndexes[j];
|
|
if ((sp >=i) && (sp < i+n)) {
|
|
const P = G.fromRprCompressed(buffC, (sp-i)*scG);
|
|
singularPoints.push(P);
|
|
}
|
|
}
|
|
}
|
|
|
|
return singularPoints;
|
|
}
|
|
|
|
|
|
async function hashSection(nextChallengeHasher, fdTo, groupName, sectionId, nPoints, sectionName, logger) {
|
|
|
|
const G = curve[groupName];
|
|
const sG = G.F.n8*2;
|
|
const nPointsChunk = Math.floor((1<<24)/sG);
|
|
|
|
const oldPos = fdTo.pos;
|
|
fdTo.pos = startSections[sectionId];
|
|
|
|
for (let i=0; i< nPoints; i += nPointsChunk) {
|
|
if (logger) logger.debug(`Hashing ${sectionName}: ${i}/${nPoints}`);
|
|
const n = Math.min(nPoints-i, nPointsChunk);
|
|
|
|
const buffLEM = await fdTo.read(n * sG);
|
|
|
|
const buffU = await G.batchLEMtoU(buffLEM);
|
|
|
|
nextChallengeHasher.update(buffU);
|
|
}
|
|
|
|
fdTo.pos = oldPos;
|
|
}
|
|
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
const sameRatio$1 = sameRatio$2;
|
|
|
|
async function verifyContribution(curve, cur, prev, logger) {
|
|
let sr;
|
|
if (cur.type == 1) { // Verify the beacon.
|
|
const beaconKey = keyFromBeacon(curve, prev.nextChallenge, cur.beaconHash, cur.numIterationsExp);
|
|
|
|
if (!curve.G1.eq(cur.key.tau.g1_s, beaconKey.tau.g1_s)) {
|
|
if (logger) logger.error(`BEACON key (tauG1_s) is not generated correctly in challenge #${cur.id} ${cur.name || ""}` );
|
|
return false;
|
|
}
|
|
if (!curve.G1.eq(cur.key.tau.g1_sx, beaconKey.tau.g1_sx)) {
|
|
if (logger) logger.error(`BEACON key (tauG1_sx) is not generated correctly in challenge #${cur.id} ${cur.name || ""}` );
|
|
return false;
|
|
}
|
|
if (!curve.G2.eq(cur.key.tau.g2_spx, beaconKey.tau.g2_spx)) {
|
|
if (logger) logger.error(`BEACON key (tauG2_spx) is not generated correctly in challenge #${cur.id} ${cur.name || ""}` );
|
|
return false;
|
|
}
|
|
|
|
if (!curve.G1.eq(cur.key.alpha.g1_s, beaconKey.alpha.g1_s)) {
|
|
if (logger) logger.error(`BEACON key (alphaG1_s) is not generated correctly in challenge #${cur.id} ${cur.name || ""}` );
|
|
return false;
|
|
}
|
|
if (!curve.G1.eq(cur.key.alpha.g1_sx, beaconKey.alpha.g1_sx)) {
|
|
if (logger) logger.error(`BEACON key (alphaG1_sx) is not generated correctly in challenge #${cur.id} ${cur.name || ""}` );
|
|
return false;
|
|
}
|
|
if (!curve.G2.eq(cur.key.alpha.g2_spx, beaconKey.alpha.g2_spx)) {
|
|
if (logger) logger.error(`BEACON key (alphaG2_spx) is not generated correctly in challenge #${cur.id} ${cur.name || ""}` );
|
|
return false;
|
|
}
|
|
|
|
if (!curve.G1.eq(cur.key.beta.g1_s, beaconKey.beta.g1_s)) {
|
|
if (logger) logger.error(`BEACON key (betaG1_s) is not generated correctly in challenge #${cur.id} ${cur.name || ""}` );
|
|
return false;
|
|
}
|
|
if (!curve.G1.eq(cur.key.beta.g1_sx, beaconKey.beta.g1_sx)) {
|
|
if (logger) logger.error(`BEACON key (betaG1_sx) is not generated correctly in challenge #${cur.id} ${cur.name || ""}` );
|
|
return false;
|
|
}
|
|
if (!curve.G2.eq(cur.key.beta.g2_spx, beaconKey.beta.g2_spx)) {
|
|
if (logger) logger.error(`BEACON key (betaG2_spx) is not generated correctly in challenge #${cur.id} ${cur.name || ""}` );
|
|
return false;
|
|
}
|
|
}
|
|
|
|
cur.key.tau.g2_sp = curve.G2.toAffine(getG2sp(curve, 0, prev.nextChallenge, cur.key.tau.g1_s, cur.key.tau.g1_sx));
|
|
cur.key.alpha.g2_sp = curve.G2.toAffine(getG2sp(curve, 1, prev.nextChallenge, cur.key.alpha.g1_s, cur.key.alpha.g1_sx));
|
|
cur.key.beta.g2_sp = curve.G2.toAffine(getG2sp(curve, 2, prev.nextChallenge, cur.key.beta.g1_s, cur.key.beta.g1_sx));
|
|
|
|
sr = await sameRatio$1(curve, cur.key.tau.g1_s, cur.key.tau.g1_sx, cur.key.tau.g2_sp, cur.key.tau.g2_spx);
|
|
if (sr !== true) {
|
|
if (logger) logger.error("INVALID key (tau) in challenge #"+cur.id);
|
|
return false;
|
|
}
|
|
|
|
sr = await sameRatio$1(curve, cur.key.alpha.g1_s, cur.key.alpha.g1_sx, cur.key.alpha.g2_sp, cur.key.alpha.g2_spx);
|
|
if (sr !== true) {
|
|
if (logger) logger.error("INVALID key (alpha) in challenge #"+cur.id);
|
|
return false;
|
|
}
|
|
|
|
sr = await sameRatio$1(curve, cur.key.beta.g1_s, cur.key.beta.g1_sx, cur.key.beta.g2_sp, cur.key.beta.g2_spx);
|
|
if (sr !== true) {
|
|
if (logger) logger.error("INVALID key (beta) in challenge #"+cur.id);
|
|
return false;
|
|
}
|
|
|
|
sr = await sameRatio$1(curve, prev.tauG1, cur.tauG1, cur.key.tau.g2_sp, cur.key.tau.g2_spx);
|
|
if (sr !== true) {
|
|
if (logger) logger.error("INVALID tau*G1. challenge #"+cur.id+" It does not follow the previous contribution");
|
|
return false;
|
|
}
|
|
|
|
sr = await sameRatio$1(curve, cur.key.tau.g1_s, cur.key.tau.g1_sx, prev.tauG2, cur.tauG2);
|
|
if (sr !== true) {
|
|
if (logger) logger.error("INVALID tau*G2. challenge #"+cur.id+" It does not follow the previous contribution");
|
|
return false;
|
|
}
|
|
|
|
sr = await sameRatio$1(curve, prev.alphaG1, cur.alphaG1, cur.key.alpha.g2_sp, cur.key.alpha.g2_spx);
|
|
if (sr !== true) {
|
|
if (logger) logger.error("INVALID alpha*G1. challenge #"+cur.id+" It does not follow the previous contribution");
|
|
return false;
|
|
}
|
|
|
|
sr = await sameRatio$1(curve, prev.betaG1, cur.betaG1, cur.key.beta.g2_sp, cur.key.beta.g2_spx);
|
|
if (sr !== true) {
|
|
if (logger) logger.error("INVALID beta*G1. challenge #"+cur.id+" It does not follow the previous contribution");
|
|
return false;
|
|
}
|
|
|
|
sr = await sameRatio$1(curve, cur.key.beta.g1_s, cur.key.beta.g1_sx, prev.betaG2, cur.betaG2);
|
|
if (sr !== true) {
|
|
if (logger) logger.error("INVALID beta*G2. challenge #"+cur.id+"It does not follow the previous contribution");
|
|
return false;
|
|
}
|
|
|
|
if (logger) logger.info("Powers Of tau file OK!");
|
|
return true;
|
|
}
|
|
|
|
async function verify(tauFilename, logger) {
|
|
let sr;
|
|
await Blake2b__default["default"].ready();
|
|
|
|
const {fd, sections} = await binFileUtils__namespace.readBinFile(tauFilename, "ptau", 1);
|
|
const {curve, power, ceremonyPower} = await readPTauHeader(fd, sections);
|
|
const contrs = await readContributions(fd, curve, sections);
|
|
|
|
if (logger) logger.debug("power: 2**" + power);
|
|
// Verify Last contribution
|
|
|
|
if (logger) logger.debug("Computing initial contribution hash");
|
|
const initialContribution = {
|
|
tauG1: curve.G1.g,
|
|
tauG2: curve.G2.g,
|
|
alphaG1: curve.G1.g,
|
|
betaG1: curve.G1.g,
|
|
betaG2: curve.G2.g,
|
|
nextChallenge: calculateFirstChallengeHash(curve, ceremonyPower, logger),
|
|
responseHash: Blake2b__default["default"](64).digest()
|
|
};
|
|
|
|
if (contrs.length == 0) {
|
|
if (logger) logger.error("This file has no contribution! It cannot be used in production");
|
|
return false;
|
|
}
|
|
|
|
let prevContr;
|
|
if (contrs.length>1) {
|
|
prevContr = contrs[contrs.length-2];
|
|
} else {
|
|
prevContr = initialContribution;
|
|
}
|
|
const curContr = contrs[contrs.length-1];
|
|
if (logger) logger.debug("Validating contribution #"+contrs[contrs.length-1].id);
|
|
const res = await verifyContribution(curve, curContr, prevContr, logger);
|
|
if (!res) return false;
|
|
|
|
|
|
const nextContributionHasher = Blake2b__default["default"](64);
|
|
nextContributionHasher.update(curContr.responseHash);
|
|
|
|
// Verify powers and compute nextChallengeHash
|
|
|
|
// await test();
|
|
|
|
// Verify Section tau*G1
|
|
if (logger) logger.debug("Verifying powers in tau*G1 section");
|
|
const rTau1 = await processSection(2, "G1", "tauG1", (2 ** power)*2-1, [0, 1], logger);
|
|
sr = await sameRatio$1(curve, rTau1.R1, rTau1.R2, curve.G2.g, curContr.tauG2);
|
|
if (sr !== true) {
|
|
if (logger) logger.error("tauG1 section. Powers do not match");
|
|
return false;
|
|
}
|
|
if (!curve.G1.eq(curve.G1.g, rTau1.singularPoints[0])) {
|
|
if (logger) logger.error("First element of tau*G1 section must be the generator");
|
|
return false;
|
|
}
|
|
if (!curve.G1.eq(curContr.tauG1, rTau1.singularPoints[1])) {
|
|
if (logger) logger.error("Second element of tau*G1 section does not match the one in the contribution section");
|
|
return false;
|
|
}
|
|
|
|
// await test();
|
|
|
|
// Verify Section tau*G2
|
|
if (logger) logger.debug("Verifying powers in tau*G2 section");
|
|
const rTau2 = await processSection(3, "G2", "tauG2", 2 ** power, [0, 1], logger);
|
|
sr = await sameRatio$1(curve, curve.G1.g, curContr.tauG1, rTau2.R1, rTau2.R2);
|
|
if (sr !== true) {
|
|
if (logger) logger.error("tauG2 section. Powers do not match");
|
|
return false;
|
|
}
|
|
if (!curve.G2.eq(curve.G2.g, rTau2.singularPoints[0])) {
|
|
if (logger) logger.error("First element of tau*G2 section must be the generator");
|
|
return false;
|
|
}
|
|
if (!curve.G2.eq(curContr.tauG2, rTau2.singularPoints[1])) {
|
|
if (logger) logger.error("Second element of tau*G2 section does not match the one in the contribution section");
|
|
return false;
|
|
}
|
|
|
|
// Verify Section alpha*tau*G1
|
|
if (logger) logger.debug("Verifying powers in alpha*tau*G1 section");
|
|
const rAlphaTauG1 = await processSection(4, "G1", "alphatauG1", 2 ** power, [0], logger);
|
|
sr = await sameRatio$1(curve, rAlphaTauG1.R1, rAlphaTauG1.R2, curve.G2.g, curContr.tauG2);
|
|
if (sr !== true) {
|
|
if (logger) logger.error("alphaTauG1 section. Powers do not match");
|
|
return false;
|
|
}
|
|
if (!curve.G1.eq(curContr.alphaG1, rAlphaTauG1.singularPoints[0])) {
|
|
if (logger) logger.error("First element of alpha*tau*G1 section (alpha*G1) does not match the one in the contribution section");
|
|
return false;
|
|
}
|
|
|
|
// Verify Section beta*tau*G1
|
|
if (logger) logger.debug("Verifying powers in beta*tau*G1 section");
|
|
const rBetaTauG1 = await processSection(5, "G1", "betatauG1", 2 ** power, [0], logger);
|
|
sr = await sameRatio$1(curve, rBetaTauG1.R1, rBetaTauG1.R2, curve.G2.g, curContr.tauG2);
|
|
if (sr !== true) {
|
|
if (logger) logger.error("betaTauG1 section. Powers do not match");
|
|
return false;
|
|
}
|
|
if (!curve.G1.eq(curContr.betaG1, rBetaTauG1.singularPoints[0])) {
|
|
if (logger) logger.error("First element of beta*tau*G1 section (beta*G1) does not match the one in the contribution section");
|
|
return false;
|
|
}
|
|
|
|
//Verify Beta G2
|
|
const betaG2 = await processSectionBetaG2(logger);
|
|
if (!curve.G2.eq(curContr.betaG2, betaG2)) {
|
|
if (logger) logger.error("betaG2 element in betaG2 section does not match the one in the contribution section");
|
|
return false;
|
|
}
|
|
|
|
|
|
const nextContributionHash = nextContributionHasher.digest();
|
|
|
|
// Check the nextChallengeHash
|
|
if (power == ceremonyPower) {
|
|
if (!hashIsEqual(nextContributionHash,curContr.nextChallenge)) {
|
|
if (logger) logger.error("Hash of the values does not match the next challenge of the last contributor in the contributions section");
|
|
return false;
|
|
}
|
|
}
|
|
|
|
if (logger) logger.info(formatHash(nextContributionHash, "Next challenge hash: "));
|
|
|
|
// Verify Previous contributions
|
|
|
|
printContribution(curContr, prevContr);
|
|
for (let i = contrs.length-2; i>=0; i--) {
|
|
const curContr = contrs[i];
|
|
const prevContr = (i>0) ? contrs[i-1] : initialContribution;
|
|
const res = await verifyContribution(curve, curContr, prevContr, logger);
|
|
if (!res) return false;
|
|
printContribution(curContr, prevContr);
|
|
}
|
|
if (logger) logger.info("-----------------------------------------------------");
|
|
|
|
if ((!sections[12]) || (!sections[13]) || (!sections[14]) || (!sections[15])) {
|
|
if (logger) logger.warn(
|
|
"this file does not contain phase2 precalculated values. Please run: \n" +
|
|
" snarkjs \"powersoftau preparephase2\" to prepare this file to be used in the phase2 ceremony."
|
|
);
|
|
} else {
|
|
let res;
|
|
res = await verifyLagrangeEvaluations("G1", 2, 12, "tauG1", logger);
|
|
if (!res) return false;
|
|
res = await verifyLagrangeEvaluations("G2", 3, 13, "tauG2", logger);
|
|
if (!res) return false;
|
|
res = await verifyLagrangeEvaluations("G1", 4, 14, "alphaTauG1", logger);
|
|
if (!res) return false;
|
|
res = await verifyLagrangeEvaluations("G1", 5, 15, "betaTauG1", logger);
|
|
if (!res) return false;
|
|
}
|
|
|
|
await fd.close();
|
|
|
|
if (logger) logger.info("Powers of Tau Ok!");
|
|
|
|
return true;
|
|
|
|
function printContribution(curContr, prevContr) {
|
|
if (!logger) return;
|
|
logger.info("-----------------------------------------------------");
|
|
logger.info(`Contribution #${curContr.id}: ${curContr.name ||""}`);
|
|
|
|
logger.info(formatHash(curContr.nextChallenge, "Next Challenge: "));
|
|
|
|
const buffV = new Uint8Array(curve.G1.F.n8*2*6+curve.G2.F.n8*2*3);
|
|
toPtauPubKeyRpr(buffV, 0, curve, curContr.key, false);
|
|
|
|
const responseHasher = Blake2b__default["default"](64);
|
|
responseHasher.setPartialHash(curContr.partialHash);
|
|
responseHasher.update(buffV);
|
|
const responseHash = responseHasher.digest();
|
|
|
|
logger.info(formatHash(responseHash, "Response Hash:"));
|
|
|
|
logger.info(formatHash(prevContr.nextChallenge, "Response Hash:"));
|
|
|
|
if (curContr.type == 1) {
|
|
logger.info(`Beacon generator: ${byteArray2hex(curContr.beaconHash)}`);
|
|
logger.info(`Beacon iterations Exp: ${curContr.numIterationsExp}`);
|
|
}
|
|
|
|
}
|
|
|
|
async function processSectionBetaG2(logger) {
|
|
const G = curve.G2;
|
|
const sG = G.F.n8*2;
|
|
const buffUv = new Uint8Array(sG);
|
|
|
|
if (!sections[6]) {
|
|
logger.error("File has no BetaG2 section");
|
|
throw new Error("File has no BetaG2 section");
|
|
}
|
|
if (sections[6].length>1) {
|
|
logger.error("File has no BetaG2 section");
|
|
throw new Error("File has more than one GetaG2 section");
|
|
}
|
|
fd.pos = sections[6][0].p;
|
|
|
|
const buff = await fd.read(sG);
|
|
const P = G.fromRprLEM(buff);
|
|
|
|
G.toRprUncompressed(buffUv, 0, P);
|
|
nextContributionHasher.update(buffUv);
|
|
|
|
return P;
|
|
}
|
|
|
|
async function processSection(idSection, groupName, sectionName, nPoints, singularPointIndexes, logger) {
|
|
const MAX_CHUNK_SIZE = 1<<16;
|
|
const G = curve[groupName];
|
|
const sG = G.F.n8*2;
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, idSection);
|
|
|
|
const singularPoints = [];
|
|
|
|
let R1 = G.zero;
|
|
let R2 = G.zero;
|
|
|
|
let lastBase = G.zero;
|
|
|
|
for (let i=0; i<nPoints; i += MAX_CHUNK_SIZE) {
|
|
if (logger) logger.debug(`points relations: ${sectionName}: ${i}/${nPoints} `);
|
|
const n = Math.min(nPoints - i, MAX_CHUNK_SIZE);
|
|
const bases = await fd.read(n*sG);
|
|
|
|
const basesU = await G.batchLEMtoU(bases);
|
|
nextContributionHasher.update(basesU);
|
|
|
|
const scalars = new Uint8Array(4*(n-1));
|
|
crypto__default["default"].randomFillSync(scalars);
|
|
|
|
|
|
if (i>0) {
|
|
const firstBase = G.fromRprLEM(bases, 0);
|
|
const r = crypto__default["default"].randomBytes(4).readUInt32BE(0, true);
|
|
|
|
R1 = G.add(R1, G.timesScalar(lastBase, r));
|
|
R2 = G.add(R2, G.timesScalar(firstBase, r));
|
|
}
|
|
|
|
const r1 = await G.multiExpAffine(bases.slice(0, (n-1)*sG), scalars);
|
|
const r2 = await G.multiExpAffine(bases.slice(sG), scalars);
|
|
|
|
R1 = G.add(R1, r1);
|
|
R2 = G.add(R2, r2);
|
|
|
|
lastBase = G.fromRprLEM( bases, (n-1)*sG);
|
|
|
|
for (let j=0; j<singularPointIndexes.length; j++) {
|
|
const sp = singularPointIndexes[j];
|
|
if ((sp >=i) && (sp < i+n)) {
|
|
const P = G.fromRprLEM(bases, (sp-i)*sG);
|
|
singularPoints.push(P);
|
|
}
|
|
}
|
|
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
return {
|
|
R1: R1,
|
|
R2: R2,
|
|
singularPoints: singularPoints
|
|
};
|
|
|
|
}
|
|
|
|
async function verifyLagrangeEvaluations(gName, tauSection, lagrangeSection, sectionName, logger) {
|
|
|
|
if (logger) logger.debug(`Verifying phase2 calculated values ${sectionName}...`);
|
|
const G = curve[gName];
|
|
const sG = G.F.n8*2;
|
|
|
|
const seed= new Array(8);
|
|
for (let i=0; i<8; i++) {
|
|
seed[i] = crypto__default["default"].randomBytes(4).readUInt32BE(0, true);
|
|
}
|
|
|
|
for (let p=0; p<= power; p ++) {
|
|
const res = await verifyPower(p);
|
|
if (!res) return false;
|
|
}
|
|
|
|
if (tauSection == 2) {
|
|
const res = await verifyPower(power+1);
|
|
if (!res) return false;
|
|
}
|
|
|
|
return true;
|
|
|
|
async function verifyPower(p) {
|
|
if (logger) logger.debug(`Power ${p}...`);
|
|
const n8r = curve.Fr.n8;
|
|
const nPoints = 2 ** p;
|
|
let buff_r = new Uint32Array(nPoints);
|
|
let buffG;
|
|
|
|
let rng = new ffjavascript.ChaCha(seed);
|
|
|
|
if (logger) logger.debug(`Creating random numbers Powers${p}...`);
|
|
for (let i=0; i<nPoints; i++) {
|
|
if ((p == power+1)&&(i == nPoints-1)) {
|
|
buff_r[i] = 0;
|
|
} else {
|
|
buff_r[i] = rng.nextU32();
|
|
}
|
|
}
|
|
|
|
buff_r = new Uint8Array(buff_r.buffer, buff_r.byteOffset, buff_r.byteLength);
|
|
|
|
if (logger) logger.debug(`reading points Powers${p}...`);
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, tauSection);
|
|
buffG = new ffjavascript.BigBuffer(nPoints*sG);
|
|
if (p == power+1) {
|
|
await fd.readToBuffer(buffG, 0, (nPoints-1)*sG);
|
|
buffG.set(curve.G1.zeroAffine, (nPoints-1)*sG);
|
|
} else {
|
|
await fd.readToBuffer(buffG, 0, nPoints*sG);
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd, true);
|
|
|
|
const resTau = await G.multiExpAffine(buffG, buff_r, logger, sectionName + "_" + p);
|
|
|
|
buff_r = new ffjavascript.BigBuffer(nPoints * n8r);
|
|
|
|
rng = new ffjavascript.ChaCha(seed);
|
|
|
|
const buff4 = new Uint8Array(4);
|
|
const buff4V = new DataView(buff4.buffer);
|
|
|
|
if (logger) logger.debug(`Creating random numbers Powers${p}...`);
|
|
for (let i=0; i<nPoints; i++) {
|
|
if ((i != nPoints-1) || (p != power+1)) {
|
|
buff4V.setUint32(0, rng.nextU32(), true);
|
|
buff_r.set(buff4, i*n8r);
|
|
}
|
|
}
|
|
|
|
if (logger) logger.debug(`batchToMontgomery ${p}...`);
|
|
buff_r = await curve.Fr.batchToMontgomery(buff_r);
|
|
if (logger) logger.debug(`fft ${p}...`);
|
|
buff_r = await curve.Fr.fft(buff_r);
|
|
if (logger) logger.debug(`batchFromMontgomery ${p}...`);
|
|
buff_r = await curve.Fr.batchFromMontgomery(buff_r);
|
|
|
|
if (logger) logger.debug(`reading points Lagrange${p}...`);
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, lagrangeSection);
|
|
fd.pos += sG*((2 ** p)-1);
|
|
await fd.readToBuffer(buffG, 0, nPoints*sG);
|
|
await binFileUtils__namespace.endReadSection(fd, true);
|
|
|
|
const resLagrange = await G.multiExpAffine(buffG, buff_r, logger, sectionName + "_" + p + "_transformed");
|
|
|
|
if (!G.eq(resTau, resLagrange)) {
|
|
if (logger) logger.error("Phase2 caclutation does not match with powers of tau");
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
}
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
/*
|
|
This function creates a new section in the fdTo file with id idSection.
|
|
It multiplies the pooints in fdFrom by first, first*inc, first*inc^2, ....
|
|
nPoint Times.
|
|
It also updates the newChallengeHasher with the new points
|
|
*/
|
|
|
|
async function applyKeyToSection(fdOld, sections, fdNew, idSection, curve, groupName, first, inc, sectionName, logger) {
|
|
const MAX_CHUNK_SIZE = 1 << 16;
|
|
const G = curve[groupName];
|
|
const sG = G.F.n8*2;
|
|
const nPoints = sections[idSection][0].size / sG;
|
|
|
|
await binFileUtils__namespace.startReadUniqueSection(fdOld, sections,idSection );
|
|
await binFileUtils__namespace.startWriteSection(fdNew, idSection);
|
|
|
|
let t = first;
|
|
for (let i=0; i<nPoints; i += MAX_CHUNK_SIZE) {
|
|
if (logger) logger.debug(`Applying key: ${sectionName}: ${i}/${nPoints}`);
|
|
const n= Math.min(nPoints - i, MAX_CHUNK_SIZE);
|
|
let buff;
|
|
buff = await fdOld.read(n*sG);
|
|
buff = await G.batchApplyKey(buff, t, inc);
|
|
await fdNew.write(buff);
|
|
t = curve.Fr.mul(t, curve.Fr.exp(inc, n));
|
|
}
|
|
|
|
await binFileUtils__namespace.endWriteSection(fdNew);
|
|
await binFileUtils__namespace.endReadSection(fdOld);
|
|
}
|
|
|
|
|
|
|
|
async function applyKeyToChallengeSection(fdOld, fdNew, responseHasher, curve, groupName, nPoints, first, inc, formatOut, sectionName, logger) {
|
|
const G = curve[groupName];
|
|
const sG = G.F.n8*2;
|
|
const chunkSize = Math.floor((1<<20) / sG); // 128Mb chunks
|
|
let t = first;
|
|
for (let i=0 ; i<nPoints ; i+= chunkSize) {
|
|
if (logger) logger.debug(`Applying key ${sectionName}: ${i}/${nPoints}`);
|
|
const n= Math.min(nPoints-i, chunkSize );
|
|
const buffInU = await fdOld.read(n * sG);
|
|
const buffInLEM = await G.batchUtoLEM(buffInU);
|
|
const buffOutLEM = await G.batchApplyKey(buffInLEM, t, inc);
|
|
let buffOut;
|
|
if (formatOut == "COMPRESSED") {
|
|
buffOut = await G.batchLEMtoC(buffOutLEM);
|
|
} else {
|
|
buffOut = await G.batchLEMtoU(buffOutLEM);
|
|
}
|
|
|
|
if (responseHasher) responseHasher.update(buffOut);
|
|
await fdNew.write(buffOut);
|
|
t = curve.Fr.mul(t, curve.Fr.exp(inc, n));
|
|
}
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function challengeContribute(curve, challengeFilename, responesFileName, entropy, logger) {
|
|
await Blake2b__default["default"].ready();
|
|
|
|
const fdFrom = await fastFile__namespace.readExisting(challengeFilename);
|
|
|
|
|
|
const sG1 = curve.F1.n64*8*2;
|
|
const sG2 = curve.F2.n64*8*2;
|
|
const domainSize = (fdFrom.totalSize + sG1 - 64 - sG2) / (4*sG1 + sG2);
|
|
let e = domainSize;
|
|
let power = 0;
|
|
while (e>1) {
|
|
e = e /2;
|
|
power += 1;
|
|
}
|
|
|
|
if (2 ** power != domainSize) throw new Error("Invalid file size");
|
|
if (logger) logger.debug("Power to tau size: "+power);
|
|
|
|
const rng = await getRandomRng(entropy);
|
|
|
|
const fdTo = await fastFile__namespace.createOverride(responesFileName);
|
|
|
|
// Calculate the hash
|
|
const challengeHasher = Blake2b__default["default"](64);
|
|
for (let i=0; i<fdFrom.totalSize; i+= fdFrom.pageSize) {
|
|
if (logger) logger.debug(`Hashing challenge ${i}/${fdFrom.totalSize}`);
|
|
const s = Math.min(fdFrom.totalSize - i, fdFrom.pageSize);
|
|
const buff = await fdFrom.read(s);
|
|
challengeHasher.update(buff);
|
|
}
|
|
|
|
const claimedHash = await fdFrom.read(64, 0);
|
|
if (logger) logger.info(formatHash(claimedHash, "Claimed Previous Response Hash: "));
|
|
|
|
const challengeHash = challengeHasher.digest();
|
|
if (logger) logger.info(formatHash(challengeHash, "Current Challenge Hash: "));
|
|
|
|
const key = createPTauKey(curve, challengeHash, rng);
|
|
|
|
if (logger) {
|
|
["tau", "alpha", "beta"].forEach( (k) => {
|
|
logger.debug(k + ".g1_s: " + curve.G1.toString(key[k].g1_s, 16));
|
|
logger.debug(k + ".g1_sx: " + curve.G1.toString(key[k].g1_sx, 16));
|
|
logger.debug(k + ".g2_sp: " + curve.G2.toString(key[k].g2_sp, 16));
|
|
logger.debug(k + ".g2_spx: " + curve.G2.toString(key[k].g2_spx, 16));
|
|
logger.debug("");
|
|
});
|
|
}
|
|
|
|
const responseHasher = Blake2b__default["default"](64);
|
|
|
|
await fdTo.write(challengeHash);
|
|
responseHasher.update(challengeHash);
|
|
|
|
await applyKeyToChallengeSection(fdFrom, fdTo, responseHasher, curve, "G1", (2 ** power)*2-1, curve.Fr.one , key.tau.prvKey, "COMPRESSED", "tauG1" , logger );
|
|
await applyKeyToChallengeSection(fdFrom, fdTo, responseHasher, curve, "G2", (2 ** power) , curve.Fr.one , key.tau.prvKey, "COMPRESSED", "tauG2" , logger );
|
|
await applyKeyToChallengeSection(fdFrom, fdTo, responseHasher, curve, "G1", (2 ** power) , key.alpha.prvKey, key.tau.prvKey, "COMPRESSED", "alphaTauG1", logger );
|
|
await applyKeyToChallengeSection(fdFrom, fdTo, responseHasher, curve, "G1", (2 ** power) , key.beta.prvKey , key.tau.prvKey, "COMPRESSED", "betaTauG1" , logger );
|
|
await applyKeyToChallengeSection(fdFrom, fdTo, responseHasher, curve, "G2", 1 , key.beta.prvKey , key.tau.prvKey, "COMPRESSED", "betaTauG2" , logger );
|
|
|
|
// Write and hash key
|
|
const buffKey = new Uint8Array(curve.F1.n8*2*6+curve.F2.n8*2*3);
|
|
toPtauPubKeyRpr(buffKey, 0, curve, key, false);
|
|
await fdTo.write(buffKey);
|
|
responseHasher.update(buffKey);
|
|
const responseHash = responseHasher.digest();
|
|
if (logger) logger.info(formatHash(responseHash, "Contribution Response Hash: "));
|
|
|
|
await fdTo.close();
|
|
await fdFrom.close();
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function beacon$1(oldPtauFilename, newPTauFilename, name, beaconHashStr,numIterationsExp, logger) {
|
|
const beaconHash = hex2ByteArray(beaconHashStr);
|
|
if ( (beaconHash.byteLength == 0)
|
|
|| (beaconHash.byteLength*2 !=beaconHashStr.length))
|
|
{
|
|
if (logger) logger.error("Invalid Beacon Hash. (It must be a valid hexadecimal sequence)");
|
|
return false;
|
|
}
|
|
if (beaconHash.length>=256) {
|
|
if (logger) logger.error("Maximum lenght of beacon hash is 255 bytes");
|
|
return false;
|
|
}
|
|
|
|
numIterationsExp = parseInt(numIterationsExp);
|
|
if ((numIterationsExp<10)||(numIterationsExp>63)) {
|
|
if (logger) logger.error("Invalid numIterationsExp. (Must be between 10 and 63)");
|
|
return false;
|
|
}
|
|
|
|
|
|
await Blake2b__default["default"].ready();
|
|
|
|
const {fd: fdOld, sections} = await binFileUtils__namespace.readBinFile(oldPtauFilename, "ptau", 1);
|
|
const {curve, power, ceremonyPower} = await readPTauHeader(fdOld, sections);
|
|
if (power != ceremonyPower) {
|
|
if (logger) logger.error("This file has been reduced. You cannot contribute into a reduced file.");
|
|
return false;
|
|
}
|
|
if (sections[12]) {
|
|
if (logger) logger.warn("Contributing into a file that has phase2 calculated. You will have to prepare phase2 again.");
|
|
}
|
|
const contributions = await readContributions(fdOld, curve, sections);
|
|
const curContribution = {
|
|
name: name,
|
|
type: 1, // Beacon
|
|
numIterationsExp: numIterationsExp,
|
|
beaconHash: beaconHash
|
|
};
|
|
|
|
let lastChallengeHash;
|
|
|
|
if (contributions.length>0) {
|
|
lastChallengeHash = contributions[contributions.length-1].nextChallenge;
|
|
} else {
|
|
lastChallengeHash = calculateFirstChallengeHash(curve, power, logger);
|
|
}
|
|
|
|
curContribution.key = keyFromBeacon(curve, lastChallengeHash, beaconHash, numIterationsExp);
|
|
|
|
const responseHasher = new Blake2b__default["default"](64);
|
|
responseHasher.update(lastChallengeHash);
|
|
|
|
const fdNew = await binFileUtils__namespace.createBinFile(newPTauFilename, "ptau", 1, 7);
|
|
await writePTauHeader(fdNew, curve, power);
|
|
|
|
const startSections = [];
|
|
|
|
let firstPoints;
|
|
firstPoints = await processSection(2, "G1", (2 ** power) * 2 -1, curve.Fr.e(1), curContribution.key.tau.prvKey, "tauG1", logger );
|
|
curContribution.tauG1 = firstPoints[1];
|
|
firstPoints = await processSection(3, "G2", (2 ** power) , curve.Fr.e(1), curContribution.key.tau.prvKey, "tauG2", logger );
|
|
curContribution.tauG2 = firstPoints[1];
|
|
firstPoints = await processSection(4, "G1", (2 ** power) , curContribution.key.alpha.prvKey, curContribution.key.tau.prvKey, "alphaTauG1", logger );
|
|
curContribution.alphaG1 = firstPoints[0];
|
|
firstPoints = await processSection(5, "G1", (2 ** power) , curContribution.key.beta.prvKey, curContribution.key.tau.prvKey, "betaTauG1", logger );
|
|
curContribution.betaG1 = firstPoints[0];
|
|
firstPoints = await processSection(6, "G2", 1, curContribution.key.beta.prvKey, curContribution.key.tau.prvKey, "betaTauG2", logger );
|
|
curContribution.betaG2 = firstPoints[0];
|
|
|
|
curContribution.partialHash = responseHasher.getPartialHash();
|
|
|
|
const buffKey = new Uint8Array(curve.F1.n8*2*6+curve.F2.n8*2*3);
|
|
|
|
toPtauPubKeyRpr(buffKey, 0, curve, curContribution.key, false);
|
|
|
|
responseHasher.update(new Uint8Array(buffKey));
|
|
const hashResponse = responseHasher.digest();
|
|
|
|
if (logger) logger.info(formatHash(hashResponse, "Contribution Response Hash imported: "));
|
|
|
|
const nextChallengeHasher = new Blake2b__default["default"](64);
|
|
nextChallengeHasher.update(hashResponse);
|
|
|
|
await hashSection(fdNew, "G1", 2, (2 ** power) * 2 -1, "tauG1", logger);
|
|
await hashSection(fdNew, "G2", 3, (2 ** power) , "tauG2", logger);
|
|
await hashSection(fdNew, "G1", 4, (2 ** power) , "alphaTauG1", logger);
|
|
await hashSection(fdNew, "G1", 5, (2 ** power) , "betaTauG1", logger);
|
|
await hashSection(fdNew, "G2", 6, 1 , "betaG2", logger);
|
|
|
|
curContribution.nextChallenge = nextChallengeHasher.digest();
|
|
|
|
if (logger) logger.info(formatHash(curContribution.nextChallenge, "Next Challenge Hash: "));
|
|
|
|
contributions.push(curContribution);
|
|
|
|
await writeContributions(fdNew, curve, contributions);
|
|
|
|
await fdOld.close();
|
|
await fdNew.close();
|
|
|
|
return hashResponse;
|
|
|
|
async function processSection(sectionId, groupName, NPoints, first, inc, sectionName, logger) {
|
|
const res = [];
|
|
fdOld.pos = sections[sectionId][0].p;
|
|
|
|
await binFileUtils__namespace.startWriteSection(fdNew, sectionId);
|
|
|
|
startSections[sectionId] = fdNew.pos;
|
|
|
|
const G = curve[groupName];
|
|
const sG = G.F.n8*2;
|
|
const chunkSize = Math.floor((1<<20) / sG); // 128Mb chunks
|
|
let t = first;
|
|
for (let i=0 ; i<NPoints ; i+= chunkSize) {
|
|
if (logger) logger.debug(`applying key${sectionName}: ${i}/${NPoints}`);
|
|
const n= Math.min(NPoints-i, chunkSize );
|
|
const buffIn = await fdOld.read(n * sG);
|
|
const buffOutLEM = await G.batchApplyKey(buffIn, t, inc);
|
|
|
|
/* Code to test the case where we don't have the 2^m-2 component
|
|
if (sectionName== "tauG1") {
|
|
const bz = new Uint8Array(64);
|
|
buffOutLEM.set(bz, 64*((2 ** power) - 1 ));
|
|
}
|
|
*/
|
|
|
|
const promiseWrite = fdNew.write(buffOutLEM);
|
|
const buffOutC = await G.batchLEMtoC(buffOutLEM);
|
|
|
|
responseHasher.update(buffOutC);
|
|
await promiseWrite;
|
|
if (i==0) // Return the 2 first points.
|
|
for (let j=0; j<Math.min(2, NPoints); j++)
|
|
res.push(G.fromRprLEM(buffOutLEM, j*sG));
|
|
t = curve.Fr.mul(t, curve.Fr.exp(inc, n));
|
|
}
|
|
|
|
await binFileUtils__namespace.endWriteSection(fdNew);
|
|
|
|
return res;
|
|
}
|
|
|
|
|
|
async function hashSection(fdTo, groupName, sectionId, nPoints, sectionName, logger) {
|
|
|
|
const G = curve[groupName];
|
|
const sG = G.F.n8*2;
|
|
const nPointsChunk = Math.floor((1<<24)/sG);
|
|
|
|
const oldPos = fdTo.pos;
|
|
fdTo.pos = startSections[sectionId];
|
|
|
|
for (let i=0; i< nPoints; i += nPointsChunk) {
|
|
if (logger) logger.debug(`Hashing ${sectionName}: ${i}/${nPoints}`);
|
|
const n = Math.min(nPoints-i, nPointsChunk);
|
|
|
|
const buffLEM = await fdTo.read(n * sG);
|
|
|
|
const buffU = await G.batchLEMtoU(buffLEM);
|
|
|
|
nextChallengeHasher.update(buffU);
|
|
}
|
|
|
|
fdTo.pos = oldPos;
|
|
}
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function contribute(oldPtauFilename, newPTauFilename, name, entropy, logger) {
|
|
await Blake2b__default["default"].ready();
|
|
|
|
const {fd: fdOld, sections} = await binFileUtils__namespace.readBinFile(oldPtauFilename, "ptau", 1);
|
|
const {curve, power, ceremonyPower} = await readPTauHeader(fdOld, sections);
|
|
if (power != ceremonyPower) {
|
|
if (logger) logger.error("This file has been reduced. You cannot contribute into a reduced file.");
|
|
throw new Error("This file has been reduced. You cannot contribute into a reduced file.");
|
|
}
|
|
if (sections[12]) {
|
|
if (logger) logger.warn("WARNING: Contributing into a file that has phase2 calculated. You will have to prepare phase2 again.");
|
|
}
|
|
const contributions = await readContributions(fdOld, curve, sections);
|
|
const curContribution = {
|
|
name: name,
|
|
type: 0, // Beacon
|
|
};
|
|
|
|
let lastChallengeHash;
|
|
|
|
const rng = await getRandomRng(entropy);
|
|
|
|
if (contributions.length>0) {
|
|
lastChallengeHash = contributions[contributions.length-1].nextChallenge;
|
|
} else {
|
|
lastChallengeHash = calculateFirstChallengeHash(curve, power, logger);
|
|
}
|
|
|
|
// Generate a random key
|
|
|
|
|
|
curContribution.key = createPTauKey(curve, lastChallengeHash, rng);
|
|
|
|
|
|
const responseHasher = new Blake2b__default["default"](64);
|
|
responseHasher.update(lastChallengeHash);
|
|
|
|
const fdNew = await binFileUtils__namespace.createBinFile(newPTauFilename, "ptau", 1, 7);
|
|
await writePTauHeader(fdNew, curve, power);
|
|
|
|
const startSections = [];
|
|
|
|
let firstPoints;
|
|
firstPoints = await processSection(2, "G1", (2 ** power) * 2 -1, curve.Fr.e(1), curContribution.key.tau.prvKey, "tauG1" );
|
|
curContribution.tauG1 = firstPoints[1];
|
|
firstPoints = await processSection(3, "G2", (2 ** power) , curve.Fr.e(1), curContribution.key.tau.prvKey, "tauG2" );
|
|
curContribution.tauG2 = firstPoints[1];
|
|
firstPoints = await processSection(4, "G1", (2 ** power) , curContribution.key.alpha.prvKey, curContribution.key.tau.prvKey, "alphaTauG1" );
|
|
curContribution.alphaG1 = firstPoints[0];
|
|
firstPoints = await processSection(5, "G1", (2 ** power) , curContribution.key.beta.prvKey, curContribution.key.tau.prvKey, "betaTauG1" );
|
|
curContribution.betaG1 = firstPoints[0];
|
|
firstPoints = await processSection(6, "G2", 1, curContribution.key.beta.prvKey, curContribution.key.tau.prvKey, "betaTauG2" );
|
|
curContribution.betaG2 = firstPoints[0];
|
|
|
|
curContribution.partialHash = responseHasher.getPartialHash();
|
|
|
|
const buffKey = new Uint8Array(curve.F1.n8*2*6+curve.F2.n8*2*3);
|
|
|
|
toPtauPubKeyRpr(buffKey, 0, curve, curContribution.key, false);
|
|
|
|
responseHasher.update(new Uint8Array(buffKey));
|
|
const hashResponse = responseHasher.digest();
|
|
|
|
if (logger) logger.info(formatHash(hashResponse, "Contribution Response Hash imported: "));
|
|
|
|
const nextChallengeHasher = new Blake2b__default["default"](64);
|
|
nextChallengeHasher.update(hashResponse);
|
|
|
|
await hashSection(fdNew, "G1", 2, (2 ** power) * 2 -1, "tauG1");
|
|
await hashSection(fdNew, "G2", 3, (2 ** power) , "tauG2");
|
|
await hashSection(fdNew, "G1", 4, (2 ** power) , "alphaTauG1");
|
|
await hashSection(fdNew, "G1", 5, (2 ** power) , "betaTauG1");
|
|
await hashSection(fdNew, "G2", 6, 1 , "betaG2");
|
|
|
|
curContribution.nextChallenge = nextChallengeHasher.digest();
|
|
|
|
if (logger) logger.info(formatHash(curContribution.nextChallenge, "Next Challenge Hash: "));
|
|
|
|
contributions.push(curContribution);
|
|
|
|
await writeContributions(fdNew, curve, contributions);
|
|
|
|
await fdOld.close();
|
|
await fdNew.close();
|
|
|
|
return hashResponse;
|
|
|
|
async function processSection(sectionId, groupName, NPoints, first, inc, sectionName) {
|
|
const res = [];
|
|
fdOld.pos = sections[sectionId][0].p;
|
|
|
|
await binFileUtils__namespace.startWriteSection(fdNew, sectionId);
|
|
|
|
startSections[sectionId] = fdNew.pos;
|
|
|
|
const G = curve[groupName];
|
|
const sG = G.F.n8*2;
|
|
const chunkSize = Math.floor((1<<20) / sG); // 128Mb chunks
|
|
let t = first;
|
|
for (let i=0 ; i<NPoints ; i+= chunkSize) {
|
|
if (logger) logger.debug(`processing: ${sectionName}: ${i}/${NPoints}`);
|
|
const n= Math.min(NPoints-i, chunkSize );
|
|
const buffIn = await fdOld.read(n * sG);
|
|
const buffOutLEM = await G.batchApplyKey(buffIn, t, inc);
|
|
|
|
/* Code to test the case where we don't have the 2^m-2 component
|
|
if (sectionName== "tauG1") {
|
|
const bz = new Uint8Array(64);
|
|
buffOutLEM.set(bz, 64*((2 ** power) - 1 ));
|
|
}
|
|
*/
|
|
|
|
const promiseWrite = fdNew.write(buffOutLEM);
|
|
const buffOutC = await G.batchLEMtoC(buffOutLEM);
|
|
|
|
responseHasher.update(buffOutC);
|
|
await promiseWrite;
|
|
if (i==0) // Return the 2 first points.
|
|
for (let j=0; j<Math.min(2, NPoints); j++)
|
|
res.push(G.fromRprLEM(buffOutLEM, j*sG));
|
|
t = curve.Fr.mul(t, curve.Fr.exp(inc, n));
|
|
}
|
|
|
|
await binFileUtils__namespace.endWriteSection(fdNew);
|
|
|
|
return res;
|
|
}
|
|
|
|
|
|
async function hashSection(fdTo, groupName, sectionId, nPoints, sectionName) {
|
|
|
|
const G = curve[groupName];
|
|
const sG = G.F.n8*2;
|
|
const nPointsChunk = Math.floor((1<<24)/sG);
|
|
|
|
const oldPos = fdTo.pos;
|
|
fdTo.pos = startSections[sectionId];
|
|
|
|
for (let i=0; i< nPoints; i += nPointsChunk) {
|
|
if ((logger)&&i) logger.debug(`Hashing ${sectionName}: ` + i);
|
|
const n = Math.min(nPoints-i, nPointsChunk);
|
|
|
|
const buffLEM = await fdTo.read(n * sG);
|
|
|
|
const buffU = await G.batchLEMtoU(buffLEM);
|
|
|
|
nextChallengeHasher.update(buffU);
|
|
}
|
|
|
|
fdTo.pos = oldPos;
|
|
}
|
|
|
|
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function preparePhase2(oldPtauFilename, newPTauFilename, logger) {
|
|
|
|
const {fd: fdOld, sections} = await binFileUtils__namespace.readBinFile(oldPtauFilename, "ptau", 1);
|
|
const {curve, power} = await readPTauHeader(fdOld, sections);
|
|
|
|
const fdNew = await binFileUtils__namespace.createBinFile(newPTauFilename, "ptau", 1, 11);
|
|
await writePTauHeader(fdNew, curve, power);
|
|
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 2);
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 3);
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 4);
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 5);
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 6);
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 7);
|
|
|
|
await processSection(2, 12, "G1", "tauG1" );
|
|
await processSection(3, 13, "G2", "tauG2" );
|
|
await processSection(4, 14, "G1", "alphaTauG1" );
|
|
await processSection(5, 15, "G1", "betaTauG1" );
|
|
|
|
await fdOld.close();
|
|
await fdNew.close();
|
|
|
|
// await fs.promises.unlink(newPTauFilename+ ".tmp");
|
|
|
|
return;
|
|
|
|
async function processSection(oldSectionId, newSectionId, Gstr, sectionName) {
|
|
if (logger) logger.debug("Starting section: "+sectionName);
|
|
|
|
await binFileUtils__namespace.startWriteSection(fdNew, newSectionId);
|
|
|
|
for (let p=0; p<=power; p++) {
|
|
await processSectionPower(p);
|
|
}
|
|
|
|
if (oldSectionId == 2) {
|
|
await processSectionPower(power+1);
|
|
}
|
|
|
|
await binFileUtils__namespace.endWriteSection(fdNew);
|
|
|
|
|
|
async function processSectionPower(p) {
|
|
const nPoints = 2 ** p;
|
|
const G = curve[Gstr];
|
|
curve.Fr;
|
|
const sGin = G.F.n8*2;
|
|
G.F.n8*3;
|
|
|
|
let buff;
|
|
buff = new ffjavascript.BigBuffer(nPoints*sGin);
|
|
|
|
await binFileUtils__namespace.startReadUniqueSection(fdOld, sections, oldSectionId);
|
|
if ((oldSectionId == 2)&&(p==power+1)) {
|
|
await fdOld.readToBuffer(buff, 0,(nPoints-1)*sGin );
|
|
buff.set(curve.G1.zeroAffine, (nPoints-1)*sGin );
|
|
} else {
|
|
await fdOld.readToBuffer(buff, 0,nPoints*sGin );
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fdOld, true);
|
|
|
|
|
|
buff = await G.lagrangeEvaluations(buff, "affine", "affine", logger, sectionName);
|
|
await fdNew.write(buff);
|
|
|
|
/*
|
|
if (p <= curve.Fr.s) {
|
|
buff = await G.ifft(buff, "affine", "affine", logger, sectionName);
|
|
await fdNew.write(buff);
|
|
} else if (p == curve.Fr.s+1) {
|
|
const smallM = 1<<curve.Fr.s;
|
|
let t0 = new BigBuffer( smallM * sGmid );
|
|
let t1 = new BigBuffer( smallM * sGmid );
|
|
|
|
const shift_to_small_m = Fr.exp(Fr.shift, smallM);
|
|
const one_over_denom = Fr.inv(Fr.sub(shift_to_small_m, Fr.one));
|
|
|
|
let sInvAcc = Fr.one;
|
|
for (let i=0; i<smallM; i++) {
|
|
const ti = buff.slice(i*sGin, (i+1)*sGin);
|
|
const tmi = buff.slice((i+smallM)*sGin, (i+smallM+1)*sGin);
|
|
|
|
t0.set(
|
|
G.timesFr(
|
|
G.sub(
|
|
G.timesFr(ti , shift_to_small_m),
|
|
tmi
|
|
),
|
|
one_over_denom
|
|
),
|
|
i*sGmid
|
|
);
|
|
t1.set(
|
|
G.timesFr(
|
|
G.sub( tmi, ti),
|
|
Fr.mul(sInvAcc, one_over_denom)
|
|
),
|
|
i*sGmid
|
|
);
|
|
|
|
|
|
sInvAcc = Fr.mul(sInvAcc, Fr.shiftInv);
|
|
}
|
|
t0 = await G.ifft(t0, "jacobian", "affine", logger, sectionName + " t0");
|
|
await fdNew.write(t0);
|
|
t0 = null;
|
|
t1 = await G.ifft(t1, "jacobian", "affine", logger, sectionName + " t0");
|
|
await fdNew.write(t1);
|
|
|
|
} else {
|
|
if (logger) logger.error("Power too big");
|
|
throw new Error("Power to big");
|
|
}
|
|
*/
|
|
}
|
|
}
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function truncate(ptauFilename, template, logger) {
|
|
|
|
const {fd: fdOld, sections} = await binFileUtils__namespace.readBinFile(ptauFilename, "ptau", 1);
|
|
const {curve, power, ceremonyPower} = await readPTauHeader(fdOld, sections);
|
|
|
|
const sG1 = curve.G1.F.n8*2;
|
|
const sG2 = curve.G2.F.n8*2;
|
|
|
|
for (let p=1; p<power; p++) {
|
|
await generateTruncate(p);
|
|
}
|
|
|
|
await fdOld.close();
|
|
|
|
return true;
|
|
|
|
async function generateTruncate(p) {
|
|
|
|
let sP = p.toString();
|
|
while (sP.length<2) sP = "0" + sP;
|
|
|
|
if (logger) logger.debug("Writing Power: "+sP);
|
|
|
|
const fdNew = await binFileUtils__namespace.createBinFile(template + sP + ".ptau", "ptau", 1, 11);
|
|
await writePTauHeader(fdNew, curve, p, ceremonyPower);
|
|
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 2, ((2 ** p)*2-1) * sG1 ); // tagG1
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 3, (2 ** p) * sG2); // tauG2
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 4, (2 ** p) * sG1); // alfaTauG1
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 5, (2 ** p) * sG1); // betaTauG1
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 6, sG2); // betaTauG2
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 7); // contributions
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 12, ((2 ** (p+1))*2 -1) * sG1); // L_tauG1
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 13, ((2 ** p)*2 -1) * sG2); // L_tauG2
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 14, ((2 ** p)*2 -1) * sG1); // L_alfaTauG1
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 15, ((2 ** p)*2 -1) * sG1); // L_betaTauG1
|
|
|
|
await fdNew.close();
|
|
}
|
|
|
|
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function convert(oldPtauFilename, newPTauFilename, logger) {
|
|
|
|
const {fd: fdOld, sections} = await binFileUtils__namespace.readBinFile(oldPtauFilename, "ptau", 1);
|
|
const {curve, power} = await readPTauHeader(fdOld, sections);
|
|
|
|
const fdNew = await binFileUtils__namespace.createBinFile(newPTauFilename, "ptau", 1, 11);
|
|
await writePTauHeader(fdNew, curve, power);
|
|
|
|
// const fdTmp = await fastFile.createOverride(newPTauFilename+ ".tmp");
|
|
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 2);
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 3);
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 4);
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 5);
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 6);
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 7);
|
|
|
|
await processSection(2, 12, "G1", "tauG1" );
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 13);
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 14);
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 15);
|
|
|
|
await fdOld.close();
|
|
await fdNew.close();
|
|
|
|
// await fs.promises.unlink(newPTauFilename+ ".tmp");
|
|
|
|
return;
|
|
|
|
async function processSection(oldSectionId, newSectionId, Gstr, sectionName) {
|
|
if (logger) logger.debug("Starting section: "+sectionName);
|
|
|
|
await binFileUtils__namespace.startWriteSection(fdNew, newSectionId);
|
|
|
|
const size = sections[newSectionId][0].size;
|
|
const chunkSize = fdOld.pageSize;
|
|
await binFileUtils__namespace.startReadUniqueSection(fdOld, sections, newSectionId);
|
|
for (let p=0; p<size; p+=chunkSize) {
|
|
const l = Math.min(size -p, chunkSize);
|
|
const buff = await fdOld.read(l);
|
|
await fdNew.write(buff);
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fdOld);
|
|
|
|
if (oldSectionId == 2) {
|
|
await processSectionPower(power+1);
|
|
}
|
|
|
|
await binFileUtils__namespace.endWriteSection(fdNew);
|
|
|
|
async function processSectionPower(p) {
|
|
const nPoints = 2 ** p;
|
|
const G = curve[Gstr];
|
|
const sGin = G.F.n8*2;
|
|
|
|
let buff;
|
|
buff = new ffjavascript.BigBuffer(nPoints*sGin);
|
|
|
|
await binFileUtils__namespace.startReadUniqueSection(fdOld, sections, oldSectionId);
|
|
if ((oldSectionId == 2)&&(p==power+1)) {
|
|
await fdOld.readToBuffer(buff, 0,(nPoints-1)*sGin );
|
|
buff.set(curve.G1.zeroAffine, (nPoints-1)*sGin );
|
|
} else {
|
|
await fdOld.readToBuffer(buff, 0,nPoints*sGin );
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fdOld, true);
|
|
|
|
buff = await G.lagrangeEvaluations(buff, "affine", "affine", logger, sectionName);
|
|
await fdNew.write(buff);
|
|
|
|
/*
|
|
if (p <= curve.Fr.s) {
|
|
buff = await G.ifft(buff, "affine", "affine", logger, sectionName);
|
|
await fdNew.write(buff);
|
|
} else if (p == curve.Fr.s+1) {
|
|
const smallM = 1<<curve.Fr.s;
|
|
let t0 = new BigBuffer( smallM * sGmid );
|
|
let t1 = new BigBuffer( smallM * sGmid );
|
|
|
|
const shift_to_small_m = Fr.exp(Fr.shift, smallM);
|
|
const one_over_denom = Fr.inv(Fr.sub(shift_to_small_m, Fr.one));
|
|
|
|
let sInvAcc = Fr.one;
|
|
for (let i=0; i<smallM; i++) {
|
|
if (i%10000) logger.debug(`sectionName prepare L calc: ${sectionName}, ${i}/${smallM}`);
|
|
const ti = buff.slice(i*sGin, (i+1)*sGin);
|
|
const tmi = buff.slice((i+smallM)*sGin, (i+smallM+1)*sGin);
|
|
|
|
t0.set(
|
|
G.timesFr(
|
|
G.sub(
|
|
G.timesFr(ti , shift_to_small_m),
|
|
tmi
|
|
),
|
|
one_over_denom
|
|
),
|
|
i*sGmid
|
|
);
|
|
t1.set(
|
|
G.timesFr(
|
|
G.sub( tmi, ti),
|
|
Fr.mul(sInvAcc, one_over_denom)
|
|
),
|
|
i*sGmid
|
|
);
|
|
|
|
|
|
sInvAcc = Fr.mul(sInvAcc, Fr.shiftInv);
|
|
}
|
|
t0 = await G.ifft(t0, "jacobian", "affine", logger, sectionName + " t0");
|
|
await fdNew.write(t0);
|
|
t0 = null;
|
|
t1 = await G.ifft(t1, "jacobian", "affine", logger, sectionName + " t1");
|
|
await fdNew.write(t1);
|
|
|
|
} else {
|
|
if (logger) logger.error("Power too big");
|
|
throw new Error("Power to big");
|
|
}
|
|
*/
|
|
}
|
|
|
|
|
|
}
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function exportJson(pTauFilename, verbose) {
|
|
const {fd, sections} = await binFileUtils__namespace.readBinFile(pTauFilename, "ptau", 1);
|
|
|
|
const {curve, power} = await readPTauHeader(fd, sections);
|
|
|
|
const pTau = {};
|
|
pTau.q = curve.q;
|
|
pTau.power = power;
|
|
pTau.contributions = await readContributions(fd, curve, sections);
|
|
|
|
pTau.tauG1 = await exportSection(2, "G1", (2 ** power)*2 -1, "tauG1");
|
|
pTau.tauG2 = await exportSection(3, "G2", (2 ** power), "tauG2");
|
|
pTau.alphaTauG1 = await exportSection(4, "G1", (2 ** power), "alphaTauG1");
|
|
pTau.betaTauG1 = await exportSection(5, "G1", (2 ** power), "betaTauG1");
|
|
pTau.betaG2 = await exportSection(6, "G2", 1, "betaG2");
|
|
|
|
pTau.lTauG1 = await exportLagrange(12, "G1", "lTauG1");
|
|
pTau.lTauG2 = await exportLagrange(13, "G2", "lTauG2");
|
|
pTau.lAlphaTauG1 = await exportLagrange(14, "G1", "lAlphaTauG2");
|
|
pTau.lBetaTauG1 = await exportLagrange(15, "G1", "lBetaTauG2");
|
|
|
|
await fd.close();
|
|
|
|
return pTau;
|
|
|
|
|
|
|
|
async function exportSection(sectionId, groupName, nPoints, sectionName) {
|
|
const G = curve[groupName];
|
|
const sG = G.F.n8*2;
|
|
|
|
const res = [];
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, sectionId);
|
|
for (let i=0; i< nPoints; i++) {
|
|
if ((verbose)&&i&&(i%10000 == 0)) console.log(`${sectionName}: ` + i);
|
|
const buff = await fd.read(sG);
|
|
res.push(G.fromRprLEM(buff, 0));
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
return res;
|
|
}
|
|
|
|
async function exportLagrange(sectionId, groupName, sectionName) {
|
|
const G = curve[groupName];
|
|
const sG = G.F.n8*2;
|
|
|
|
const res = [];
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, sectionId);
|
|
for (let p=0; p<=power; p++) {
|
|
if (verbose) console.log(`${sectionName}: Power: ${p}`);
|
|
res[p] = [];
|
|
const nPoints = (2 ** p);
|
|
for (let i=0; i<nPoints; i++) {
|
|
if ((verbose)&&i&&(i%10000 == 0)) console.log(`${sectionName}: ${i}/${nPoints}`);
|
|
const buff = await fd.read(sG);
|
|
res[p].push(G.fromRprLEM(buff, 0));
|
|
}
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
return res;
|
|
}
|
|
|
|
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
var powersoftau = /*#__PURE__*/Object.freeze({
|
|
__proto__: null,
|
|
newAccumulator: newAccumulator,
|
|
exportChallenge: exportChallenge,
|
|
importResponse: importResponse,
|
|
verify: verify,
|
|
challengeContribute: challengeContribute,
|
|
beacon: beacon$1,
|
|
contribute: contribute,
|
|
preparePhase2: preparePhase2,
|
|
truncate: truncate,
|
|
convert: convert,
|
|
exportJson: exportJson
|
|
});
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
function r1csPrint(r1cs, syms, logger) {
|
|
for (let i=0; i<r1cs.constraints.length; i++) {
|
|
printCostraint(r1cs.constraints[i]);
|
|
}
|
|
function printCostraint(c) {
|
|
const lc2str = (lc) => {
|
|
let S = "";
|
|
const keys = Object.keys(lc);
|
|
keys.forEach( (k) => {
|
|
let name = syms.varIdx2Name[k];
|
|
if (name == "one") name = "";
|
|
|
|
let vs = r1cs.curve.Fr.toString(lc[k]);
|
|
if (vs == "1") vs = ""; // Do not show ones
|
|
if (vs == "-1") vs = "-"; // Do not show ones
|
|
if ((S!="")&&(vs[0]!="-")) vs = "+"+vs;
|
|
if (S!="") vs = " "+vs;
|
|
S= S + vs + name;
|
|
});
|
|
return S;
|
|
};
|
|
const S = `[ ${lc2str(c[0])} ] * [ ${lc2str(c[1])} ] - [ ${lc2str(c[2])} ] = 0`;
|
|
if (logger) logger.info(S);
|
|
}
|
|
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
const bls12381r = ffjavascript.Scalar.e("73eda753299d7d483339d80809a1d80553bda402fffe5bfeffffffff00000001", 16);
|
|
const bn128r = ffjavascript.Scalar.e("21888242871839275222246405745257275088548364400416034343698204186575808495617");
|
|
|
|
async function r1csInfo(r1csName, logger) {
|
|
|
|
const cir = await r1csfile.readR1cs(r1csName);
|
|
|
|
if (ffjavascript.Scalar.eq(cir.prime, bn128r)) {
|
|
if (logger) logger.info("Curve: bn-128");
|
|
} else if (ffjavascript.Scalar.eq(cir.prime, bls12381r)) {
|
|
if (logger) logger.info("Curve: bls12-381");
|
|
} else {
|
|
if (logger) logger.info(`Unknown Curve. Prime: ${ffjavascript.Scalar.toString(cir.prime)}`);
|
|
}
|
|
if (logger) logger.info(`# of Wires: ${cir.nVars}`);
|
|
if (logger) logger.info(`# of Constraints: ${cir.nConstraints}`);
|
|
if (logger) logger.info(`# of Private Inputs: ${cir.nPrvInputs}`);
|
|
if (logger) logger.info(`# of Public Inputs: ${cir.nPubInputs}`);
|
|
if (logger) logger.info(`# of Labels: ${cir.nLabels}`);
|
|
if (logger) logger.info(`# of Outputs: ${cir.nOutputs}`);
|
|
|
|
return cir;
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
function stringifyBigInts$2(Fr, o) {
|
|
if (o instanceof Uint8Array) {
|
|
return Fr.toString(o);
|
|
} else if (Array.isArray(o)) {
|
|
return o.map(stringifyBigInts$2.bind(null, Fr));
|
|
} else if (typeof o == "object") {
|
|
const res = {};
|
|
const keys = Object.keys(o);
|
|
keys.forEach( (k) => {
|
|
res[k] = stringifyBigInts$2(Fr, o[k]);
|
|
});
|
|
return res;
|
|
} else if ((typeof(o) == "bigint") || o.eq !== undefined) {
|
|
return o.toString(10);
|
|
} else {
|
|
return o;
|
|
}
|
|
}
|
|
|
|
|
|
async function r1csExportJson(r1csFileName, logger) {
|
|
|
|
const cir = await r1csfile.readR1cs(r1csFileName, true, true, true, logger);
|
|
const Fr=cir.curve.Fr;
|
|
delete cir.curve;
|
|
|
|
return stringifyBigInts$2(Fr, cir);
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
var r1cs = /*#__PURE__*/Object.freeze({
|
|
__proto__: null,
|
|
print: r1csPrint,
|
|
info: r1csInfo,
|
|
exportJson: r1csExportJson
|
|
});
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function loadSymbols(symFileName) {
|
|
const sym = {
|
|
labelIdx2Name: [ "one" ],
|
|
varIdx2Name: [ "one" ],
|
|
componentIdx2Name: []
|
|
};
|
|
const fd = await fastFile__namespace.readExisting(symFileName);
|
|
const buff = await fd.read(fd.totalSize);
|
|
const symsStr = new TextDecoder("utf-8").decode(buff);
|
|
const lines = symsStr.split("\n");
|
|
for (let i=0; i<lines.length; i++) {
|
|
const arr = lines[i].split(",");
|
|
if (arr.length!=4) continue;
|
|
if (sym.varIdx2Name[arr[1]]) {
|
|
sym.varIdx2Name[arr[1]] += "|" + arr[3];
|
|
} else {
|
|
sym.varIdx2Name[arr[1]] = arr[3];
|
|
}
|
|
sym.labelIdx2Name[arr[0]] = arr[3];
|
|
if (!sym.componentIdx2Name[arr[2]]) {
|
|
sym.componentIdx2Name[arr[2]] = extractComponent(arr[3]);
|
|
}
|
|
}
|
|
|
|
await fd.close();
|
|
|
|
return sym;
|
|
|
|
function extractComponent(name) {
|
|
const arr = name.split(".");
|
|
arr.pop(); // Remove the lasr element
|
|
return arr.join(".");
|
|
}
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function wtnsDebug(input, wasmFileName, wtnsFileName, symName, options, logger) {
|
|
|
|
const fdWasm = await fastFile__namespace.readExisting(wasmFileName);
|
|
const wasm = await fdWasm.read(fdWasm.totalSize);
|
|
await fdWasm.close();
|
|
|
|
|
|
let wcOps = {
|
|
sanityCheck: true
|
|
};
|
|
let sym = await loadSymbols(symName);
|
|
if (options.set) {
|
|
if (!sym) sym = await loadSymbols(symName);
|
|
wcOps.logSetSignal= function(labelIdx, value) {
|
|
if (logger) logger.info("SET " + sym.labelIdx2Name[labelIdx] + " <-- " + value.toString());
|
|
};
|
|
}
|
|
if (options.get) {
|
|
if (!sym) sym = await loadSymbols(symName);
|
|
wcOps.logGetSignal= function(varIdx, value) {
|
|
if (logger) logger.info("GET " + sym.labelIdx2Name[varIdx] + " --> " + value.toString());
|
|
};
|
|
}
|
|
if (options.trigger) {
|
|
if (!sym) sym = await loadSymbols(symName);
|
|
wcOps.logStartComponent= function(cIdx) {
|
|
if (logger) logger.info("START: " + sym.componentIdx2Name[cIdx]);
|
|
};
|
|
wcOps.logFinishComponent= function(cIdx) {
|
|
if (logger) logger.info("FINISH: " + sym.componentIdx2Name[cIdx]);
|
|
};
|
|
}
|
|
wcOps.sym = sym;
|
|
|
|
const wc = await circom_runtime.WitnessCalculatorBuilder(wasm, wcOps);
|
|
const w = await wc.calculateWitness(input);
|
|
|
|
const fdWtns = await binFileUtils__namespace.createBinFile(wtnsFileName, "wtns", 2, 2);
|
|
|
|
await write(fdWtns, w, wc.prime);
|
|
|
|
await fdWtns.close();
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function wtnsExportJson(wtnsFileName) {
|
|
|
|
const w = await read(wtnsFileName);
|
|
|
|
return w;
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
var wtns = /*#__PURE__*/Object.freeze({
|
|
__proto__: null,
|
|
calculate: wtnsCalculate,
|
|
debug: wtnsDebug,
|
|
exportJson: wtnsExportJson
|
|
});
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
const SUBARRAY_SIZE = 0x40000;
|
|
|
|
const BigArrayHandler = {
|
|
get: function(obj, prop) {
|
|
if (!isNaN(prop)) {
|
|
return obj.getElement(prop);
|
|
} else return obj[prop];
|
|
},
|
|
set: function(obj, prop, value) {
|
|
if (!isNaN(prop)) {
|
|
return obj.setElement(prop, value);
|
|
} else {
|
|
obj[prop] = value;
|
|
return true;
|
|
}
|
|
}
|
|
};
|
|
|
|
class _BigArray {
|
|
constructor (initSize) {
|
|
this.length = initSize || 0;
|
|
this.arr = new Array(SUBARRAY_SIZE);
|
|
|
|
for (let i=0; i<initSize; i+=SUBARRAY_SIZE) {
|
|
this.arr[i/SUBARRAY_SIZE] = new Array(Math.min(SUBARRAY_SIZE, initSize - i));
|
|
}
|
|
return this;
|
|
}
|
|
push () {
|
|
for (let i=0; i<arguments.length; i++) {
|
|
this.setElement (this.length, arguments[i]);
|
|
}
|
|
}
|
|
|
|
slice (f, t) {
|
|
const arr = new Array(t-f);
|
|
for (let i=f; i< t; i++) arr[i-f] = this.getElement(i);
|
|
return arr;
|
|
}
|
|
getElement(idx) {
|
|
idx = parseInt(idx);
|
|
const idx1 = Math.floor(idx / SUBARRAY_SIZE);
|
|
const idx2 = idx % SUBARRAY_SIZE;
|
|
return this.arr[idx1] ? this.arr[idx1][idx2] : undefined;
|
|
}
|
|
setElement(idx, value) {
|
|
idx = parseInt(idx);
|
|
const idx1 = Math.floor(idx / SUBARRAY_SIZE);
|
|
if (!this.arr[idx1]) {
|
|
this.arr[idx1] = new Array(SUBARRAY_SIZE);
|
|
}
|
|
const idx2 = idx % SUBARRAY_SIZE;
|
|
this.arr[idx1][idx2] = value;
|
|
if (idx >= this.length) this.length = idx+1;
|
|
return true;
|
|
}
|
|
getKeys() {
|
|
const newA = new BigArray();
|
|
for (let i=0; i<this.arr.length; i++) {
|
|
if (this.arr[i]) {
|
|
for (let j=0; j<this.arr[i].length; j++) {
|
|
if (typeof this.arr[i][j] !== "undefined") {
|
|
newA.push(i*SUBARRAY_SIZE+j);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return newA;
|
|
}
|
|
}
|
|
|
|
class BigArray {
|
|
constructor( initSize ) {
|
|
const obj = new _BigArray(initSize);
|
|
const extObj = new Proxy(obj, BigArrayHandler);
|
|
return extObj;
|
|
}
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
|
|
async function newZKey(r1csName, ptauName, zkeyName, logger) {
|
|
|
|
const TAU_G1 = 0;
|
|
const TAU_G2 = 1;
|
|
const ALPHATAU_G1 = 2;
|
|
const BETATAU_G1 = 3;
|
|
await Blake2b__default["default"].ready();
|
|
const csHasher = Blake2b__default["default"](64);
|
|
|
|
const {fd: fdPTau, sections: sectionsPTau} = await binFileUtils.readBinFile(ptauName, "ptau", 1, 1<<22, 1<<24);
|
|
const {curve, power} = await readPTauHeader(fdPTau, sectionsPTau);
|
|
const {fd: fdR1cs, sections: sectionsR1cs} = await binFileUtils.readBinFile(r1csName, "r1cs", 1, 1<<22, 1<<24);
|
|
const r1cs = await r1csfile.readR1csHeader(fdR1cs, sectionsR1cs, false);
|
|
|
|
const fdZKey = await binFileUtils.createBinFile(zkeyName, "zkey", 1, 10, 1<<22, 1<<24);
|
|
|
|
const sG1 = curve.G1.F.n8*2;
|
|
const sG2 = curve.G2.F.n8*2;
|
|
|
|
if (r1cs.prime != curve.r) {
|
|
if (logger) logger.error("r1cs curve does not match powers of tau ceremony curve");
|
|
return -1;
|
|
}
|
|
|
|
const cirPower = log2(r1cs.nConstraints + r1cs.nPubInputs + r1cs.nOutputs +1 -1) +1;
|
|
|
|
if (cirPower > power) {
|
|
if (logger) logger.error(`circuit too big for this power of tau ceremony. ${r1cs.nConstraints}*2 > 2**${power}`);
|
|
return -1;
|
|
}
|
|
|
|
if (!sectionsPTau[12]) {
|
|
if (logger) logger.error("Powers of tau is not prepared.");
|
|
return -1;
|
|
}
|
|
|
|
const nPublic = r1cs.nOutputs + r1cs.nPubInputs;
|
|
const domainSize = 2 ** cirPower;
|
|
|
|
// Write the header
|
|
///////////
|
|
await binFileUtils.startWriteSection(fdZKey, 1);
|
|
await fdZKey.writeULE32(1); // Groth
|
|
await binFileUtils.endWriteSection(fdZKey);
|
|
|
|
// Write the Groth header section
|
|
///////////
|
|
|
|
await binFileUtils.startWriteSection(fdZKey, 2);
|
|
const primeQ = curve.q;
|
|
const n8q = (Math.floor( (ffjavascript.Scalar.bitLength(primeQ) - 1) / 64) +1)*8;
|
|
|
|
const primeR = curve.r;
|
|
const n8r = (Math.floor( (ffjavascript.Scalar.bitLength(primeR) - 1) / 64) +1)*8;
|
|
const Rr = ffjavascript.Scalar.mod(ffjavascript.Scalar.shl(1, n8r*8), primeR);
|
|
const R2r = curve.Fr.e(ffjavascript.Scalar.mod(ffjavascript.Scalar.mul(Rr,Rr), primeR));
|
|
|
|
await fdZKey.writeULE32(n8q);
|
|
await binFileUtils.writeBigInt(fdZKey, primeQ, n8q);
|
|
await fdZKey.writeULE32(n8r);
|
|
await binFileUtils.writeBigInt(fdZKey, primeR, n8r);
|
|
await fdZKey.writeULE32(r1cs.nVars); // Total number of bars
|
|
await fdZKey.writeULE32(nPublic); // Total number of public vars (not including ONE)
|
|
await fdZKey.writeULE32(domainSize); // domainSize
|
|
|
|
let bAlpha1;
|
|
bAlpha1 = await fdPTau.read(sG1, sectionsPTau[4][0].p);
|
|
await fdZKey.write(bAlpha1);
|
|
bAlpha1 = await curve.G1.batchLEMtoU(bAlpha1);
|
|
csHasher.update(bAlpha1);
|
|
|
|
let bBeta1;
|
|
bBeta1 = await fdPTau.read(sG1, sectionsPTau[5][0].p);
|
|
await fdZKey.write(bBeta1);
|
|
bBeta1 = await curve.G1.batchLEMtoU(bBeta1);
|
|
csHasher.update(bBeta1);
|
|
|
|
let bBeta2;
|
|
bBeta2 = await fdPTau.read(sG2, sectionsPTau[6][0].p);
|
|
await fdZKey.write(bBeta2);
|
|
bBeta2 = await curve.G2.batchLEMtoU(bBeta2);
|
|
csHasher.update(bBeta2);
|
|
|
|
const bg1 = new Uint8Array(sG1);
|
|
curve.G1.toRprLEM(bg1, 0, curve.G1.g);
|
|
const bg2 = new Uint8Array(sG2);
|
|
curve.G2.toRprLEM(bg2, 0, curve.G2.g);
|
|
const bg1U = new Uint8Array(sG1);
|
|
curve.G1.toRprUncompressed(bg1U, 0, curve.G1.g);
|
|
const bg2U = new Uint8Array(sG2);
|
|
curve.G2.toRprUncompressed(bg2U, 0, curve.G2.g);
|
|
|
|
await fdZKey.write(bg2); // gamma2
|
|
await fdZKey.write(bg1); // delta1
|
|
await fdZKey.write(bg2); // delta2
|
|
csHasher.update(bg2U); // gamma2
|
|
csHasher.update(bg1U); // delta1
|
|
csHasher.update(bg2U); // delta2
|
|
await binFileUtils.endWriteSection(fdZKey);
|
|
|
|
if (logger) logger.info("Reading r1cs");
|
|
let sR1cs = await binFileUtils.readSection(fdR1cs, sectionsR1cs, 2);
|
|
|
|
const A = new BigArray(r1cs.nVars);
|
|
const B1 = new BigArray(r1cs.nVars);
|
|
const B2 = new BigArray(r1cs.nVars);
|
|
const C = new BigArray(r1cs.nVars- nPublic -1);
|
|
const IC = new Array(nPublic+1);
|
|
|
|
if (logger) logger.info("Reading tauG1");
|
|
let sTauG1 = await binFileUtils.readSection(fdPTau, sectionsPTau, 12, (domainSize -1)*sG1, domainSize*sG1);
|
|
if (logger) logger.info("Reading tauG2");
|
|
let sTauG2 = await binFileUtils.readSection(fdPTau, sectionsPTau, 13, (domainSize -1)*sG2, domainSize*sG2);
|
|
if (logger) logger.info("Reading alphatauG1");
|
|
let sAlphaTauG1 = await binFileUtils.readSection(fdPTau, sectionsPTau, 14, (domainSize -1)*sG1, domainSize*sG1);
|
|
if (logger) logger.info("Reading betatauG1");
|
|
let sBetaTauG1 = await binFileUtils.readSection(fdPTau, sectionsPTau, 15, (domainSize -1)*sG1, domainSize*sG1);
|
|
|
|
await processConstraints();
|
|
|
|
await composeAndWritePoints(3, "G1", IC, "IC");
|
|
|
|
await writeHs();
|
|
|
|
await hashHPoints();
|
|
|
|
await composeAndWritePoints(8, "G1", C, "C");
|
|
await composeAndWritePoints(5, "G1", A, "A");
|
|
await composeAndWritePoints(6, "G1", B1, "B1");
|
|
await composeAndWritePoints(7, "G2", B2, "B2");
|
|
|
|
const csHash = csHasher.digest();
|
|
// Contributions section
|
|
await binFileUtils.startWriteSection(fdZKey, 10);
|
|
await fdZKey.write(csHash);
|
|
await fdZKey.writeULE32(0);
|
|
await binFileUtils.endWriteSection(fdZKey);
|
|
|
|
if (logger) logger.info(formatHash(csHash, "Circuit hash: "));
|
|
|
|
|
|
await fdZKey.close();
|
|
await fdR1cs.close();
|
|
await fdPTau.close();
|
|
|
|
return csHash;
|
|
|
|
async function writeHs() {
|
|
await binFileUtils.startWriteSection(fdZKey, 9);
|
|
const buffOut = new ffjavascript.BigBuffer(domainSize*sG1);
|
|
if (cirPower < curve.Fr.s) {
|
|
let sTauG1 = await binFileUtils.readSection(fdPTau, sectionsPTau, 12, (domainSize*2-1)*sG1, domainSize*2*sG1);
|
|
for (let i=0; i< domainSize; i++) {
|
|
if ((logger)&&(i%10000 == 0)) logger.debug(`spliting buffer: ${i}/${domainSize}`);
|
|
const buff = sTauG1.slice( (i*2+1)*sG1, (i*2+1)*sG1 + sG1 );
|
|
buffOut.set(buff, i*sG1);
|
|
}
|
|
} else if (cirPower == curve.Fr.s) {
|
|
const o = sectionsPTau[12][0].p + ((2 ** (cirPower+1)) -1)*sG1;
|
|
await fdPTau.readToBuffer(buffOut, 0, domainSize*sG1, o + domainSize*sG1);
|
|
} else {
|
|
if (logger) logger.error("Circuit too big");
|
|
throw new Error("Circuit too big for this curve");
|
|
}
|
|
await fdZKey.write(buffOut);
|
|
await binFileUtils.endWriteSection(fdZKey);
|
|
}
|
|
|
|
async function processConstraints() {
|
|
const buffCoeff = new Uint8Array(12 + curve.Fr.n8);
|
|
const buffCoeffV = new DataView(buffCoeff.buffer);
|
|
const bOne = new Uint8Array(curve.Fr.n8);
|
|
curve.Fr.toRprLE(bOne, 0, curve.Fr.e(1));
|
|
|
|
let r1csPos = 0;
|
|
|
|
function r1cs_readULE32() {
|
|
const buff = sR1cs.slice(r1csPos, r1csPos+4);
|
|
r1csPos += 4;
|
|
const buffV = new DataView(buff.buffer);
|
|
return buffV.getUint32(0, true);
|
|
}
|
|
|
|
const coefs = new BigArray();
|
|
for (let c=0; c<r1cs.nConstraints; c++) {
|
|
if ((logger)&&(c%10000 == 0)) logger.debug(`processing constraints: ${c}/${r1cs.nConstraints}`);
|
|
const nA = r1cs_readULE32();
|
|
for (let i=0; i<nA; i++) {
|
|
const s = r1cs_readULE32();
|
|
const coefp = r1csPos;
|
|
r1csPos += curve.Fr.n8;
|
|
|
|
const l1t = TAU_G1;
|
|
const l1 = sG1*c;
|
|
const l2t = BETATAU_G1;
|
|
const l2 = sG1*c;
|
|
if (typeof A[s] === "undefined") A[s] = [];
|
|
A[s].push([l1t, l1, coefp]);
|
|
|
|
if (s <= nPublic) {
|
|
if (typeof IC[s] === "undefined") IC[s] = [];
|
|
IC[s].push([l2t, l2, coefp]);
|
|
} else {
|
|
if (typeof C[s- nPublic -1] === "undefined") C[s- nPublic -1] = [];
|
|
C[s - nPublic -1].push([l2t, l2, coefp]);
|
|
}
|
|
coefs.push([0, c, s, coefp]);
|
|
}
|
|
|
|
const nB = r1cs_readULE32();
|
|
for (let i=0; i<nB; i++) {
|
|
const s = r1cs_readULE32();
|
|
const coefp = r1csPos;
|
|
r1csPos += curve.Fr.n8;
|
|
|
|
const l1t = TAU_G1;
|
|
const l1 = sG1*c;
|
|
const l2t = TAU_G2;
|
|
const l2 = sG2*c;
|
|
const l3t = ALPHATAU_G1;
|
|
const l3 = sG1*c;
|
|
if (typeof B1[s] === "undefined") B1[s] = [];
|
|
B1[s].push([l1t, l1, coefp]);
|
|
if (typeof B2[s] === "undefined") B2[s] = [];
|
|
B2[s].push([l2t, l2, coefp]);
|
|
|
|
if (s <= nPublic) {
|
|
if (typeof IC[s] === "undefined") IC[s] = [];
|
|
IC[s].push([l3t, l3, coefp]);
|
|
} else {
|
|
if (typeof C[s- nPublic -1] === "undefined") C[s- nPublic -1] = [];
|
|
C[s- nPublic -1].push([l3t, l3, coefp]);
|
|
}
|
|
|
|
coefs.push([1, c, s, coefp]);
|
|
}
|
|
|
|
const nC = r1cs_readULE32();
|
|
for (let i=0; i<nC; i++) {
|
|
const s = r1cs_readULE32();
|
|
const coefp = r1csPos;
|
|
r1csPos += curve.Fr.n8;
|
|
|
|
const l1t = TAU_G1;
|
|
const l1 = sG1*c;
|
|
if (s <= nPublic) {
|
|
if (typeof IC[s] === "undefined") IC[s] = [];
|
|
IC[s].push([l1t, l1, coefp]);
|
|
} else {
|
|
if (typeof C[s- nPublic -1] === "undefined") C[s- nPublic -1] = [];
|
|
C[s- nPublic -1].push([l1t, l1, coefp]);
|
|
}
|
|
}
|
|
}
|
|
|
|
for (let s = 0; s <= nPublic ; s++) {
|
|
const l1t = TAU_G1;
|
|
const l1 = sG1*(r1cs.nConstraints + s);
|
|
const l2t = BETATAU_G1;
|
|
const l2 = sG1*(r1cs.nConstraints + s);
|
|
if (typeof A[s] === "undefined") A[s] = [];
|
|
A[s].push([l1t, l1, -1]);
|
|
if (typeof IC[s] === "undefined") IC[s] = [];
|
|
IC[s].push([l2t, l2, -1]);
|
|
coefs.push([0, r1cs.nConstraints + s, s, -1]);
|
|
}
|
|
|
|
|
|
await binFileUtils.startWriteSection(fdZKey, 4);
|
|
|
|
const buffSection = new ffjavascript.BigBuffer(coefs.length*(12+curve.Fr.n8) + 4);
|
|
|
|
const buff4 = new Uint8Array(4);
|
|
const buff4V = new DataView(buff4.buffer);
|
|
buff4V.setUint32(0, coefs.length, true);
|
|
buffSection.set(buff4);
|
|
let coefsPos = 4;
|
|
for (let i=0; i<coefs.length; i++) {
|
|
if ((logger)&&(i%100000 == 0)) logger.debug(`writing coeffs: ${i}/${coefs.length}`);
|
|
writeCoef(coefs[i]);
|
|
}
|
|
|
|
await fdZKey.write(buffSection);
|
|
await binFileUtils.endWriteSection(fdZKey);
|
|
|
|
function writeCoef(c) {
|
|
buffCoeffV.setUint32(0, c[0], true);
|
|
buffCoeffV.setUint32(4, c[1], true);
|
|
buffCoeffV.setUint32(8, c[2], true);
|
|
let n;
|
|
if (c[3]>=0) {
|
|
n = curve.Fr.fromRprLE(sR1cs.slice(c[3], c[3] + curve.Fr.n8), 0);
|
|
} else {
|
|
n = curve.Fr.fromRprLE(bOne, 0);
|
|
}
|
|
const nR2 = curve.Fr.mul(n, R2r);
|
|
curve.Fr.toRprLE(buffCoeff, 12, nR2);
|
|
buffSection.set(buffCoeff, coefsPos);
|
|
coefsPos += buffCoeff.length;
|
|
}
|
|
|
|
}
|
|
|
|
async function composeAndWritePoints(idSection, groupName, arr, sectionName) {
|
|
const CHUNK_SIZE= 1<<15;
|
|
const G = curve[groupName];
|
|
|
|
hashU32(arr.length);
|
|
await binFileUtils.startWriteSection(fdZKey, idSection);
|
|
|
|
let opPromises = [];
|
|
|
|
let i=0;
|
|
while (i<arr.length) {
|
|
|
|
let t=0;
|
|
while ((i<arr.length)&&(t<curve.tm.concurrency)) {
|
|
if (logger) logger.debug(`Writing points start ${sectionName}: ${i}/${arr.length}`);
|
|
let n = 1;
|
|
let nP = (arr[i] ? arr[i].length : 0);
|
|
while ((i + n < arr.length) && (nP + (arr[i+n] ? arr[i+n].length : 0) < CHUNK_SIZE) && (n<CHUNK_SIZE)) {
|
|
nP += (arr[i+n] ? arr[i+n].length : 0);
|
|
n ++;
|
|
}
|
|
const subArr = arr.slice(i, i + n);
|
|
const _i = i;
|
|
opPromises.push(composeAndWritePointsThread(groupName, subArr, logger, sectionName).then( (r) => {
|
|
if (logger) logger.debug(`Writing points end ${sectionName}: ${_i}/${arr.length}`);
|
|
return r;
|
|
}));
|
|
i += n;
|
|
t++;
|
|
}
|
|
|
|
const result = await Promise.all(opPromises);
|
|
|
|
for (let k=0; k<result.length; k++) {
|
|
await fdZKey.write(result[k][0]);
|
|
const buff = await G.batchLEMtoU(result[k][0]);
|
|
csHasher.update(buff);
|
|
}
|
|
opPromises = [];
|
|
|
|
}
|
|
await binFileUtils.endWriteSection(fdZKey);
|
|
|
|
}
|
|
|
|
async function composeAndWritePointsThread(groupName, arr, logger, sectionName) {
|
|
const G = curve[groupName];
|
|
const sGin = G.F.n8*2;
|
|
const sGmid = G.F.n8*3;
|
|
const sGout = G.F.n8*2;
|
|
let fnExp, fnMultiExp, fnBatchToAffine, fnZero;
|
|
if (groupName == "G1") {
|
|
fnExp = "g1m_timesScalarAffine";
|
|
fnMultiExp = "g1m_multiexpAffine";
|
|
fnBatchToAffine = "g1m_batchToAffine";
|
|
fnZero = "g1m_zero";
|
|
} else if (groupName == "G2") {
|
|
fnExp = "g2m_timesScalarAffine";
|
|
fnMultiExp = "g2m_multiexpAffine";
|
|
fnBatchToAffine = "g2m_batchToAffine";
|
|
fnZero = "g2m_zero";
|
|
} else {
|
|
throw new Error("Invalid group");
|
|
}
|
|
let acc =0;
|
|
for (let i=0; i<arr.length; i++) acc += arr[i] ? arr[i].length : 0;
|
|
let bBases, bScalars;
|
|
if (acc> 2<<14) {
|
|
bBases = new ffjavascript.BigBuffer(acc*sGin);
|
|
bScalars = new ffjavascript.BigBuffer(acc*curve.Fr.n8);
|
|
} else {
|
|
bBases = new Uint8Array(acc*sGin);
|
|
bScalars = new Uint8Array(acc*curve.Fr.n8);
|
|
}
|
|
let pB =0;
|
|
let pS =0;
|
|
|
|
const sBuffs = [
|
|
sTauG1,
|
|
sTauG2,
|
|
sAlphaTauG1,
|
|
sBetaTauG1
|
|
];
|
|
|
|
const bOne = new Uint8Array(curve.Fr.n8);
|
|
curve.Fr.toRprLE(bOne, 0, curve.Fr.e(1));
|
|
|
|
let offset = 0;
|
|
for (let i=0; i<arr.length; i++) {
|
|
if (!arr[i]) continue;
|
|
for (let j=0; j<arr[i].length; j++) {
|
|
if ((logger)&&(j)&&(j%10000 == 0)) logger.debug(`Configuring big array ${sectionName}: ${j}/${arr[i].length}`);
|
|
bBases.set(
|
|
sBuffs[arr[i][j][0]].slice(
|
|
arr[i][j][1],
|
|
arr[i][j][1] + sGin
|
|
), offset*sGin
|
|
);
|
|
if (arr[i][j][2]>=0) {
|
|
bScalars.set(
|
|
sR1cs.slice(
|
|
arr[i][j][2],
|
|
arr[i][j][2] + curve.Fr.n8
|
|
),
|
|
offset*curve.Fr.n8
|
|
);
|
|
} else {
|
|
bScalars.set(bOne, offset*curve.Fr.n8);
|
|
}
|
|
offset ++;
|
|
}
|
|
}
|
|
|
|
if (arr.length>1) {
|
|
const task = [];
|
|
task.push({cmd: "ALLOCSET", var: 0, buff: bBases});
|
|
task.push({cmd: "ALLOCSET", var: 1, buff: bScalars});
|
|
task.push({cmd: "ALLOC", var: 2, len: arr.length*sGmid});
|
|
pB = 0;
|
|
pS = 0;
|
|
let pD =0;
|
|
for (let i=0; i<arr.length; i++) {
|
|
if (!arr[i]) {
|
|
task.push({cmd: "CALL", fnName: fnZero, params: [
|
|
{var: 2, offset: pD}
|
|
]});
|
|
pD += sGmid;
|
|
continue;
|
|
}
|
|
if (arr[i].length == 1) {
|
|
task.push({cmd: "CALL", fnName: fnExp, params: [
|
|
{var: 0, offset: pB},
|
|
{var: 1, offset: pS},
|
|
{val: curve.Fr.n8},
|
|
{var: 2, offset: pD}
|
|
]});
|
|
} else {
|
|
task.push({cmd: "CALL", fnName: fnMultiExp, params: [
|
|
{var: 0, offset: pB},
|
|
{var: 1, offset: pS},
|
|
{val: curve.Fr.n8},
|
|
{val: arr[i].length},
|
|
{var: 2, offset: pD}
|
|
]});
|
|
}
|
|
pB += sGin*arr[i].length;
|
|
pS += curve.Fr.n8*arr[i].length;
|
|
pD += sGmid;
|
|
}
|
|
task.push({cmd: "CALL", fnName: fnBatchToAffine, params: [
|
|
{var: 2},
|
|
{val: arr.length},
|
|
{var: 2},
|
|
]});
|
|
task.push({cmd: "GET", out: 0, var: 2, len: arr.length*sGout});
|
|
|
|
const res = await curve.tm.queueAction(task);
|
|
return res;
|
|
} else {
|
|
let res = await G.multiExpAffine(bBases, bScalars, logger, sectionName);
|
|
res = [ G.toAffine(res) ];
|
|
return res;
|
|
}
|
|
}
|
|
|
|
|
|
async function hashHPoints() {
|
|
const CHUNK_SIZE = 1<<14;
|
|
|
|
hashU32(domainSize-1);
|
|
|
|
for (let i=0; i<domainSize-1; i+= CHUNK_SIZE) {
|
|
if (logger) logger.debug(`HashingHPoints: ${i}/${domainSize}`);
|
|
const n = Math.min(domainSize-1, CHUNK_SIZE);
|
|
await hashHPointsChunk(i, n);
|
|
}
|
|
}
|
|
|
|
async function hashHPointsChunk(offset, nPoints) {
|
|
const buff1 = await fdPTau.read(nPoints *sG1, sectionsPTau[2][0].p + (offset + domainSize)*sG1);
|
|
const buff2 = await fdPTau.read(nPoints *sG1, sectionsPTau[2][0].p + offset*sG1);
|
|
const concurrency= curve.tm.concurrency;
|
|
const nPointsPerThread = Math.floor(nPoints / concurrency);
|
|
const opPromises = [];
|
|
for (let i=0; i<concurrency; i++) {
|
|
let n;
|
|
if (i< concurrency-1) {
|
|
n = nPointsPerThread;
|
|
} else {
|
|
n = nPoints - i*nPointsPerThread;
|
|
}
|
|
if (n==0) continue;
|
|
|
|
const subBuff1 = buff1.slice(i*nPointsPerThread*sG1, (i*nPointsPerThread+n)*sG1);
|
|
const subBuff2 = buff2.slice(i*nPointsPerThread*sG1, (i*nPointsPerThread+n)*sG1);
|
|
opPromises.push(hashHPointsThread(subBuff1, subBuff2));
|
|
}
|
|
|
|
|
|
const result = await Promise.all(opPromises);
|
|
|
|
for (let i=0; i<result.length; i++) {
|
|
csHasher.update(result[i][0]);
|
|
}
|
|
}
|
|
|
|
async function hashHPointsThread(buff1, buff2) {
|
|
const nPoints = buff1.byteLength/sG1;
|
|
const sGmid = curve.G1.F.n8*3;
|
|
const task = [];
|
|
task.push({cmd: "ALLOCSET", var: 0, buff: buff1});
|
|
task.push({cmd: "ALLOCSET", var: 1, buff: buff2});
|
|
task.push({cmd: "ALLOC", var: 2, len: nPoints*sGmid});
|
|
for (let i=0; i<nPoints; i++) {
|
|
task.push({
|
|
cmd: "CALL",
|
|
fnName: "g1m_subAffine",
|
|
params: [
|
|
{var: 0, offset: i*sG1},
|
|
{var: 1, offset: i*sG1},
|
|
{var: 2, offset: i*sGmid},
|
|
]
|
|
});
|
|
}
|
|
task.push({cmd: "CALL", fnName: "g1m_batchToAffine", params: [
|
|
{var: 2},
|
|
{val: nPoints},
|
|
{var: 2},
|
|
]});
|
|
task.push({cmd: "CALL", fnName: "g1m_batchLEMtoU", params: [
|
|
{var: 2},
|
|
{val: nPoints},
|
|
{var: 2},
|
|
]});
|
|
task.push({cmd: "GET", out: 0, var: 2, len: nPoints*sG1});
|
|
|
|
const res = await curve.tm.queueAction(task);
|
|
|
|
return res;
|
|
}
|
|
|
|
function hashU32(n) {
|
|
const buff = new Uint8Array(4);
|
|
const buffV = new DataView(buff.buffer, buff.byteOffset, buff.byteLength);
|
|
buffV.setUint32(0, n, false);
|
|
csHasher.update(buff);
|
|
}
|
|
|
|
}
|
|
|
|
async function phase2exportMPCParams(zkeyName, mpcparamsName, logger) {
|
|
|
|
const {fd: fdZKey, sections: sectionsZKey} = await binFileUtils__namespace.readBinFile(zkeyName, "zkey", 2);
|
|
const zkey = await readHeader$1(fdZKey, sectionsZKey);
|
|
if (zkey.protocol != "groth16") {
|
|
throw new Error("zkey file is not groth16");
|
|
}
|
|
|
|
const curve = await getCurveFromQ(zkey.q);
|
|
const sG1 = curve.G1.F.n8*2;
|
|
const sG2 = curve.G2.F.n8*2;
|
|
|
|
const mpcParams = await readMPCParams(fdZKey, curve, sectionsZKey);
|
|
|
|
const fdMPCParams = await fastFile__namespace.createOverride(mpcparamsName);
|
|
|
|
/////////////////////
|
|
// Verification Key Section
|
|
/////////////////////
|
|
await writeG1(zkey.vk_alpha_1);
|
|
await writeG1(zkey.vk_beta_1);
|
|
await writeG2(zkey.vk_beta_2);
|
|
await writeG2(zkey.vk_gamma_2);
|
|
await writeG1(zkey.vk_delta_1);
|
|
await writeG2(zkey.vk_delta_2);
|
|
|
|
// IC
|
|
let buffBasesIC;
|
|
buffBasesIC = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 3);
|
|
buffBasesIC = await curve.G1.batchLEMtoU(buffBasesIC);
|
|
|
|
await writePointArray("G1", buffBasesIC);
|
|
|
|
/////////////////////
|
|
// h Section
|
|
/////////////////////
|
|
const buffBasesH_Lodd = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 9);
|
|
|
|
let buffBasesH_Tau;
|
|
buffBasesH_Tau = await curve.G1.fft(buffBasesH_Lodd, "affine", "jacobian", logger);
|
|
buffBasesH_Tau = await curve.G1.batchApplyKey(buffBasesH_Tau, curve.Fr.neg(curve.Fr.e(2)), curve.Fr.w[zkey.power+1], "jacobian", "affine", logger);
|
|
|
|
// Remove last element. (The degree of H will be allways m-2)
|
|
buffBasesH_Tau = buffBasesH_Tau.slice(0, buffBasesH_Tau.byteLength - sG1);
|
|
buffBasesH_Tau = await curve.G1.batchLEMtoU(buffBasesH_Tau);
|
|
await writePointArray("G1", buffBasesH_Tau);
|
|
|
|
/////////////////////
|
|
// L section
|
|
/////////////////////
|
|
let buffBasesC;
|
|
buffBasesC = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 8);
|
|
buffBasesC = await curve.G1.batchLEMtoU(buffBasesC);
|
|
await writePointArray("G1", buffBasesC);
|
|
|
|
/////////////////////
|
|
// A Section (C section)
|
|
/////////////////////
|
|
let buffBasesA;
|
|
buffBasesA = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 5);
|
|
buffBasesA = await curve.G1.batchLEMtoU(buffBasesA);
|
|
await writePointArray("G1", buffBasesA);
|
|
|
|
/////////////////////
|
|
// B1 Section
|
|
/////////////////////
|
|
let buffBasesB1;
|
|
buffBasesB1 = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 6);
|
|
buffBasesB1 = await curve.G1.batchLEMtoU(buffBasesB1);
|
|
await writePointArray("G1", buffBasesB1);
|
|
|
|
/////////////////////
|
|
// B2 Section
|
|
/////////////////////
|
|
let buffBasesB2;
|
|
buffBasesB2 = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 7);
|
|
buffBasesB2 = await curve.G2.batchLEMtoU(buffBasesB2);
|
|
await writePointArray("G2", buffBasesB2);
|
|
|
|
await fdMPCParams.write(mpcParams.csHash);
|
|
await writeU32(mpcParams.contributions.length);
|
|
|
|
for (let i=0; i<mpcParams.contributions.length; i++) {
|
|
const c = mpcParams.contributions[i];
|
|
await writeG1(c.deltaAfter);
|
|
await writeG1(c.delta.g1_s);
|
|
await writeG1(c.delta.g1_sx);
|
|
await writeG2(c.delta.g2_spx);
|
|
await fdMPCParams.write(c.transcript);
|
|
}
|
|
|
|
await fdZKey.close();
|
|
await fdMPCParams.close();
|
|
|
|
async function writeG1(P) {
|
|
const buff = new Uint8Array(sG1);
|
|
curve.G1.toRprUncompressed(buff, 0, P);
|
|
await fdMPCParams.write(buff);
|
|
}
|
|
|
|
async function writeG2(P) {
|
|
const buff = new Uint8Array(sG2);
|
|
curve.G2.toRprUncompressed(buff, 0, P);
|
|
await fdMPCParams.write(buff);
|
|
}
|
|
|
|
async function writePointArray(groupName, buff) {
|
|
let sG;
|
|
if (groupName == "G1") {
|
|
sG = sG1;
|
|
} else {
|
|
sG = sG2;
|
|
}
|
|
|
|
const buffSize = new Uint8Array(4);
|
|
const buffSizeV = new DataView(buffSize.buffer, buffSize.byteOffset, buffSize.byteLength);
|
|
buffSizeV.setUint32(0, buff.byteLength / sG, false);
|
|
|
|
await fdMPCParams.write(buffSize);
|
|
await fdMPCParams.write(buff);
|
|
}
|
|
|
|
async function writeU32(n) {
|
|
const buffSize = new Uint8Array(4);
|
|
const buffSizeV = new DataView(buffSize.buffer, buffSize.byteOffset, buffSize.byteLength);
|
|
buffSizeV.setUint32(0, n, false);
|
|
|
|
await fdMPCParams.write(buffSize);
|
|
}
|
|
|
|
|
|
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function phase2importMPCParams(zkeyNameOld, mpcparamsName, zkeyNameNew, name, logger) {
|
|
|
|
const {fd: fdZKeyOld, sections: sectionsZKeyOld} = await binFileUtils__namespace.readBinFile(zkeyNameOld, "zkey", 2);
|
|
const zkeyHeader = await readHeader$1(fdZKeyOld, sectionsZKeyOld, false);
|
|
if (zkeyHeader.protocol != "groth16") {
|
|
throw new Error("zkey file is not groth16");
|
|
}
|
|
|
|
const curve = await getCurveFromQ(zkeyHeader.q);
|
|
const sG1 = curve.G1.F.n8*2;
|
|
const sG2 = curve.G2.F.n8*2;
|
|
|
|
const oldMPCParams = await readMPCParams(fdZKeyOld, curve, sectionsZKeyOld);
|
|
const newMPCParams = {};
|
|
|
|
const fdMPCParams = await fastFile__namespace.readExisting(mpcparamsName);
|
|
|
|
fdMPCParams.pos =
|
|
sG1*3 + sG2*3 + // vKey
|
|
8 + sG1*zkeyHeader.nVars + // IC + C
|
|
4 + sG1*(zkeyHeader.domainSize-1) + // H
|
|
4 + sG1*zkeyHeader.nVars + // A
|
|
4 + sG1*zkeyHeader.nVars + // B1
|
|
4 + sG2*zkeyHeader.nVars; // B2
|
|
|
|
// csHash
|
|
newMPCParams.csHash = await fdMPCParams.read(64);
|
|
|
|
const nConttributions = await fdMPCParams.readUBE32();
|
|
newMPCParams.contributions = [];
|
|
for (let i=0; i<nConttributions; i++) {
|
|
const c = { delta:{} };
|
|
c.deltaAfter = await readG1(fdMPCParams);
|
|
c.delta.g1_s = await readG1(fdMPCParams);
|
|
c.delta.g1_sx = await readG1(fdMPCParams);
|
|
c.delta.g2_spx = await readG2(fdMPCParams);
|
|
c.transcript = await fdMPCParams.read(64);
|
|
if (i<oldMPCParams.contributions.length) {
|
|
c.type = oldMPCParams.contributions[i].type;
|
|
if (c.type==1) {
|
|
c.beaconHash = oldMPCParams.contributions[i].beaconHash;
|
|
c.numIterationsExp = oldMPCParams.contributions[i].numIterationsExp;
|
|
}
|
|
if (oldMPCParams.contributions[i].name) {
|
|
c.name = oldMPCParams.contributions[i].name;
|
|
}
|
|
}
|
|
newMPCParams.contributions.push(c);
|
|
}
|
|
|
|
if (!hashIsEqual(newMPCParams.csHash, oldMPCParams.csHash)) {
|
|
if (logger) logger.error("Hash of the original circuit does not match with the MPC one");
|
|
return false;
|
|
}
|
|
|
|
if (oldMPCParams.contributions.length > newMPCParams.contributions.length) {
|
|
if (logger) logger.error("The impoerted file does not include new contributions");
|
|
return false;
|
|
}
|
|
|
|
for (let i=0; i<oldMPCParams.contributions.length; i++) {
|
|
if (!contributionIsEqual(oldMPCParams.contributions[i], newMPCParams.contributions[i])) {
|
|
if (logger) logger.error(`Previos contribution ${i} does not match`);
|
|
return false;
|
|
}
|
|
}
|
|
|
|
|
|
// Set the same name to all new controbutions
|
|
if (name) {
|
|
for (let i=oldMPCParams.contributions.length; i<newMPCParams.contributions.length; i++) {
|
|
newMPCParams.contributions[i].name = name;
|
|
}
|
|
}
|
|
|
|
const fdZKeyNew = await binFileUtils__namespace.createBinFile(zkeyNameNew, "zkey", 1, 10);
|
|
fdMPCParams.pos = 0;
|
|
|
|
// Header
|
|
fdMPCParams.pos += sG1; // ignore alpha1 (keep original)
|
|
fdMPCParams.pos += sG1; // ignore beta1
|
|
fdMPCParams.pos += sG2; // ignore beta2
|
|
fdMPCParams.pos += sG2; // ignore gamma2
|
|
zkeyHeader.vk_delta_1 = await readG1(fdMPCParams);
|
|
zkeyHeader.vk_delta_2 = await readG2(fdMPCParams);
|
|
await writeHeader(fdZKeyNew, zkeyHeader);
|
|
|
|
// IC (Keep original)
|
|
const nIC = await fdMPCParams.readUBE32();
|
|
if (nIC != zkeyHeader.nPublic +1) {
|
|
if (logger) logger.error("Invalid number of points in IC");
|
|
await fdZKeyNew.discard();
|
|
return false;
|
|
}
|
|
fdMPCParams.pos += sG1*(zkeyHeader.nPublic+1);
|
|
await binFileUtils__namespace.copySection(fdZKeyOld, sectionsZKeyOld, fdZKeyNew, 3);
|
|
|
|
// Coeffs (Keep original)
|
|
await binFileUtils__namespace.copySection(fdZKeyOld, sectionsZKeyOld, fdZKeyNew, 4);
|
|
|
|
// H Section
|
|
const nH = await fdMPCParams.readUBE32();
|
|
if (nH != zkeyHeader.domainSize-1) {
|
|
if (logger) logger.error("Invalid number of points in H");
|
|
await fdZKeyNew.discard();
|
|
return false;
|
|
}
|
|
let buffH;
|
|
const buffTauU = await fdMPCParams.read(sG1*(zkeyHeader.domainSize-1));
|
|
const buffTauLEM = await curve.G1.batchUtoLEM(buffTauU);
|
|
buffH = new Uint8Array(zkeyHeader.domainSize*sG1);
|
|
buffH.set(buffTauLEM); // Let the last one to zero.
|
|
curve.G1.toRprLEM(buffH, sG1*(zkeyHeader.domainSize-1), curve.G1.zeroAffine);
|
|
const n2Inv = curve.Fr.neg(curve.Fr.inv(curve.Fr.e(2)));
|
|
const wInv = curve.Fr.inv(curve.Fr.w[zkeyHeader.power+1]);
|
|
buffH = await curve.G1.batchApplyKey(buffH, n2Inv, wInv, "affine", "jacobian", logger);
|
|
buffH = await curve.G1.ifft(buffH, "jacobian", "affine", logger);
|
|
await binFileUtils__namespace.startWriteSection(fdZKeyNew, 9);
|
|
await fdZKeyNew.write(buffH);
|
|
await binFileUtils__namespace.endWriteSection(fdZKeyNew);
|
|
|
|
// C Secion (L section)
|
|
const nL = await fdMPCParams.readUBE32();
|
|
if (nL != (zkeyHeader.nVars-zkeyHeader.nPublic-1)) {
|
|
if (logger) logger.error("Invalid number of points in L");
|
|
await fdZKeyNew.discard();
|
|
return false;
|
|
}
|
|
let buffL;
|
|
buffL = await fdMPCParams.read(sG1*(zkeyHeader.nVars-zkeyHeader.nPublic-1));
|
|
buffL = await curve.G1.batchUtoLEM(buffL);
|
|
await binFileUtils__namespace.startWriteSection(fdZKeyNew, 8);
|
|
await fdZKeyNew.write(buffL);
|
|
await binFileUtils__namespace.endWriteSection(fdZKeyNew);
|
|
|
|
// A Section
|
|
const nA = await fdMPCParams.readUBE32();
|
|
if (nA != zkeyHeader.nVars) {
|
|
if (logger) logger.error("Invalid number of points in A");
|
|
await fdZKeyNew.discard();
|
|
return false;
|
|
}
|
|
fdMPCParams.pos += sG1*(zkeyHeader.nVars);
|
|
await binFileUtils__namespace.copySection(fdZKeyOld, sectionsZKeyOld, fdZKeyNew, 5);
|
|
|
|
// B1 Section
|
|
const nB1 = await fdMPCParams.readUBE32();
|
|
if (nB1 != zkeyHeader.nVars) {
|
|
if (logger) logger.error("Invalid number of points in B1");
|
|
await fdZKeyNew.discard();
|
|
return false;
|
|
}
|
|
fdMPCParams.pos += sG1*(zkeyHeader.nVars);
|
|
await binFileUtils__namespace.copySection(fdZKeyOld, sectionsZKeyOld, fdZKeyNew, 6);
|
|
|
|
// B2 Section
|
|
const nB2 = await fdMPCParams.readUBE32();
|
|
if (nB2 != zkeyHeader.nVars) {
|
|
if (logger) logger.error("Invalid number of points in B2");
|
|
await fdZKeyNew.discard();
|
|
return false;
|
|
}
|
|
fdMPCParams.pos += sG2*(zkeyHeader.nVars);
|
|
await binFileUtils__namespace.copySection(fdZKeyOld, sectionsZKeyOld, fdZKeyNew, 7);
|
|
|
|
await writeMPCParams(fdZKeyNew, curve, newMPCParams);
|
|
|
|
await fdMPCParams.close();
|
|
await fdZKeyNew.close();
|
|
await fdZKeyOld.close();
|
|
|
|
return true;
|
|
|
|
async function readG1(fd) {
|
|
const buff = await fd.read(curve.G1.F.n8*2);
|
|
return curve.G1.fromRprUncompressed(buff, 0);
|
|
}
|
|
|
|
async function readG2(fd) {
|
|
const buff = await fd.read(curve.G2.F.n8*2);
|
|
return curve.G2.fromRprUncompressed(buff, 0);
|
|
}
|
|
|
|
|
|
function contributionIsEqual(c1, c2) {
|
|
if (!curve.G1.eq(c1.deltaAfter , c2.deltaAfter)) return false;
|
|
if (!curve.G1.eq(c1.delta.g1_s , c2.delta.g1_s)) return false;
|
|
if (!curve.G1.eq(c1.delta.g1_sx , c2.delta.g1_sx)) return false;
|
|
if (!curve.G2.eq(c1.delta.g2_spx , c2.delta.g2_spx)) return false;
|
|
if (!hashIsEqual(c1.transcript, c2.transcript)) return false;
|
|
return true;
|
|
}
|
|
|
|
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
const sameRatio = sameRatio$2;
|
|
|
|
|
|
|
|
async function phase2verifyFromInit(initFileName, pTauFileName, zkeyFileName, logger) {
|
|
|
|
let sr;
|
|
await Blake2b__default["default"].ready();
|
|
|
|
const {fd, sections} = await binFileUtils__namespace.readBinFile(zkeyFileName, "zkey", 2);
|
|
const zkey = await readHeader$1(fd, sections, false);
|
|
if (zkey.protocol != "groth16") {
|
|
throw new Error("zkey file is not groth16");
|
|
}
|
|
|
|
const curve = await getCurveFromQ(zkey.q);
|
|
const sG1 = curve.G1.F.n8*2;
|
|
|
|
const mpcParams = await readMPCParams(fd, curve, sections);
|
|
|
|
const accumulatedHasher = Blake2b__default["default"](64);
|
|
accumulatedHasher.update(mpcParams.csHash);
|
|
let curDelta = curve.G1.g;
|
|
for (let i=0; i<mpcParams.contributions.length; i++) {
|
|
const c = mpcParams.contributions[i];
|
|
const ourHasher = cloneHasher(accumulatedHasher);
|
|
|
|
hashG1(ourHasher, curve, c.delta.g1_s);
|
|
hashG1(ourHasher, curve, c.delta.g1_sx);
|
|
|
|
if (!hashIsEqual(ourHasher.digest(), c.transcript)) {
|
|
console.log(`INVALID(${i}): Inconsistent transcript `);
|
|
return false;
|
|
}
|
|
|
|
const delta_g2_sp = hashToG2(curve, c.transcript);
|
|
|
|
sr = await sameRatio(curve, c.delta.g1_s, c.delta.g1_sx, delta_g2_sp, c.delta.g2_spx);
|
|
if (sr !== true) {
|
|
console.log(`INVALID(${i}): public key G1 and G2 do not have the same ration `);
|
|
return false;
|
|
}
|
|
|
|
sr = await sameRatio(curve, curDelta, c.deltaAfter, delta_g2_sp, c.delta.g2_spx);
|
|
if (sr !== true) {
|
|
console.log(`INVALID(${i}): deltaAfter does not fillow the public key `);
|
|
return false;
|
|
}
|
|
|
|
if (c.type == 1) {
|
|
const rng = rngFromBeaconParams(c.beaconHash, c.numIterationsExp);
|
|
const expected_prvKey = curve.Fr.fromRng(rng);
|
|
const expected_g1_s = curve.G1.toAffine(curve.G1.fromRng(rng));
|
|
const expected_g1_sx = curve.G1.toAffine(curve.G1.timesFr(expected_g1_s, expected_prvKey));
|
|
if (curve.G1.eq(expected_g1_s, c.delta.g1_s) !== true) {
|
|
console.log(`INVALID(${i}): Key of the beacon does not match. g1_s `);
|
|
return false;
|
|
}
|
|
if (curve.G1.eq(expected_g1_sx, c.delta.g1_sx) !== true) {
|
|
console.log(`INVALID(${i}): Key of the beacon does not match. g1_sx `);
|
|
return false;
|
|
}
|
|
}
|
|
|
|
hashPubKey(accumulatedHasher, curve, c);
|
|
|
|
const contributionHasher = Blake2b__default["default"](64);
|
|
hashPubKey(contributionHasher, curve, c);
|
|
|
|
c.contributionHash = contributionHasher.digest();
|
|
|
|
curDelta = c.deltaAfter;
|
|
}
|
|
|
|
|
|
const {fd: fdInit, sections: sectionsInit} = await binFileUtils__namespace.readBinFile(initFileName, "zkey", 2);
|
|
const zkeyInit = await readHeader$1(fdInit, sectionsInit, false);
|
|
|
|
if (zkeyInit.protocol != "groth16") {
|
|
throw new Error("zkeyinit file is not groth16");
|
|
}
|
|
|
|
if ( (!ffjavascript.Scalar.eq(zkeyInit.q, zkey.q))
|
|
||(!ffjavascript.Scalar.eq(zkeyInit.r, zkey.r))
|
|
||(zkeyInit.n8q != zkey.n8q)
|
|
||(zkeyInit.n8r != zkey.n8r))
|
|
{
|
|
if (logger) logger.error("INVALID: Different curves");
|
|
return false;
|
|
}
|
|
|
|
if ( (zkeyInit.nVars != zkey.nVars)
|
|
||(zkeyInit.nPublic != zkey.nPublic)
|
|
||(zkeyInit.domainSize != zkey.domainSize))
|
|
{
|
|
if (logger) logger.error("INVALID: Different circuit parameters");
|
|
return false;
|
|
}
|
|
|
|
if (!curve.G1.eq(zkey.vk_alpha_1, zkeyInit.vk_alpha_1)) {
|
|
if (logger) logger.error("INVALID: Invalid alpha1");
|
|
return false;
|
|
}
|
|
if (!curve.G1.eq(zkey.vk_beta_1, zkeyInit.vk_beta_1)) {
|
|
if (logger) logger.error("INVALID: Invalid beta1");
|
|
return false;
|
|
}
|
|
if (!curve.G2.eq(zkey.vk_beta_2, zkeyInit.vk_beta_2)) {
|
|
if (logger) logger.error("INVALID: Invalid beta2");
|
|
return false;
|
|
}
|
|
if (!curve.G2.eq(zkey.vk_gamma_2, zkeyInit.vk_gamma_2)) {
|
|
if (logger) logger.error("INVALID: Invalid gamma2");
|
|
return false;
|
|
}
|
|
if (!curve.G1.eq(zkey.vk_delta_1, curDelta)) {
|
|
if (logger) logger.error("INVALID: Invalid delta1");
|
|
return false;
|
|
}
|
|
sr = await sameRatio(curve, curve.G1.g, curDelta, curve.G2.g, zkey.vk_delta_2);
|
|
if (sr !== true) {
|
|
if (logger) logger.error("INVALID: Invalid delta2");
|
|
return false;
|
|
}
|
|
|
|
const mpcParamsInit = await readMPCParams(fdInit, curve, sectionsInit);
|
|
if (!hashIsEqual(mpcParams.csHash, mpcParamsInit.csHash)) {
|
|
if (logger) logger.error("INVALID: Circuit does not match");
|
|
return false;
|
|
}
|
|
|
|
// Check sizes of sections
|
|
if (sections[8][0].size != sG1*(zkey.nVars-zkey.nPublic-1)) {
|
|
if (logger) logger.error("INVALID: Invalid L section size");
|
|
return false;
|
|
}
|
|
|
|
if (sections[9][0].size != sG1*(zkey.domainSize)) {
|
|
if (logger) logger.error("INVALID: Invalid H section size");
|
|
return false;
|
|
}
|
|
|
|
let ss;
|
|
ss = await binFileUtils__namespace.sectionIsEqual(fd, sections, fdInit, sectionsInit, 3);
|
|
if (!ss) {
|
|
if (logger) logger.error("INVALID: IC section is not identical");
|
|
return false;
|
|
}
|
|
|
|
ss = await binFileUtils__namespace.sectionIsEqual(fd, sections, fdInit, sectionsInit, 4);
|
|
if (!ss) {
|
|
if (logger) logger.error("Coeffs section is not identical");
|
|
return false;
|
|
}
|
|
|
|
ss = await binFileUtils__namespace.sectionIsEqual(fd, sections, fdInit, sectionsInit, 5);
|
|
if (!ss) {
|
|
if (logger) logger.error("A section is not identical");
|
|
return false;
|
|
}
|
|
|
|
ss = await binFileUtils__namespace.sectionIsEqual(fd, sections, fdInit, sectionsInit, 6);
|
|
if (!ss) {
|
|
if (logger) logger.error("B1 section is not identical");
|
|
return false;
|
|
}
|
|
|
|
ss = await binFileUtils__namespace.sectionIsEqual(fd, sections, fdInit, sectionsInit, 7);
|
|
if (!ss) {
|
|
if (logger) logger.error("B2 section is not identical");
|
|
return false;
|
|
}
|
|
|
|
// Check L
|
|
sr = await sectionHasSameRatio("G1", fdInit, sectionsInit, fd, sections, 8, zkey.vk_delta_2, zkeyInit.vk_delta_2, "L section");
|
|
if (sr!==true) {
|
|
if (logger) logger.error("L section does not match");
|
|
return false;
|
|
}
|
|
|
|
// Check H
|
|
sr = await sameRatioH();
|
|
if (sr!==true) {
|
|
if (logger) logger.error("H section does not match");
|
|
return false;
|
|
}
|
|
|
|
if (logger) logger.info(formatHash(mpcParams.csHash, "Circuit Hash: "));
|
|
|
|
await fd.close();
|
|
await fdInit.close();
|
|
|
|
for (let i=mpcParams.contributions.length-1; i>=0; i--) {
|
|
const c = mpcParams.contributions[i];
|
|
if (logger) logger.info("-------------------------");
|
|
if (logger) logger.info(formatHash(c.contributionHash, `contribution #${i+1} ${c.name ? c.name : ""}:`));
|
|
if (c.type == 1) {
|
|
if (logger) logger.info(`Beacon generator: ${byteArray2hex(c.beaconHash)}`);
|
|
if (logger) logger.info(`Beacon iterations Exp: ${c.numIterationsExp}`);
|
|
}
|
|
}
|
|
if (logger) logger.info("-------------------------");
|
|
|
|
if (logger) logger.info("ZKey Ok!");
|
|
|
|
return true;
|
|
|
|
|
|
async function sectionHasSameRatio(groupName, fd1, sections1, fd2, sections2, idSection, g2sp, g2spx, sectionName) {
|
|
const MAX_CHUNK_SIZE = 1<<20;
|
|
const G = curve[groupName];
|
|
const sG = G.F.n8*2;
|
|
await binFileUtils__namespace.startReadUniqueSection(fd1, sections1, idSection);
|
|
await binFileUtils__namespace.startReadUniqueSection(fd2, sections2, idSection);
|
|
|
|
let R1 = G.zero;
|
|
let R2 = G.zero;
|
|
|
|
const nPoints = sections1[idSection][0].size / sG;
|
|
|
|
for (let i=0; i<nPoints; i += MAX_CHUNK_SIZE) {
|
|
if (logger) logger.debug(`Same ratio check ${sectionName}: ${i}/${nPoints}`);
|
|
const n = Math.min(nPoints - i, MAX_CHUNK_SIZE);
|
|
const bases1 = await fd1.read(n*sG);
|
|
const bases2 = await fd2.read(n*sG);
|
|
|
|
const scalars = new Uint8Array(4*n);
|
|
crypto__default["default"].randomFillSync(scalars);
|
|
|
|
|
|
const r1 = await G.multiExpAffine(bases1, scalars);
|
|
const r2 = await G.multiExpAffine(bases2, scalars);
|
|
|
|
R1 = G.add(R1, r1);
|
|
R2 = G.add(R2, r2);
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd1);
|
|
await binFileUtils__namespace.endReadSection(fd2);
|
|
|
|
if (nPoints == 0) return true;
|
|
|
|
sr = await sameRatio(curve, R1, R2, g2sp, g2spx);
|
|
if (sr !== true) return false;
|
|
|
|
return true;
|
|
}
|
|
|
|
async function sameRatioH() {
|
|
const MAX_CHUNK_SIZE = 1<<20;
|
|
const G = curve.G1;
|
|
const Fr = curve.Fr;
|
|
const sG = G.F.n8*2;
|
|
|
|
const {fd: fdPTau, sections: sectionsPTau} = await binFileUtils__namespace.readBinFile(pTauFileName, "ptau", 1);
|
|
|
|
let buff_r = new ffjavascript.BigBuffer(zkey.domainSize * zkey.n8r);
|
|
|
|
const seed= new Array(8);
|
|
for (let i=0; i<8; i++) {
|
|
seed[i] = crypto__default["default"].randomBytes(4).readUInt32BE(0, true);
|
|
}
|
|
const rng = new ffjavascript.ChaCha(seed);
|
|
for (let i=0; i<zkey.domainSize-1; i++) { // Note that last one is zero
|
|
const e = Fr.fromRng(rng);
|
|
Fr.toRprLE(buff_r, i*zkey.n8r, e);
|
|
}
|
|
Fr.toRprLE(buff_r, (zkey.domainSize-1)*zkey.n8r, Fr.zero);
|
|
|
|
let R1 = G.zero;
|
|
for (let i=0; i<zkey.domainSize; i += MAX_CHUNK_SIZE) {
|
|
if (logger) logger.debug(`H Verificaition(tau): ${i}/${zkey.domainSize}`);
|
|
const n = Math.min(zkey.domainSize - i, MAX_CHUNK_SIZE);
|
|
|
|
const buff1 = await fdPTau.read(sG*n, sectionsPTau[2][0].p + zkey.domainSize*sG + i*sG);
|
|
const buff2 = await fdPTau.read(sG*n, sectionsPTau[2][0].p + i*sG);
|
|
|
|
const buffB = await batchSubstract(buff1, buff2);
|
|
const buffS = buff_r.slice(i*zkey.n8r, (i+n)*zkey.n8r);
|
|
const r = await G.multiExpAffine(buffB, buffS);
|
|
|
|
R1 = G.add(R1, r);
|
|
}
|
|
|
|
// Caluclate odd coeficients in transformed domain
|
|
|
|
buff_r = await Fr.batchToMontgomery(buff_r);
|
|
// const first = curve.Fr.neg(curve.Fr.inv(curve.Fr.e(2)));
|
|
// Works*2 const first = curve.Fr.neg(curve.Fr.e(2));
|
|
|
|
|
|
let first;
|
|
|
|
if (zkey.power < Fr.s) {
|
|
first = Fr.neg(Fr.e(2));
|
|
} else {
|
|
const small_m = 2 ** Fr.s;
|
|
const shift_to_small_m = Fr.exp(Fr.shift, small_m);
|
|
first = Fr.sub( shift_to_small_m, Fr.one);
|
|
}
|
|
|
|
// const inc = curve.Fr.inv(curve.PFr.w[zkey.power+1]);
|
|
const inc = zkey.power < Fr.s ? Fr.w[zkey.power+1] : Fr.shift;
|
|
buff_r = await Fr.batchApplyKey(buff_r, first, inc);
|
|
buff_r = await Fr.fft(buff_r);
|
|
buff_r = await Fr.batchFromMontgomery(buff_r);
|
|
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 9);
|
|
let R2 = G.zero;
|
|
for (let i=0; i<zkey.domainSize; i += MAX_CHUNK_SIZE) {
|
|
if (logger) logger.debug(`H Verificaition(lagrange): ${i}/${zkey.domainSize}`);
|
|
const n = Math.min(zkey.domainSize - i, MAX_CHUNK_SIZE);
|
|
|
|
const buff = await fd.read(sG*n);
|
|
const buffS = buff_r.slice(i*zkey.n8r, (i+n)*zkey.n8r);
|
|
const r = await G.multiExpAffine(buff, buffS);
|
|
|
|
R2 = G.add(R2, r);
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
sr = await sameRatio(curve, R1, R2, zkey.vk_delta_2, zkeyInit.vk_delta_2);
|
|
if (sr !== true) return false;
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
async function batchSubstract(buff1, buff2) {
|
|
const sG = curve.G1.F.n8*2;
|
|
const nPoints = buff1.byteLength / sG;
|
|
const concurrency= curve.tm.concurrency;
|
|
const nPointsPerThread = Math.floor(nPoints / concurrency);
|
|
const opPromises = [];
|
|
for (let i=0; i<concurrency; i++) {
|
|
let n;
|
|
if (i< concurrency-1) {
|
|
n = nPointsPerThread;
|
|
} else {
|
|
n = nPoints - i*nPointsPerThread;
|
|
}
|
|
if (n==0) continue;
|
|
|
|
const subBuff1 = buff1.slice(i*nPointsPerThread*sG1, (i*nPointsPerThread+n)*sG1);
|
|
const subBuff2 = buff2.slice(i*nPointsPerThread*sG1, (i*nPointsPerThread+n)*sG1);
|
|
opPromises.push(batchSubstractThread(subBuff1, subBuff2));
|
|
}
|
|
|
|
|
|
const result = await Promise.all(opPromises);
|
|
|
|
const fullBuffOut = new Uint8Array(nPoints*sG);
|
|
let p =0;
|
|
for (let i=0; i<result.length; i++) {
|
|
fullBuffOut.set(result[i][0], p);
|
|
p+=result[i][0].byteLength;
|
|
}
|
|
|
|
return fullBuffOut;
|
|
}
|
|
|
|
|
|
async function batchSubstractThread(buff1, buff2) {
|
|
const sG1 = curve.G1.F.n8*2;
|
|
const sGmid = curve.G1.F.n8*3;
|
|
const nPoints = buff1.byteLength/sG1;
|
|
const task = [];
|
|
task.push({cmd: "ALLOCSET", var: 0, buff: buff1});
|
|
task.push({cmd: "ALLOCSET", var: 1, buff: buff2});
|
|
task.push({cmd: "ALLOC", var: 2, len: nPoints*sGmid});
|
|
for (let i=0; i<nPoints; i++) {
|
|
task.push({
|
|
cmd: "CALL",
|
|
fnName: "g1m_subAffine",
|
|
params: [
|
|
{var: 0, offset: i*sG1},
|
|
{var: 1, offset: i*sG1},
|
|
{var: 2, offset: i*sGmid},
|
|
]
|
|
});
|
|
}
|
|
task.push({cmd: "CALL", fnName: "g1m_batchToAffine", params: [
|
|
{var: 2},
|
|
{val: nPoints},
|
|
{var: 2},
|
|
]});
|
|
task.push({cmd: "GET", out: 0, var: 2, len: nPoints*sG1});
|
|
|
|
const res = await curve.tm.queueAction(task);
|
|
|
|
return res;
|
|
}
|
|
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function phase2verifyFromR1cs(r1csFileName, pTauFileName, zkeyFileName, logger) {
|
|
|
|
// const initFileName = "~" + zkeyFileName + ".init";
|
|
const initFileName = {type: "bigMem"};
|
|
await newZKey(r1csFileName, pTauFileName, initFileName, logger);
|
|
|
|
return await phase2verifyFromInit(initFileName, pTauFileName, zkeyFileName, logger);
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function phase2contribute(zkeyNameOld, zkeyNameNew, name, entropy, logger) {
|
|
await Blake2b__default["default"].ready();
|
|
|
|
const {fd: fdOld, sections: sections} = await binFileUtils__namespace.readBinFile(zkeyNameOld, "zkey", 2);
|
|
const zkey = await readHeader$1(fdOld, sections);
|
|
if (zkey.protocol != "groth16") {
|
|
throw new Error("zkey file is not groth16");
|
|
}
|
|
|
|
const curve = await getCurveFromQ(zkey.q);
|
|
|
|
const mpcParams = await readMPCParams(fdOld, curve, sections);
|
|
|
|
const fdNew = await binFileUtils__namespace.createBinFile(zkeyNameNew, "zkey", 1, 10);
|
|
|
|
|
|
const rng = await getRandomRng(entropy);
|
|
|
|
const transcriptHasher = Blake2b__default["default"](64);
|
|
transcriptHasher.update(mpcParams.csHash);
|
|
for (let i=0; i<mpcParams.contributions.length; i++) {
|
|
hashPubKey(transcriptHasher, curve, mpcParams.contributions[i]);
|
|
}
|
|
|
|
const curContribution = {};
|
|
curContribution.delta = {};
|
|
curContribution.delta.prvKey = curve.Fr.fromRng(rng);
|
|
curContribution.delta.g1_s = curve.G1.toAffine(curve.G1.fromRng(rng));
|
|
curContribution.delta.g1_sx = curve.G1.toAffine(curve.G1.timesFr(curContribution.delta.g1_s, curContribution.delta.prvKey));
|
|
hashG1(transcriptHasher, curve, curContribution.delta.g1_s);
|
|
hashG1(transcriptHasher, curve, curContribution.delta.g1_sx);
|
|
curContribution.transcript = transcriptHasher.digest();
|
|
curContribution.delta.g2_sp = hashToG2(curve, curContribution.transcript);
|
|
curContribution.delta.g2_spx = curve.G2.toAffine(curve.G2.timesFr(curContribution.delta.g2_sp, curContribution.delta.prvKey));
|
|
|
|
zkey.vk_delta_1 = curve.G1.timesFr(zkey.vk_delta_1, curContribution.delta.prvKey);
|
|
zkey.vk_delta_2 = curve.G2.timesFr(zkey.vk_delta_2, curContribution.delta.prvKey);
|
|
|
|
curContribution.deltaAfter = zkey.vk_delta_1;
|
|
|
|
curContribution.type = 0;
|
|
if (name) curContribution.name = name;
|
|
|
|
mpcParams.contributions.push(curContribution);
|
|
|
|
await writeHeader(fdNew, zkey);
|
|
|
|
// IC
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 3);
|
|
|
|
// Coeffs (Keep original)
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 4);
|
|
|
|
// A Section
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 5);
|
|
|
|
// B1 Section
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 6);
|
|
|
|
// B2 Section
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 7);
|
|
|
|
const invDelta = curve.Fr.inv(curContribution.delta.prvKey);
|
|
await applyKeyToSection(fdOld, sections, fdNew, 8, curve, "G1", invDelta, curve.Fr.e(1), "L Section", logger);
|
|
await applyKeyToSection(fdOld, sections, fdNew, 9, curve, "G1", invDelta, curve.Fr.e(1), "H Section", logger);
|
|
|
|
await writeMPCParams(fdNew, curve, mpcParams);
|
|
|
|
await fdOld.close();
|
|
await fdNew.close();
|
|
|
|
const contributionHasher = Blake2b__default["default"](64);
|
|
hashPubKey(contributionHasher, curve, curContribution);
|
|
|
|
const contribuionHash = contributionHasher.digest();
|
|
|
|
if (logger) logger.info(formatHash(mpcParams.csHash, "Circuit Hash: "));
|
|
if (logger) logger.info(formatHash(contribuionHash, "Contribution Hash: "));
|
|
|
|
return contribuionHash;
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
|
|
async function beacon(zkeyNameOld, zkeyNameNew, name, beaconHashStr, numIterationsExp, logger) {
|
|
await Blake2b__default["default"].ready();
|
|
|
|
const beaconHash = hex2ByteArray(beaconHashStr);
|
|
if ( (beaconHash.byteLength == 0)
|
|
|| (beaconHash.byteLength*2 !=beaconHashStr.length))
|
|
{
|
|
if (logger) logger.error("Invalid Beacon Hash. (It must be a valid hexadecimal sequence)");
|
|
return false;
|
|
}
|
|
if (beaconHash.length>=256) {
|
|
if (logger) logger.error("Maximum lenght of beacon hash is 255 bytes");
|
|
return false;
|
|
}
|
|
|
|
numIterationsExp = parseInt(numIterationsExp);
|
|
if ((numIterationsExp<10)||(numIterationsExp>63)) {
|
|
if (logger) logger.error("Invalid numIterationsExp. (Must be between 10 and 63)");
|
|
return false;
|
|
}
|
|
|
|
|
|
const {fd: fdOld, sections: sections} = await binFileUtils__namespace.readBinFile(zkeyNameOld, "zkey", 2);
|
|
const zkey = await readHeader$1(fdOld, sections);
|
|
|
|
if (zkey.protocol != "groth16") {
|
|
throw new Error("zkey file is not groth16");
|
|
}
|
|
|
|
|
|
const curve = await getCurveFromQ(zkey.q);
|
|
|
|
const mpcParams = await readMPCParams(fdOld, curve, sections);
|
|
|
|
const fdNew = await binFileUtils__namespace.createBinFile(zkeyNameNew, "zkey", 1, 10);
|
|
|
|
const rng = await rngFromBeaconParams(beaconHash, numIterationsExp);
|
|
|
|
const transcriptHasher = Blake2b__default["default"](64);
|
|
transcriptHasher.update(mpcParams.csHash);
|
|
for (let i=0; i<mpcParams.contributions.length; i++) {
|
|
hashPubKey(transcriptHasher, curve, mpcParams.contributions[i]);
|
|
}
|
|
|
|
const curContribution = {};
|
|
curContribution.delta = {};
|
|
curContribution.delta.prvKey = curve.Fr.fromRng(rng);
|
|
curContribution.delta.g1_s = curve.G1.toAffine(curve.G1.fromRng(rng));
|
|
curContribution.delta.g1_sx = curve.G1.toAffine(curve.G1.timesFr(curContribution.delta.g1_s, curContribution.delta.prvKey));
|
|
hashG1(transcriptHasher, curve, curContribution.delta.g1_s);
|
|
hashG1(transcriptHasher, curve, curContribution.delta.g1_sx);
|
|
curContribution.transcript = transcriptHasher.digest();
|
|
curContribution.delta.g2_sp = hashToG2(curve, curContribution.transcript);
|
|
curContribution.delta.g2_spx = curve.G2.toAffine(curve.G2.timesFr(curContribution.delta.g2_sp, curContribution.delta.prvKey));
|
|
|
|
zkey.vk_delta_1 = curve.G1.timesFr(zkey.vk_delta_1, curContribution.delta.prvKey);
|
|
zkey.vk_delta_2 = curve.G2.timesFr(zkey.vk_delta_2, curContribution.delta.prvKey);
|
|
|
|
curContribution.deltaAfter = zkey.vk_delta_1;
|
|
|
|
curContribution.type = 1;
|
|
curContribution.numIterationsExp = numIterationsExp;
|
|
curContribution.beaconHash = beaconHash;
|
|
|
|
if (name) curContribution.name = name;
|
|
|
|
mpcParams.contributions.push(curContribution);
|
|
|
|
await writeHeader(fdNew, zkey);
|
|
|
|
// IC
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 3);
|
|
|
|
// Coeffs (Keep original)
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 4);
|
|
|
|
// A Section
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 5);
|
|
|
|
// B1 Section
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 6);
|
|
|
|
// B2 Section
|
|
await binFileUtils__namespace.copySection(fdOld, sections, fdNew, 7);
|
|
|
|
const invDelta = curve.Fr.inv(curContribution.delta.prvKey);
|
|
await applyKeyToSection(fdOld, sections, fdNew, 8, curve, "G1", invDelta, curve.Fr.e(1), "L Section", logger);
|
|
await applyKeyToSection(fdOld, sections, fdNew, 9, curve, "G1", invDelta, curve.Fr.e(1), "H Section", logger);
|
|
|
|
await writeMPCParams(fdNew, curve, mpcParams);
|
|
|
|
await fdOld.close();
|
|
await fdNew.close();
|
|
|
|
const contributionHasher = Blake2b__default["default"](64);
|
|
hashPubKey(contributionHasher, curve, curContribution);
|
|
|
|
const contribuionHash = contributionHasher.digest();
|
|
|
|
if (logger) logger.info(formatHash(contribuionHash, "Contribution Hash: "));
|
|
|
|
return contribuionHash;
|
|
}
|
|
|
|
async function zkeyExportJson(zkeyFileName) {
|
|
|
|
const zKey = await readZKey(zkeyFileName, true);
|
|
|
|
return zKey;
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function bellmanContribute(curve, challengeFilename, responesFileName, entropy, logger) {
|
|
await Blake2b__default["default"].ready();
|
|
|
|
const rng = await getRandomRng(entropy);
|
|
|
|
const delta = curve.Fr.fromRng(rng);
|
|
const invDelta = curve.Fr.inv(delta);
|
|
|
|
const sG1 = curve.G1.F.n8*2;
|
|
const sG2 = curve.G2.F.n8*2;
|
|
|
|
const fdFrom = await fastFile__namespace.readExisting(challengeFilename);
|
|
const fdTo = await fastFile__namespace.createOverride(responesFileName);
|
|
|
|
|
|
await copy(sG1); // alpha1
|
|
await copy(sG1); // beta1
|
|
await copy(sG2); // beta2
|
|
await copy(sG2); // gamma2
|
|
const oldDelta1 = await readG1();
|
|
const delta1 = curve.G1.timesFr(oldDelta1, delta);
|
|
await writeG1(delta1);
|
|
const oldDelta2 = await readG2();
|
|
const delta2 = curve.G2.timesFr(oldDelta2, delta);
|
|
await writeG2(delta2);
|
|
|
|
// IC
|
|
const nIC = await fdFrom.readUBE32();
|
|
await fdTo.writeUBE32(nIC);
|
|
await copy(nIC*sG1);
|
|
|
|
// H
|
|
const nH = await fdFrom.readUBE32();
|
|
await fdTo.writeUBE32(nH);
|
|
await applyKeyToChallengeSection(fdFrom, fdTo, null, curve, "G1", nH, invDelta, curve.Fr.e(1), "UNCOMPRESSED", "H", logger);
|
|
|
|
// L
|
|
const nL = await fdFrom.readUBE32();
|
|
await fdTo.writeUBE32(nL);
|
|
await applyKeyToChallengeSection(fdFrom, fdTo, null, curve, "G1", nL, invDelta, curve.Fr.e(1), "UNCOMPRESSED", "L", logger);
|
|
|
|
// A
|
|
const nA = await fdFrom.readUBE32();
|
|
await fdTo.writeUBE32(nA);
|
|
await copy(nA*sG1);
|
|
|
|
// B1
|
|
const nB1 = await fdFrom.readUBE32();
|
|
await fdTo.writeUBE32(nB1);
|
|
await copy(nB1*sG1);
|
|
|
|
// B2
|
|
const nB2 = await fdFrom.readUBE32();
|
|
await fdTo.writeUBE32(nB2);
|
|
await copy(nB2*sG2);
|
|
|
|
|
|
//////////
|
|
/// Read contributions
|
|
//////////
|
|
const transcriptHasher = Blake2b__default["default"](64);
|
|
|
|
const mpcParams = {};
|
|
// csHash
|
|
mpcParams.csHash = await fdFrom.read(64);
|
|
transcriptHasher.update(mpcParams.csHash);
|
|
|
|
const nConttributions = await fdFrom.readUBE32();
|
|
mpcParams.contributions = [];
|
|
for (let i=0; i<nConttributions; i++) {
|
|
const c = { delta:{} };
|
|
c.deltaAfter = await readG1();
|
|
c.delta.g1_s = await readG1();
|
|
c.delta.g1_sx = await readG1();
|
|
c.delta.g2_spx = await readG2();
|
|
c.transcript = await fdFrom.read(64);
|
|
mpcParams.contributions.push(c);
|
|
hashPubKey(transcriptHasher, curve, c);
|
|
}
|
|
|
|
const curContribution = {};
|
|
curContribution.delta = {};
|
|
curContribution.delta.prvKey = delta;
|
|
curContribution.delta.g1_s = curve.G1.toAffine(curve.G1.fromRng(rng));
|
|
curContribution.delta.g1_sx = curve.G1.toAffine(curve.G1.timesFr(curContribution.delta.g1_s, delta));
|
|
hashG1(transcriptHasher, curve, curContribution.delta.g1_s);
|
|
hashG1(transcriptHasher, curve, curContribution.delta.g1_sx);
|
|
curContribution.transcript = transcriptHasher.digest();
|
|
curContribution.delta.g2_sp = hashToG2(curve, curContribution.transcript);
|
|
curContribution.delta.g2_spx = curve.G2.toAffine(curve.G2.timesFr(curContribution.delta.g2_sp, delta));
|
|
curContribution.deltaAfter = delta1;
|
|
curContribution.type = 0;
|
|
mpcParams.contributions.push(curContribution);
|
|
|
|
|
|
//////////
|
|
/// Write COntribution
|
|
//////////
|
|
|
|
await fdTo.write(mpcParams.csHash);
|
|
await fdTo.writeUBE32(mpcParams.contributions.length);
|
|
|
|
for (let i=0; i<mpcParams.contributions.length; i++) {
|
|
const c = mpcParams.contributions[i];
|
|
await writeG1(c.deltaAfter);
|
|
await writeG1(c.delta.g1_s);
|
|
await writeG1(c.delta.g1_sx);
|
|
await writeG2(c.delta.g2_spx);
|
|
await fdTo.write(c.transcript);
|
|
}
|
|
|
|
const contributionHasher = Blake2b__default["default"](64);
|
|
hashPubKey(contributionHasher, curve, curContribution);
|
|
|
|
const contributionHash = contributionHasher.digest();
|
|
|
|
if (logger) logger.info(formatHash(contributionHash, "Contribution Hash: "));
|
|
|
|
await fdTo.close();
|
|
await fdFrom.close();
|
|
|
|
return contributionHash;
|
|
|
|
async function copy(nBytes) {
|
|
const CHUNK_SIZE = fdFrom.pageSize*2;
|
|
for (let i=0; i<nBytes; i+= CHUNK_SIZE) {
|
|
const n = Math.min(nBytes -i, CHUNK_SIZE);
|
|
const buff = await fdFrom.read(n);
|
|
await fdTo.write(buff);
|
|
}
|
|
}
|
|
|
|
async function readG1() {
|
|
const buff = await fdFrom.read(curve.G1.F.n8*2);
|
|
return curve.G1.fromRprUncompressed(buff, 0);
|
|
}
|
|
|
|
async function readG2() {
|
|
const buff = await fdFrom.read(curve.G2.F.n8*2);
|
|
return curve.G2.fromRprUncompressed(buff, 0);
|
|
}
|
|
|
|
async function writeG1(P) {
|
|
const buff = new Uint8Array(sG1);
|
|
curve.G1.toRprUncompressed(buff, 0, P);
|
|
await fdTo.write(buff);
|
|
}
|
|
|
|
async function writeG2(P) {
|
|
const buff = new Uint8Array(sG2);
|
|
curve.G2.toRprUncompressed(buff, 0, P);
|
|
await fdTo.write(buff);
|
|
}
|
|
|
|
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
const {stringifyBigInts: stringifyBigInts$1} = ffjavascript.utils;
|
|
|
|
async function zkeyExportVerificationKey(zkeyName, /* logger */ ) {
|
|
|
|
const {fd, sections} = await binFileUtils__namespace.readBinFile(zkeyName, "zkey", 2);
|
|
const zkey = await readHeader$1(fd, sections);
|
|
|
|
let res;
|
|
if (zkey.protocol == "groth16") {
|
|
res = await groth16Vk(zkey, fd, sections);
|
|
} else if (zkey.protocol == "plonk") {
|
|
res = await plonkVk(zkey);
|
|
} else {
|
|
throw new Error("zkey file is not groth16");
|
|
}
|
|
|
|
await fd.close();
|
|
|
|
return res;
|
|
}
|
|
|
|
|
|
async function groth16Vk(zkey, fd, sections) {
|
|
const curve = await getCurveFromQ(zkey.q);
|
|
const sG1 = curve.G1.F.n8*2;
|
|
|
|
const alphaBeta = await curve.pairing( zkey.vk_alpha_1 , zkey.vk_beta_2 );
|
|
|
|
let vKey = {
|
|
protocol: zkey.protocol,
|
|
curve: curve.name,
|
|
nPublic: zkey.nPublic,
|
|
|
|
vk_alpha_1: curve.G1.toObject(zkey.vk_alpha_1),
|
|
|
|
vk_beta_2: curve.G2.toObject(zkey.vk_beta_2),
|
|
vk_gamma_2: curve.G2.toObject(zkey.vk_gamma_2),
|
|
vk_delta_2: curve.G2.toObject(zkey.vk_delta_2),
|
|
|
|
vk_alphabeta_12: curve.Gt.toObject(alphaBeta)
|
|
};
|
|
|
|
// Read IC Section
|
|
///////////
|
|
await binFileUtils__namespace.startReadUniqueSection(fd, sections, 3);
|
|
vKey.IC = [];
|
|
for (let i=0; i<= zkey.nPublic; i++) {
|
|
const buff = await fd.read(sG1);
|
|
const P = curve.G1.toObject(buff);
|
|
vKey.IC.push(P);
|
|
}
|
|
await binFileUtils__namespace.endReadSection(fd);
|
|
|
|
vKey = stringifyBigInts$1(vKey);
|
|
|
|
return vKey;
|
|
}
|
|
|
|
|
|
async function plonkVk(zkey) {
|
|
const curve = await getCurveFromQ(zkey.q);
|
|
|
|
let vKey = {
|
|
protocol: zkey.protocol,
|
|
curve: curve.name,
|
|
nPublic: zkey.nPublic,
|
|
power: zkey.power,
|
|
|
|
k1: curve.Fr.toObject(zkey.k1),
|
|
k2: curve.Fr.toObject(zkey.k2),
|
|
|
|
Qm: curve.G1.toObject(zkey.Qm),
|
|
Ql: curve.G1.toObject(zkey.Ql),
|
|
Qr: curve.G1.toObject(zkey.Qr),
|
|
Qo: curve.G1.toObject(zkey.Qo),
|
|
Qc: curve.G1.toObject(zkey.Qc),
|
|
S1: curve.G1.toObject(zkey.S1),
|
|
S2: curve.G1.toObject(zkey.S2),
|
|
S3: curve.G1.toObject(zkey.S3),
|
|
|
|
X_2: curve.G2.toObject(zkey.X_2),
|
|
|
|
w: curve.Fr.toObject(curve.Fr.w[zkey.power])
|
|
};
|
|
|
|
vKey = stringifyBigInts$1(vKey);
|
|
|
|
return vKey;
|
|
}
|
|
|
|
// Not ready yet
|
|
// module.exports.generateVerifier_kimleeoh = generateVerifier_kimleeoh;
|
|
|
|
|
|
|
|
async function exportSolidityVerifier(zKeyName, templates, logger) {
|
|
|
|
const verificationKey = await zkeyExportVerificationKey(zKeyName);
|
|
|
|
let template = templates[verificationKey.protocol];
|
|
|
|
return ejs__default["default"].render(template , verificationKey);
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
var zkey = /*#__PURE__*/Object.freeze({
|
|
__proto__: null,
|
|
newZKey: newZKey,
|
|
exportBellman: phase2exportMPCParams,
|
|
importBellman: phase2importMPCParams,
|
|
verifyFromR1cs: phase2verifyFromR1cs,
|
|
verifyFromInit: phase2verifyFromInit,
|
|
contribute: phase2contribute,
|
|
beacon: beacon,
|
|
exportJson: zkeyExportJson,
|
|
bellmanContribute: bellmanContribute,
|
|
exportVerificationKey: zkeyExportVerificationKey,
|
|
exportSolidityVerifier: exportSolidityVerifier
|
|
});
|
|
|
|
/*
|
|
Copyright 2021 0kims association.
|
|
|
|
This file is part of snarkjs.
|
|
|
|
snarkjs is a free software: you can redistribute it and/or
|
|
modify it under the terms of the GNU General Public License as published by the
|
|
Free Software Foundation, either version 3 of the License, or (at your option)
|
|
any later version.
|
|
|
|
snarkjs is distributed in the hope that it will be useful,
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
|
|
more details.
|
|
|
|
You should have received a copy of the GNU General Public License along with
|
|
snarkjs. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
|
|
async function plonkSetup(r1csName, ptauName, zkeyName, logger) {
|
|
|
|
await Blake2b__default["default"].ready();
|
|
|
|
const {fd: fdPTau, sections: sectionsPTau} = await binFileUtils.readBinFile(ptauName, "ptau", 1, 1<<22, 1<<24);
|
|
const {curve, power} = await readPTauHeader(fdPTau, sectionsPTau);
|
|
const {fd: fdR1cs, sections: sectionsR1cs} = await binFileUtils.readBinFile(r1csName, "r1cs", 1, 1<<22, 1<<24);
|
|
const r1cs = await r1csfile.readR1csHeader(fdR1cs, sectionsR1cs, false);
|
|
|
|
const sG1 = curve.G1.F.n8*2;
|
|
const G1 = curve.G1;
|
|
const sG2 = curve.G2.F.n8*2;
|
|
const Fr = curve.Fr;
|
|
const n8r = curve.Fr.n8;
|
|
|
|
if (logger) logger.info("Reading r1cs");
|
|
let sR1cs = await binFileUtils.readSection(fdR1cs, sectionsR1cs, 2);
|
|
|
|
const plonkConstraints = new BigArray();
|
|
const plonkAdditions = new BigArray();
|
|
let plonkNVars = r1cs.nVars;
|
|
|
|
const nPublic = r1cs.nOutputs + r1cs.nPubInputs;
|
|
|
|
await processConstraints();
|
|
|
|
const fdZKey = await binFileUtils.createBinFile(zkeyName, "zkey", 1, 14, 1<<22, 1<<24);
|
|
|
|
|
|
if (r1cs.prime != curve.r) {
|
|
if (logger) logger.error("r1cs curve does not match powers of tau ceremony curve");
|
|
return -1;
|
|
}
|
|
|
|
let cirPower = log2(plonkConstraints.length -1) +1;
|
|
if (cirPower < 3) cirPower = 3; // As the t polinomal is n+5 whe need at least a power of 4
|
|
const domainSize = 2 ** cirPower;
|
|
|
|
if (logger) logger.info("Plonk constraints: " + plonkConstraints.length);
|
|
if (cirPower > power) {
|
|
if (logger) logger.error(`circuit too big for this power of tau ceremony. ${plonkConstraints.length} > 2**${power}`);
|
|
return -1;
|
|
}
|
|
|
|
if (!sectionsPTau[12]) {
|
|
if (logger) logger.error("Powers of tau is not prepared.");
|
|
return -1;
|
|
}
|
|
|
|
|
|
const LPoints = new ffjavascript.BigBuffer(domainSize*sG1);
|
|
const o = sectionsPTau[12][0].p + ((2 ** (cirPower)) -1)*sG1;
|
|
await fdPTau.readToBuffer(LPoints, 0, domainSize*sG1, o);
|
|
|
|
const [k1, k2] = getK1K2();
|
|
|
|
const vk = {};
|
|
|
|
|
|
await writeAdditions(3, "Additions");
|
|
await writeWitnessMap(4, 0, "Amap");
|
|
await writeWitnessMap(5, 1, "Bmap");
|
|
await writeWitnessMap(6, 2, "Cmap");
|
|
await writeQMap(7, 3, "Qm");
|
|
await writeQMap(8, 4, "Ql");
|
|
await writeQMap(9, 5, "Qr");
|
|
await writeQMap(10, 6, "Qo");
|
|
await writeQMap(11, 7, "Qc");
|
|
await writeSigma(12, "sigma");
|
|
await writeLs(13, "lagrange polynomials");
|
|
|
|
// Write PTau points
|
|
////////////
|
|
|
|
await binFileUtils.startWriteSection(fdZKey, 14);
|
|
const buffOut = new ffjavascript.BigBuffer((domainSize+6)*sG1);
|
|
await fdPTau.readToBuffer(buffOut, 0, (domainSize+6)*sG1, sectionsPTau[2][0].p);
|
|
await fdZKey.write(buffOut);
|
|
await binFileUtils.endWriteSection(fdZKey);
|
|
|
|
|
|
await writeHeaders();
|
|
|
|
await fdZKey.close();
|
|
await fdR1cs.close();
|
|
await fdPTau.close();
|
|
|
|
if (logger) logger.info("Setup Finished");
|
|
|
|
return ;
|
|
|
|
async function processConstraints() {
|
|
|
|
let r1csPos = 0;
|
|
|
|
function r1cs_readULE32() {
|
|
const buff = sR1cs.slice(r1csPos, r1csPos+4);
|
|
r1csPos += 4;
|
|
const buffV = new DataView(buff.buffer);
|
|
return buffV.getUint32(0, true);
|
|
}
|
|
|
|
function r1cs_readCoef() {
|
|
const res = Fr.fromRprLE(sR1cs.slice(r1csPos, r1csPos+curve.Fr.n8));
|
|
r1csPos += curve.Fr.n8;
|
|
return res;
|
|
}
|
|
|
|
function r1cs_readCoefs() {
|
|
const coefs = [];
|
|
const res = {
|
|
k: curve.Fr.zero
|
|
};
|
|
const nA = r1cs_readULE32();
|
|
for (let i=0; i<nA; i++) {
|
|
const s = r1cs_readULE32();
|
|
const coefp = r1cs_readCoef();
|
|
|
|
if (s==0) {
|
|
res.k = coefp;
|
|
} else {
|
|
coefs.push([s, coefp]);
|
|
}
|
|
}
|
|
|
|
const resCoef = reduceCoef(coefs);
|
|
res.s = resCoef[0];
|
|
res.coef = resCoef[1];
|
|
return res;
|
|
}
|
|
|
|
function reduceCoef(coefs) {
|
|
if (coefs.length == 0) {
|
|
return [0, curve.Fr.zero];
|
|
}
|
|
if (coefs.length == 1) {
|
|
return coefs[0];
|
|
}
|
|
const arr1 = coefs.slice(0, coefs.length >> 1);
|
|
const arr2 = coefs.slice(coefs.length >> 1);
|
|
const coef1 = reduceCoef(arr1);
|
|
const coef2 = reduceCoef(arr2);
|
|
|
|
const sl = coef1[0];
|
|
const sr = coef2[0];
|
|
const so = plonkNVars++;
|
|
const qm = curve.Fr.zero;
|
|
const ql = Fr.neg(coef1[1]);
|
|
const qr = Fr.neg(coef2[1]);
|
|
const qo = curve.Fr.one;
|
|
const qc = curve.Fr.zero;
|
|
|
|
plonkConstraints.push([sl, sr, so, qm, ql, qr, qo, qc]);
|
|
|
|
plonkAdditions.push([sl, sr, coef1[1], coef2[1]]);
|
|
|
|
return [so, curve.Fr.one];
|
|
}
|
|
|
|
for (let s = 1; s <= nPublic ; s++) {
|
|
const sl = s;
|
|
const sr = 0;
|
|
const so = 0;
|
|
const qm = curve.Fr.zero;
|
|
const ql = curve.Fr.one;
|
|
const qr = curve.Fr.zero;
|
|
const qo = curve.Fr.zero;
|
|
const qc = curve.Fr.zero;
|
|
|
|
plonkConstraints.push([sl, sr, so, qm, ql, qr, qo, qc]);
|
|
}
|
|
|
|
for (let c=0; c<r1cs.nConstraints; c++) {
|
|
if ((logger)&&(c%10000 == 0)) logger.debug(`processing constraints: ${c}/${r1cs.nConstraints}`);
|
|
const A = r1cs_readCoefs();
|
|
const B = r1cs_readCoefs();
|
|
const C = r1cs_readCoefs();
|
|
|
|
const sl = A.s;
|
|
const sr = B.s;
|
|
const so = C.s;
|
|
const qm = curve.Fr.mul(A.coef, B.coef);
|
|
const ql = curve.Fr.mul(A.coef, B.k);
|
|
const qr = curve.Fr.mul(A.k, B.coef);
|
|
const qo = curve.Fr.neg(C.coef);
|
|
const qc = curve.Fr.sub(curve.Fr.mul(A.k, B.k) , C.k);
|
|
|
|
plonkConstraints.push([sl, sr, so, qm, ql, qr, qo, qc]);
|
|
}
|
|
|
|
}
|
|
|
|
async function writeWitnessMap(sectionNum, posConstraint, name) {
|
|
await binFileUtils.startWriteSection(fdZKey, sectionNum);
|
|
for (let i=0; i<plonkConstraints.length; i++) {
|
|
await fdZKey.writeULE32(plonkConstraints[i][posConstraint]);
|
|
if ((logger)&&(i%1000000 == 0)) logger.debug(`writing ${name}: ${i}/${plonkConstraints.length}`);
|
|
}
|
|
await binFileUtils.endWriteSection(fdZKey);
|
|
}
|
|
|
|
async function writeQMap(sectionNum, posConstraint, name) {
|
|
let Q = new ffjavascript.BigBuffer(domainSize*n8r);
|
|
for (let i=0; i<plonkConstraints.length; i++) {
|
|
Q.set(plonkConstraints[i][posConstraint], i*n8r);
|
|
if ((logger)&&(i%1000000 == 0)) logger.debug(`writing ${name}: ${i}/${plonkConstraints.length}`);
|
|
}
|
|
await binFileUtils.startWriteSection(fdZKey, sectionNum);
|
|
await writeP4(Q);
|
|
await binFileUtils.endWriteSection(fdZKey);
|
|
Q = await Fr.batchFromMontgomery(Q);
|
|
vk[name]= await curve.G1.multiExpAffine(LPoints, Q, logger, "multiexp "+name);
|
|
}
|
|
|
|
async function writeP4(buff) {
|
|
const q = await Fr.ifft(buff);
|
|
const q4 = new ffjavascript.BigBuffer(domainSize*n8r*4);
|
|
q4.set(q, 0);
|
|
const Q4 = await Fr.fft(q4);
|
|
await fdZKey.write(q);
|
|
await fdZKey.write(Q4);
|
|
}
|
|
|
|
async function writeAdditions(sectionNum, name) {
|
|
await binFileUtils.startWriteSection(fdZKey, sectionNum);
|
|
const buffOut = new Uint8Array((2*4+2*n8r));
|
|
const buffOutV = new DataView(buffOut.buffer);
|
|
for (let i=0; i<plonkAdditions.length; i++) {
|
|
const addition=plonkAdditions[i];
|
|
let o=0;
|
|
buffOutV.setUint32(o, addition[0], true); o+=4;
|
|
buffOutV.setUint32(o, addition[1], true); o+=4;
|
|
// The value is storen in Montgomery. stored = v*R
|
|
// so when montgomery multiplicated by the witness it result = v*R*w/R = v*w
|
|
buffOut.set(addition[2], o); o+= n8r;
|
|
buffOut.set(addition[3], o); o+= n8r;
|
|
await fdZKey.write(buffOut);
|
|
if ((logger)&&(i%1000000 == 0)) logger.debug(`writing ${name}: ${i}/${plonkAdditions.length}`);
|
|
}
|
|
await binFileUtils.endWriteSection(fdZKey);
|
|
}
|
|
|
|
async function writeSigma(sectionNum, name) {
|
|
const sigma = new ffjavascript.BigBuffer(n8r*domainSize*3);
|
|
const lastAparence = new BigArray(plonkNVars);
|
|
const firstPos = new BigArray(plonkNVars);
|
|
let w = Fr.one;
|
|
for (let i=0; i<domainSize;i++) {
|
|
if (i<plonkConstraints.length) {
|
|
buildSigma(plonkConstraints[i][0], i);
|
|
buildSigma(plonkConstraints[i][1], domainSize + i);
|
|
buildSigma(plonkConstraints[i][2], domainSize*2 + i);
|
|
} else {
|
|
buildSigma(0, i);
|
|
buildSigma(0, domainSize + i);
|
|
buildSigma(0, domainSize*2 + i);
|
|
}
|
|
w = Fr.mul(w, Fr.w[cirPower]);
|
|
if ((logger)&&(i%1000000 == 0)) logger.debug(`writing ${name} phase1: ${i}/${plonkConstraints.length}`);
|
|
}
|
|
for (let s=0; s<plonkNVars; s++) {
|
|
if (typeof firstPos[s] !== "undefined") {
|
|
sigma.set(lastAparence[s], firstPos[s]*n8r);
|
|
} else {
|
|
throw new Error("Variable not used");
|
|
}
|
|
if ((logger)&&(s%1000000 == 0)) logger.debug(`writing ${name} phase2: ${s}/${plonkNVars}`);
|
|
}
|
|
await binFileUtils.startWriteSection(fdZKey, sectionNum);
|
|
let S1 = sigma.slice(0, domainSize*n8r);
|
|
await writeP4(S1);
|
|
let S2 = sigma.slice(domainSize*n8r, domainSize*n8r*2);
|
|
await writeP4(S2);
|
|
let S3 = sigma.slice(domainSize*n8r*2, domainSize*n8r*3);
|
|
await writeP4(S3);
|
|
await binFileUtils.endWriteSection(fdZKey);
|
|
|
|
S1 = await Fr.batchFromMontgomery(S1);
|
|
S2 = await Fr.batchFromMontgomery(S2);
|
|
S3 = await Fr.batchFromMontgomery(S3);
|
|
|
|
vk.S1= await curve.G1.multiExpAffine(LPoints, S1, logger, "multiexp S1");
|
|
vk.S2= await curve.G1.multiExpAffine(LPoints, S2, logger, "multiexp S2");
|
|
vk.S3= await curve.G1.multiExpAffine(LPoints, S3, logger, "multiexp S3");
|
|
|
|
function buildSigma(s, p) {
|
|
if (typeof lastAparence[s] === "undefined") {
|
|
firstPos[s] = p;
|
|
} else {
|
|
sigma.set(lastAparence[s], p*n8r);
|
|
}
|
|
let v;
|
|
if (p<domainSize) {
|
|
v = w;
|
|
} else if (p<2*domainSize) {
|
|
v = Fr.mul(w, k1);
|
|
} else {
|
|
v = Fr.mul(w, k2);
|
|
}
|
|
lastAparence[s]=v;
|
|
}
|
|
}
|
|
|
|
async function writeLs(sectionNum, name) {
|
|
await binFileUtils.startWriteSection(fdZKey, sectionNum);
|
|
const l=Math.max(nPublic, 1);
|
|
for (let i=0; i<l; i++) {
|
|
let buff = new ffjavascript.BigBuffer(domainSize*n8r);
|
|
buff.set(Fr.one, i*n8r);
|
|
await writeP4(buff);
|
|
if (logger) logger.debug(`writing ${name} ${i}/${l}`);
|
|
}
|
|
await binFileUtils.endWriteSection(fdZKey);
|
|
}
|
|
|
|
async function writeHeaders() {
|
|
|
|
// Write the header
|
|
///////////
|
|
await binFileUtils.startWriteSection(fdZKey, 1);
|
|
await fdZKey.writeULE32(2); // Plonk
|
|
await binFileUtils.endWriteSection(fdZKey);
|
|
|
|
// Write the Plonk header section
|
|
///////////
|
|
|
|
await binFileUtils.startWriteSection(fdZKey, 2);
|
|
const primeQ = curve.q;
|
|
const n8q = (Math.floor( (ffjavascript.Scalar.bitLength(primeQ) - 1) / 64) +1)*8;
|
|
|
|
const primeR = curve.r;
|
|
const n8r = (Math.floor( (ffjavascript.Scalar.bitLength(primeR) - 1) / 64) +1)*8;
|
|
|
|
await fdZKey.writeULE32(n8q);
|
|
await binFileUtils.writeBigInt(fdZKey, primeQ, n8q);
|
|
await fdZKey.writeULE32(n8r);
|
|
await binFileUtils.writeBigInt(fdZKey, primeR, n8r);
|
|
await fdZKey.writeULE32(plonkNVars); // Total number of bars
|
|
await fdZKey.writeULE32(nPublic); // Total number of public vars (not including ONE)
|
|
await fdZKey.writeULE32(domainSize); // domainSize
|
|
await fdZKey.writeULE32(plonkAdditions.length); // domainSize
|
|
await fdZKey.writeULE32(plonkConstraints.length);
|
|
|
|
await fdZKey.write(k1);
|
|
await fdZKey.write(k2);
|
|
|
|
await fdZKey.write(G1.toAffine(vk.Qm));
|
|
await fdZKey.write(G1.toAffine(vk.Ql));
|
|
await fdZKey.write(G1.toAffine(vk.Qr));
|
|
await fdZKey.write(G1.toAffine(vk.Qo));
|
|
await fdZKey.write(G1.toAffine(vk.Qc));
|
|
|
|
await fdZKey.write(G1.toAffine(vk.S1));
|
|
await fdZKey.write(G1.toAffine(vk.S2));
|
|
await fdZKey.write(G1.toAffine(vk.S3));
|
|
|
|
let bX_2;
|
|
bX_2 = await fdPTau.read(sG2, sectionsPTau[3][0].p + sG2);
|
|
await fdZKey.write(bX_2);
|
|
|
|
await binFileUtils.endWriteSection(fdZKey);
|
|
}
|
|
|
|
function getK1K2() {
|
|
let k1 = Fr.two;
|
|
while (isIncluded(k1, [], cirPower)) Fr.add(k1, Fr.one);
|
|
let k2 = Fr.add(k1, Fr.one);
|
|
while (isIncluded(k2, [k1], cirPower)) Fr.add(k2, Fr.one);
|
|
return [k1, k2];
|
|
|
|
|
|
function isIncluded(k, kArr, pow) {
|
|
const domainSize= 2**pow;
|
|
let w = Fr.one;
|
|
for (let i=0; i<domainSize; i++) {
|
|
if (Fr.eq(k, w)) return true;
|
|
for (let j=0; j<kArr.length; j++) {
|
|
if (Fr.eq(k, Fr.mul(kArr[j], w))) return true;
|
|
}
|
|
w = Fr.mul(w, Fr.w[pow]);
|
|
}
|
|
return false;
|
|
}
|
|
}
|
|
}
|
|
|
|
/*
|
|
Copyright 2021 0kims association.
|
|
|
|
This file is part of snarkjs.
|
|
|
|
snarkjs is a free software: you can redistribute it and/or
|
|
modify it under the terms of the GNU General Public License as published by the
|
|
Free Software Foundation, either version 3 of the License, or (at your option)
|
|
any later version.
|
|
|
|
snarkjs is distributed in the hope that it will be useful,
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
|
|
more details.
|
|
|
|
You should have received a copy of the GNU General Public License along with
|
|
snarkjs. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
const {stringifyBigInts} = ffjavascript.utils;
|
|
const { keccak256: keccak256$1 } = jsSha3__default["default"];
|
|
|
|
async function plonk16Prove(zkeyFileName, witnessFileName, logger) {
|
|
const {fd: fdWtns, sections: sectionsWtns} = await binFileUtils__namespace.readBinFile(witnessFileName, "wtns", 2, 1<<25, 1<<23);
|
|
|
|
const wtns = await readHeader(fdWtns, sectionsWtns);
|
|
|
|
const {fd: fdZKey, sections: sectionsZKey} = await binFileUtils__namespace.readBinFile(zkeyFileName, "zkey", 2, 1<<25, 1<<23);
|
|
|
|
const zkey = await readHeader$1(fdZKey, sectionsZKey);
|
|
if (zkey.protocol != "plonk") {
|
|
throw new Error("zkey file is not groth16");
|
|
}
|
|
|
|
if (!ffjavascript.Scalar.eq(zkey.r, wtns.q)) {
|
|
throw new Error("Curve of the witness does not match the curve of the proving key");
|
|
}
|
|
|
|
if (wtns.nWitness != zkey.nVars -zkey.nAdditions) {
|
|
throw new Error(`Invalid witness length. Circuit: ${zkey.nVars}, witness: ${wtns.nWitness}, ${zkey.nAdditions}`);
|
|
}
|
|
|
|
const curve = await getCurveFromQ(zkey.q);
|
|
const Fr = curve.Fr;
|
|
const G1 = curve.G1;
|
|
const n8r = curve.Fr.n8;
|
|
|
|
if (logger) logger.debug("Reading Wtns");
|
|
const buffWitness = await binFileUtils__namespace.readSection(fdWtns, sectionsWtns, 2);
|
|
// First element in plonk is not used and can be any value. (But always the same).
|
|
// We set it to zero to go faster in the exponentiations.
|
|
buffWitness.set(Fr.zero, 0);
|
|
const buffInternalWitness = new ffjavascript.BigBuffer(n8r*zkey.nAdditions);
|
|
|
|
await calculateAdditions();
|
|
|
|
let A,B,C,Z;
|
|
let A4, B4, C4, Z4;
|
|
let pol_a,pol_b,pol_c, pol_z, pol_t, pol_r;
|
|
let proof = {};
|
|
|
|
const sigmaBuff = new ffjavascript.BigBuffer(zkey.domainSize*n8r*4*3);
|
|
let o = sectionsZKey[12][0].p + zkey.domainSize*n8r;
|
|
await fdZKey.readToBuffer(sigmaBuff, 0 , zkey.domainSize*n8r*4, o);
|
|
o += zkey.domainSize*n8r*5;
|
|
await fdZKey.readToBuffer(sigmaBuff, zkey.domainSize*n8r*4 , zkey.domainSize*n8r*4, o);
|
|
o += zkey.domainSize*n8r*5;
|
|
await fdZKey.readToBuffer(sigmaBuff, zkey.domainSize*n8r*8 , zkey.domainSize*n8r*4, o);
|
|
|
|
const pol_s1 = new ffjavascript.BigBuffer(zkey.domainSize*n8r);
|
|
await fdZKey.readToBuffer(pol_s1, 0 , zkey.domainSize*n8r, sectionsZKey[12][0].p);
|
|
|
|
const pol_s2 = new ffjavascript.BigBuffer(zkey.domainSize*n8r);
|
|
await fdZKey.readToBuffer(pol_s2, 0 , zkey.domainSize*n8r, sectionsZKey[12][0].p + 5*zkey.domainSize*n8r);
|
|
|
|
const PTau = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 14);
|
|
|
|
|
|
const ch = {};
|
|
|
|
await round1();
|
|
await round2();
|
|
await round3();
|
|
await round4();
|
|
await round5();
|
|
|
|
|
|
///////////////////////
|
|
// Final adjustments //
|
|
///////////////////////
|
|
|
|
proof.protocol = "plonk";
|
|
proof.curve = curve.name;
|
|
|
|
await fdZKey.close();
|
|
await fdWtns.close();
|
|
|
|
let publicSignals = [];
|
|
|
|
for (let i=1; i<= zkey.nPublic; i++) {
|
|
const pub = buffWitness.slice(i*Fr.n8, i*Fr.n8+Fr.n8);
|
|
publicSignals.push(ffjavascript.Scalar.fromRprLE(pub));
|
|
}
|
|
|
|
proof.A = G1.toObject(proof.A);
|
|
proof.B = G1.toObject(proof.B);
|
|
proof.C = G1.toObject(proof.C);
|
|
proof.Z = G1.toObject(proof.Z);
|
|
|
|
proof.T1 = G1.toObject(proof.T1);
|
|
proof.T2 = G1.toObject(proof.T2);
|
|
proof.T3 = G1.toObject(proof.T3);
|
|
|
|
proof.eval_a = Fr.toObject(proof.eval_a);
|
|
proof.eval_b = Fr.toObject(proof.eval_b);
|
|
proof.eval_c = Fr.toObject(proof.eval_c);
|
|
proof.eval_s1 = Fr.toObject(proof.eval_s1);
|
|
proof.eval_s2 = Fr.toObject(proof.eval_s2);
|
|
proof.eval_zw = Fr.toObject(proof.eval_zw);
|
|
proof.eval_t = Fr.toObject(proof.eval_t);
|
|
proof.eval_r = Fr.toObject(proof.eval_r);
|
|
|
|
proof.Wxi = G1.toObject(proof.Wxi);
|
|
proof.Wxiw = G1.toObject(proof.Wxiw);
|
|
|
|
delete proof.eval_t;
|
|
|
|
proof = stringifyBigInts(proof);
|
|
publicSignals = stringifyBigInts(publicSignals);
|
|
|
|
return {proof, publicSignals};
|
|
|
|
async function calculateAdditions() {
|
|
const additionsBuff = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 3);
|
|
|
|
const sSum = 8+curve.Fr.n8*2;
|
|
|
|
for (let i=0; i<zkey.nAdditions; i++) {
|
|
const ai= readUInt32(additionsBuff, i*sSum);
|
|
const bi= readUInt32(additionsBuff, i*sSum+4);
|
|
const ac= additionsBuff.slice(i*sSum+8, i*sSum+8+n8r);
|
|
const bc= additionsBuff.slice(i*sSum+8+n8r, i*sSum+8+n8r*2);
|
|
const aw= getWitness(ai);
|
|
const bw= getWitness(bi);
|
|
|
|
const r = curve.Fr.add(
|
|
curve.Fr.mul(ac, aw),
|
|
curve.Fr.mul(bc, bw)
|
|
);
|
|
buffInternalWitness.set(r, n8r*i);
|
|
}
|
|
|
|
}
|
|
|
|
async function buildABC() {
|
|
let A = new ffjavascript.BigBuffer(zkey.domainSize * n8r);
|
|
let B = new ffjavascript.BigBuffer(zkey.domainSize * n8r);
|
|
let C = new ffjavascript.BigBuffer(zkey.domainSize * n8r);
|
|
|
|
const aMap = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 4);
|
|
const bMap = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 5);
|
|
const cMap = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 6);
|
|
|
|
for (let i=0; i<zkey.nConstrains; i++) {
|
|
const iA = readUInt32(aMap, i*4);
|
|
A.set(getWitness(iA), i*n8r);
|
|
const iB = readUInt32(bMap, i*4);
|
|
B.set(getWitness(iB), i*n8r);
|
|
const iC = readUInt32(cMap, i*4);
|
|
C.set(getWitness(iC), i*n8r);
|
|
}
|
|
|
|
A = await Fr.batchToMontgomery(A);
|
|
B = await Fr.batchToMontgomery(B);
|
|
C = await Fr.batchToMontgomery(C);
|
|
|
|
return [A,B,C];
|
|
}
|
|
|
|
function readUInt32(b, o) {
|
|
const buff = b.slice(o, o+4);
|
|
const buffV = new DataView(buff.buffer, buff.byteOffset, buff.byteLength);
|
|
return buffV.getUint32(0, true);
|
|
}
|
|
|
|
function getWitness(idx) {
|
|
if (idx < zkey.nVars-zkey.nAdditions) {
|
|
return buffWitness.slice(idx*n8r, idx*n8r+n8r);
|
|
} else if (idx < zkey.nVars) {
|
|
return buffInternalWitness.slice((idx - (zkey.nVars-zkey.nAdditions))*n8r, (idx-(zkey.nVars-zkey.nAdditions))*n8r + n8r);
|
|
} else {
|
|
return curve.Fr.zero;
|
|
}
|
|
}
|
|
|
|
async function round1() {
|
|
ch.b = [];
|
|
for (let i=1; i<=9; i++) {
|
|
ch.b[i] = curve.Fr.random();
|
|
}
|
|
|
|
[A, B, C] = await buildABC();
|
|
|
|
[pol_a, A4] = await to4T(A, [ch.b[2], ch.b[1]]);
|
|
[pol_b, B4] = await to4T(B, [ch.b[4], ch.b[3]]);
|
|
[pol_c, C4] = await to4T(C, [ch.b[6], ch.b[5]]);
|
|
|
|
|
|
proof.A = await expTau(pol_a, "multiexp A");
|
|
proof.B = await expTau(pol_b, "multiexp B");
|
|
proof.C = await expTau(pol_c, "multiexp C");
|
|
}
|
|
|
|
async function round2() {
|
|
|
|
const transcript1 = new Uint8Array(G1.F.n8*2*3);
|
|
G1.toRprUncompressed(transcript1, 0, proof.A);
|
|
G1.toRprUncompressed(transcript1, G1.F.n8*2, proof.B);
|
|
G1.toRprUncompressed(transcript1, G1.F.n8*4, proof.C);
|
|
|
|
ch.beta = hashToFr(transcript1);
|
|
if (logger) logger.debug("beta: " + Fr.toString(ch.beta));
|
|
|
|
const transcript2 = new Uint8Array(n8r);
|
|
Fr.toRprBE(transcript2, 0, ch.beta);
|
|
ch.gamma = hashToFr(transcript2);
|
|
if (logger) logger.debug("gamma: " + Fr.toString(ch.gamma));
|
|
|
|
let numArr = new ffjavascript.BigBuffer(Fr.n8*zkey.domainSize);
|
|
let denArr = new ffjavascript.BigBuffer(Fr.n8*zkey.domainSize);
|
|
|
|
numArr.set(Fr.one, 0);
|
|
denArr.set(Fr.one, 0);
|
|
|
|
let w = Fr.one;
|
|
for (let i=0; i<zkey.domainSize; i++) {
|
|
let n1 = A.slice(i*n8r, (i+1)*n8r);
|
|
n1 = Fr.add( n1, Fr.mul(ch.beta, w) );
|
|
n1 = Fr.add( n1, ch.gamma );
|
|
|
|
let n2 = B.slice(i*n8r, (i+1)*n8r);
|
|
n2 = Fr.add( n2, Fr.mul(zkey.k1, Fr.mul(ch.beta, w) ));
|
|
n2 = Fr.add( n2, ch.gamma );
|
|
|
|
let n3 = C.slice(i*n8r, (i+1)*n8r);
|
|
n3 = Fr.add( n3, Fr.mul(zkey.k2, Fr.mul(ch.beta, w) ));
|
|
n3 = Fr.add( n3, ch.gamma );
|
|
|
|
const num = Fr.mul(n1, Fr.mul(n2, n3));
|
|
|
|
let d1 = A.slice(i*n8r, (i+1)*n8r);
|
|
d1 = Fr.add(d1, Fr.mul( sigmaBuff.slice(i*n8r*4, i*n8r*4 + n8r) , ch.beta));
|
|
d1 = Fr.add(d1, ch.gamma);
|
|
|
|
let d2 = B.slice(i*n8r, (i+1)*n8r);
|
|
d2 = Fr.add(d2, Fr.mul( sigmaBuff.slice((zkey.domainSize + i)*4*n8r, (zkey.domainSize + i)*4*n8r+n8r) , ch.beta));
|
|
d2 = Fr.add(d2, ch.gamma);
|
|
|
|
let d3 = C.slice(i*n8r, (i+1)*n8r);
|
|
d3 = Fr.add(d3, Fr.mul( sigmaBuff.slice((zkey.domainSize*2 + i)*4*n8r, (zkey.domainSize*2 + i)*4*n8r + n8r) , ch.beta));
|
|
d3 = Fr.add(d3, ch.gamma);
|
|
|
|
const den = Fr.mul(d1, Fr.mul(d2, d3));
|
|
|
|
numArr.set(
|
|
Fr.mul(
|
|
numArr.slice(i*n8r,(i+1)*n8r) ,
|
|
num
|
|
),
|
|
((i+1)%zkey.domainSize)*n8r
|
|
);
|
|
|
|
denArr.set(
|
|
Fr.mul(
|
|
denArr.slice(i*n8r,(i+1)*n8r) ,
|
|
den
|
|
),
|
|
((i+1)%zkey.domainSize)*n8r
|
|
);
|
|
|
|
w = Fr.mul(w, Fr.w[zkey.power]);
|
|
}
|
|
|
|
denArr = await Fr.batchInverse(denArr);
|
|
|
|
// TODO: Do it in assembly and in parallel
|
|
for (let i=0; i<zkey.domainSize; i++) {
|
|
numArr.set( Fr.mul( numArr.slice(i*n8r, (i+1)*n8r), denArr.slice(i*n8r, (i+1)*n8r) ) ,i*n8r);
|
|
}
|
|
|
|
if (!Fr.eq(numArr.slice(0, n8r), Fr.one)) {
|
|
throw new Error("Copy constraints does not match");
|
|
}
|
|
|
|
Z = numArr;
|
|
|
|
[pol_z, Z4] = await to4T(Z, [ch.b[9], ch.b[8], ch.b[7]]);
|
|
|
|
proof.Z = await expTau( pol_z, "multiexp Z");
|
|
}
|
|
|
|
async function round3() {
|
|
|
|
/*
|
|
async function checkDegree(P) {
|
|
const p = await curve.Fr.ifft(P);
|
|
let deg = (P.byteLength/n8r)-1;
|
|
while ((deg>0)&&(Fr.isZero(p.slice(deg*n8r, deg*n8r+n8r)))) deg--;
|
|
return deg;
|
|
}
|
|
|
|
function printPol(P) {
|
|
const n=(P.byteLength/n8r);
|
|
console.log("[");
|
|
for (let i=0; i<n; i++) {
|
|
console.log(Fr.toString(P.slice(i*n8r, i*n8r+n8r)));
|
|
}
|
|
console.log("]");
|
|
}
|
|
*/
|
|
|
|
if (logger) logger.debug("phse3: Reading QM4");
|
|
const QM4 = new ffjavascript.BigBuffer(zkey.domainSize*4*n8r);
|
|
await fdZKey.readToBuffer(QM4, 0 , zkey.domainSize*n8r*4, sectionsZKey[7][0].p + zkey.domainSize*n8r);
|
|
|
|
if (logger) logger.debug("phse3: Reading QL4");
|
|
const QL4 = new ffjavascript.BigBuffer(zkey.domainSize*4*n8r);
|
|
await fdZKey.readToBuffer(QL4, 0 , zkey.domainSize*n8r*4, sectionsZKey[8][0].p + zkey.domainSize*n8r);
|
|
|
|
if (logger) logger.debug("phse3: Reading QR4");
|
|
const QR4 = new ffjavascript.BigBuffer(zkey.domainSize*4*n8r);
|
|
await fdZKey.readToBuffer(QR4, 0 , zkey.domainSize*n8r*4, sectionsZKey[9][0].p + zkey.domainSize*n8r);
|
|
|
|
if (logger) logger.debug("phse3: Reading QO4");
|
|
const QO4 = new ffjavascript.BigBuffer(zkey.domainSize*4*n8r);
|
|
await fdZKey.readToBuffer(QO4, 0 , zkey.domainSize*n8r*4, sectionsZKey[10][0].p + zkey.domainSize*n8r);
|
|
|
|
if (logger) logger.debug("phse3: Reading QC4");
|
|
const QC4 = new ffjavascript.BigBuffer(zkey.domainSize*4*n8r);
|
|
await fdZKey.readToBuffer(QC4, 0 , zkey.domainSize*n8r*4, sectionsZKey[11][0].p + zkey.domainSize*n8r);
|
|
|
|
const lPols = await binFileUtils__namespace.readSection(fdZKey, sectionsZKey, 13);
|
|
|
|
const transcript3 = new Uint8Array(G1.F.n8*2);
|
|
G1.toRprUncompressed(transcript3, 0, proof.Z);
|
|
|
|
ch.alpha = hashToFr(transcript3);
|
|
|
|
if (logger) logger.debug("alpha: " + Fr.toString(ch.alpha));
|
|
|
|
|
|
const Z1 = [
|
|
Fr.zero,
|
|
Fr.add(Fr.e(-1), Fr.w[2]),
|
|
Fr.e(-2),
|
|
Fr.sub(Fr.e(-1), Fr.w[2]),
|
|
];
|
|
|
|
const Z2 = [
|
|
Fr.zero,
|
|
Fr.add(Fr.zero, Fr.mul(Fr.e(-2), Fr.w[2])),
|
|
Fr.e(4),
|
|
Fr.sub(Fr.zero, Fr.mul(Fr.e(-2), Fr.w[2])),
|
|
];
|
|
|
|
const Z3 = [
|
|
Fr.zero,
|
|
Fr.add(Fr.e(2), Fr.mul(Fr.e(2), Fr.w[2])),
|
|
Fr.e(-8),
|
|
Fr.sub(Fr.e(2), Fr.mul(Fr.e(2), Fr.w[2])),
|
|
];
|
|
|
|
const T = new ffjavascript.BigBuffer(zkey.domainSize*4*n8r);
|
|
const Tz = new ffjavascript.BigBuffer(zkey.domainSize*4*n8r);
|
|
|
|
let w = Fr.one;
|
|
for (let i=0; i<zkey.domainSize*4; i++) {
|
|
if ((i%4096 == 0)&&(logger)) logger.debug(`calculating t ${i}/${zkey.domainSize*4}`);
|
|
|
|
const a = A4.slice(i*n8r, i*n8r+n8r);
|
|
const b = B4.slice(i*n8r, i*n8r+n8r);
|
|
const c = C4.slice(i*n8r, i*n8r+n8r);
|
|
const z = Z4.slice(i*n8r, i*n8r+n8r);
|
|
const zw = Z4.slice(((i+zkey.domainSize*4+4)%(zkey.domainSize*4)) *n8r, ((i+zkey.domainSize*4+4)%(zkey.domainSize*4)) *n8r +n8r);
|
|
const qm = QM4.slice(i*n8r, i*n8r+n8r);
|
|
const ql = QL4.slice(i*n8r, i*n8r+n8r);
|
|
const qr = QR4.slice(i*n8r, i*n8r+n8r);
|
|
const qo = QO4.slice(i*n8r, i*n8r+n8r);
|
|
const qc = QC4.slice(i*n8r, i*n8r+n8r);
|
|
const s1 = sigmaBuff.slice(i*n8r, i*n8r+n8r);
|
|
const s2 = sigmaBuff.slice((i+zkey.domainSize*4)*n8r, (i+zkey.domainSize*4)*n8r+n8r);
|
|
const s3 = sigmaBuff.slice((i+zkey.domainSize*8)*n8r, (i+zkey.domainSize*8)*n8r+n8r);
|
|
const ap = Fr.add(ch.b[2], Fr.mul(ch.b[1], w));
|
|
const bp = Fr.add(ch.b[4], Fr.mul(ch.b[3], w));
|
|
const cp = Fr.add(ch.b[6], Fr.mul(ch.b[5], w));
|
|
const w2 = Fr.square(w);
|
|
const zp = Fr.add(Fr.add(Fr.mul(ch.b[7], w2), Fr.mul(ch.b[8], w)), ch.b[9]);
|
|
const wW = Fr.mul(w, Fr.w[zkey.power]);
|
|
const wW2 = Fr.square(wW);
|
|
const zWp = Fr.add(Fr.add(Fr.mul(ch.b[7], wW2), Fr.mul(ch.b[8], wW)), ch.b[9]);
|
|
|
|
let pl = Fr.zero;
|
|
for (let j=0; j<zkey.nPublic; j++) {
|
|
pl = Fr.sub(pl, Fr.mul(
|
|
lPols.slice( (j*5*zkey.domainSize+ zkey.domainSize+ i)*n8r, (j*5*zkey.domainSize+ zkey.domainSize + i+1)*n8r),
|
|
A.slice(j*n8r, (j+1)*n8r)
|
|
));
|
|
}
|
|
|
|
let [e1, e1z] = mul2(a, b, ap, bp, i%4);
|
|
e1 = Fr.mul(e1, qm);
|
|
e1z = Fr.mul(e1z, qm);
|
|
|
|
e1 = Fr.add(e1, Fr.mul(a, ql));
|
|
e1z = Fr.add(e1z, Fr.mul(ap, ql));
|
|
|
|
e1 = Fr.add(e1, Fr.mul(b, qr));
|
|
e1z = Fr.add(e1z, Fr.mul(bp, qr));
|
|
|
|
e1 = Fr.add(e1, Fr.mul(c, qo));
|
|
e1z = Fr.add(e1z, Fr.mul(cp, qo));
|
|
|
|
e1 = Fr.add(e1, pl);
|
|
e1 = Fr.add(e1, qc);
|
|
|
|
const betaw = Fr.mul(ch.beta, w);
|
|
let e2a =a;
|
|
e2a = Fr.add(e2a, betaw);
|
|
e2a = Fr.add(e2a, ch.gamma);
|
|
|
|
let e2b =b;
|
|
e2b = Fr.add(e2b, Fr.mul(betaw, zkey.k1));
|
|
e2b = Fr.add(e2b, ch.gamma);
|
|
|
|
let e2c =c;
|
|
e2c = Fr.add(e2c, Fr.mul(betaw, zkey.k2));
|
|
e2c = Fr.add(e2c, ch.gamma);
|
|
|
|
let e2d = z;
|
|
|
|
let [e2, e2z] = mul4(e2a, e2b, e2c, e2d, ap, bp, cp, zp, i%4);
|
|
e2 = Fr.mul(e2, ch.alpha);
|
|
e2z = Fr.mul(e2z, ch.alpha);
|
|
|
|
let e3a = a;
|
|
e3a = Fr.add(e3a, Fr.mul(ch.beta, s1));
|
|
e3a = Fr.add(e3a, ch.gamma);
|
|
|
|
let e3b = b;
|
|
e3b = Fr.add(e3b, Fr.mul(ch.beta,s2));
|
|
e3b = Fr.add(e3b, ch.gamma);
|
|
|
|
let e3c = c;
|
|
e3c = Fr.add(e3c, Fr.mul(ch.beta,s3));
|
|
e3c = Fr.add(e3c, ch.gamma);
|
|
|
|
let e3d = zw;
|
|
let [e3, e3z] = mul4(e3a, e3b, e3c, e3d, ap, bp, cp, zWp, i%4);
|
|
|
|
e3 = Fr.mul(e3, ch.alpha);
|
|
e3z = Fr.mul(e3z, ch.alpha);
|
|
|
|
let e4 = Fr.sub(z, Fr.one);
|
|
e4 = Fr.mul(e4, lPols.slice( (zkey.domainSize + i)*n8r, (zkey.domainSize+i+1)*n8r));
|
|
e4 = Fr.mul(e4, Fr.mul(ch.alpha, ch.alpha));
|
|
|
|
let e4z = Fr.mul(zp, lPols.slice( (zkey.domainSize + i)*n8r, (zkey.domainSize+i+1)*n8r));
|
|
e4z = Fr.mul(e4z, Fr.mul(ch.alpha, ch.alpha));
|
|
|
|
let e = Fr.add(Fr.sub(Fr.add(e1, e2), e3), e4);
|
|
let ez = Fr.add(Fr.sub(Fr.add(e1z, e2z), e3z), e4z);
|
|
|
|
T.set(e, i*n8r);
|
|
Tz.set(ez, i*n8r);
|
|
|
|
w = Fr.mul(w, Fr.w[zkey.power+2]);
|
|
}
|
|
|
|
if (logger) logger.debug("ifft T");
|
|
let t = await Fr.ifft(T);
|
|
|
|
if (logger) logger.debug("dividing T/Z");
|
|
for (let i=0; i<zkey.domainSize; i++) {
|
|
t.set(Fr.neg(t.slice(i*n8r, i*n8r+n8r)), i*n8r);
|
|
}
|
|
|
|
for (let i=zkey.domainSize; i<zkey.domainSize*4; i++) {
|
|
const a = Fr.sub(
|
|
t.slice((i-zkey.domainSize)*n8r, (i-zkey.domainSize)*n8r + n8r),
|
|
t.slice(i*n8r, i*n8r+n8r)
|
|
);
|
|
t.set(a, i*n8r);
|
|
if (i > (zkey.domainSize*3 -4) ) {
|
|
if (!Fr.isZero(a)) {
|
|
throw new Error("T Polynomial is not divisible");
|
|
}
|
|
}
|
|
}
|
|
|
|
if (logger) logger.debug("ifft Tz");
|
|
const tz = await Fr.ifft(Tz);
|
|
for (let i=0; i<zkey.domainSize*4; i++) {
|
|
const a = tz.slice(i*n8r, (i+1)*n8r);
|
|
if (i > (zkey.domainSize*3 +5) ) {
|
|
if (!Fr.isZero(a)) {
|
|
throw new Error("Tz Polynomial is not well calculated");
|
|
}
|
|
} else {
|
|
t.set(
|
|
Fr.add(
|
|
t.slice(i*n8r, (i+1)*n8r),
|
|
a
|
|
),
|
|
i*n8r
|
|
);
|
|
}
|
|
}
|
|
|
|
pol_t = t.slice(0, (zkey.domainSize*3+6)*n8r);
|
|
|
|
proof.T1 = await expTau( t.slice(0, zkey.domainSize*n8r) , "multiexp T1");
|
|
proof.T2 = await expTau( t.slice(zkey.domainSize*n8r, zkey.domainSize*2*n8r) , "multiexp T2");
|
|
proof.T3 = await expTau( t.slice(zkey.domainSize*2*n8r, (zkey.domainSize*3+6)*n8r) , "multiexp T3");
|
|
|
|
function mul2(a,b, ap, bp, p) {
|
|
let r, rz;
|
|
|
|
|
|
const a_b = Fr.mul(a,b);
|
|
const a_bp = Fr.mul(a,bp);
|
|
const ap_b = Fr.mul(ap,b);
|
|
const ap_bp = Fr.mul(ap,bp);
|
|
|
|
r = a_b;
|
|
|
|
let a0 = Fr.add(a_bp, ap_b);
|
|
|
|
let a1 = ap_bp;
|
|
|
|
rz = a0;
|
|
if (p) {
|
|
rz = Fr.add(rz, Fr.mul(Z1[p], a1));
|
|
}
|
|
|
|
return [r, rz];
|
|
}
|
|
|
|
function mul4(a,b,c,d, ap, bp, cp, dp, p) {
|
|
let r, rz;
|
|
|
|
|
|
const a_b = Fr.mul(a,b);
|
|
const a_bp = Fr.mul(a,bp);
|
|
const ap_b = Fr.mul(ap,b);
|
|
const ap_bp = Fr.mul(ap,bp);
|
|
|
|
const c_d = Fr.mul(c,d);
|
|
const c_dp = Fr.mul(c,dp);
|
|
const cp_d = Fr.mul(cp,d);
|
|
const cp_dp = Fr.mul(cp,dp);
|
|
|
|
r = Fr.mul(a_b, c_d);
|
|
|
|
let a0 = Fr.mul(ap_b, c_d);
|
|
a0 = Fr.add(a0, Fr.mul(a_bp, c_d));
|
|
a0 = Fr.add(a0, Fr.mul(a_b, cp_d));
|
|
a0 = Fr.add(a0, Fr.mul(a_b, c_dp));
|
|
|
|
let a1 = Fr.mul(ap_bp, c_d);
|
|
a1 = Fr.add(a1, Fr.mul(ap_b, cp_d));
|
|
a1 = Fr.add(a1, Fr.mul(ap_b, c_dp));
|
|
a1 = Fr.add(a1, Fr.mul(a_bp, cp_d));
|
|
a1 = Fr.add(a1, Fr.mul(a_bp, c_dp));
|
|
a1 = Fr.add(a1, Fr.mul(a_b, cp_dp));
|
|
|
|
let a2 = Fr.mul(a_bp, cp_dp);
|
|
a2 = Fr.add(a2, Fr.mul(ap_b, cp_dp));
|
|
a2 = Fr.add(a2, Fr.mul(ap_bp, c_dp));
|
|
a2 = Fr.add(a2, Fr.mul(ap_bp, cp_d));
|
|
|
|
let a3 = Fr.mul(ap_bp, cp_dp);
|
|
|
|
rz = a0;
|
|
if (p) {
|
|
rz = Fr.add(rz, Fr.mul(Z1[p], a1));
|
|
rz = Fr.add(rz, Fr.mul(Z2[p], a2));
|
|
rz = Fr.add(rz, Fr.mul(Z3[p], a3));
|
|
}
|
|
|
|
return [r, rz];
|
|
}
|
|
}
|
|
|
|
async function round4() {
|
|
const pol_qm = new ffjavascript.BigBuffer(zkey.domainSize*n8r);
|
|
await fdZKey.readToBuffer(pol_qm, 0 , zkey.domainSize*n8r, sectionsZKey[7][0].p);
|
|
|
|
const pol_ql = new ffjavascript.BigBuffer(zkey.domainSize*n8r);
|
|
await fdZKey.readToBuffer(pol_ql, 0 , zkey.domainSize*n8r, sectionsZKey[8][0].p);
|
|
|
|
const pol_qr = new ffjavascript.BigBuffer(zkey.domainSize*n8r);
|
|
await fdZKey.readToBuffer(pol_qr, 0 , zkey.domainSize*n8r, sectionsZKey[9][0].p);
|
|
|
|
const pol_qo = new ffjavascript.BigBuffer(zkey.domainSize*n8r);
|
|
await fdZKey.readToBuffer(pol_qo, 0 , zkey.domainSize*n8r, sectionsZKey[10][0].p);
|
|
|
|
const pol_qc = new ffjavascript.BigBuffer(zkey.domainSize*n8r);
|
|
await fdZKey.readToBuffer(pol_qc, 0 , zkey.domainSize*n8r, sectionsZKey[11][0].p);
|
|
|
|
const pol_s3 = new ffjavascript.BigBuffer(zkey.domainSize*n8r);
|
|
await fdZKey.readToBuffer(pol_s3, 0 , zkey.domainSize*n8r, sectionsZKey[12][0].p + 10*zkey.domainSize*n8r);
|
|
|
|
const transcript4 = new Uint8Array(G1.F.n8*2*3);
|
|
G1.toRprUncompressed(transcript4, 0, proof.T1);
|
|
G1.toRprUncompressed(transcript4, G1.F.n8*2, proof.T2);
|
|
G1.toRprUncompressed(transcript4, G1.F.n8*4, proof.T3);
|
|
ch.xi = hashToFr(transcript4);
|
|
|
|
if (logger) logger.debug("xi: " + Fr.toString(ch.xi));
|
|
|
|
proof.eval_a = evalPol(pol_a, ch.xi);
|
|
proof.eval_b = evalPol(pol_b, ch.xi);
|
|
proof.eval_c = evalPol(pol_c, ch.xi);
|
|
proof.eval_s1 = evalPol(pol_s1, ch.xi);
|
|
proof.eval_s2 = evalPol(pol_s2, ch.xi);
|
|
proof.eval_t = evalPol(pol_t, ch.xi);
|
|
proof.eval_zw = evalPol(pol_z, Fr.mul(ch.xi, Fr.w[zkey.power]));
|
|
|
|
const coef_ab = Fr.mul(proof.eval_a, proof.eval_b);
|
|
|
|
let e2a = proof.eval_a;
|
|
const betaxi = Fr.mul(ch.beta, ch.xi);
|
|
e2a = Fr.add( e2a, betaxi);
|
|
e2a = Fr.add( e2a, ch.gamma);
|
|
|
|
let e2b = proof.eval_b;
|
|
e2b = Fr.add( e2b, Fr.mul(betaxi, zkey.k1));
|
|
e2b = Fr.add( e2b, ch.gamma);
|
|
|
|
let e2c = proof.eval_c;
|
|
e2c = Fr.add( e2c, Fr.mul(betaxi, zkey.k2));
|
|
e2c = Fr.add( e2c, ch.gamma);
|
|
|
|
const e2 = Fr.mul(Fr.mul(Fr.mul(e2a, e2b), e2c), ch.alpha);
|
|
|
|
let e3a = proof.eval_a;
|
|
e3a = Fr.add( e3a, Fr.mul(ch.beta, proof.eval_s1));
|
|
e3a = Fr.add( e3a, ch.gamma);
|
|
|
|
let e3b = proof.eval_b;
|
|
e3b = Fr.add( e3b, Fr.mul(ch.beta, proof.eval_s2));
|
|
e3b = Fr.add( e3b, ch.gamma);
|
|
|
|
let e3 = Fr.mul(e3a, e3b);
|
|
e3 = Fr.mul(e3, ch.beta);
|
|
e3 = Fr.mul(e3, proof.eval_zw);
|
|
e3 = Fr.mul(e3, ch.alpha);
|
|
|
|
ch.xim= ch.xi;
|
|
for (let i=0; i<zkey.power; i++) ch.xim = Fr.mul(ch.xim, ch.xim);
|
|
const eval_l1 = Fr.div(
|
|
Fr.sub(ch.xim, Fr.one),
|
|
Fr.mul(Fr.sub(ch.xi, Fr.one), Fr.e(zkey.domainSize))
|
|
);
|
|
|
|
const e4 = Fr.mul(eval_l1, Fr.mul(ch.alpha, ch.alpha));
|
|
|
|
const coefs3 = e3;
|
|
const coefz = Fr.add(e2, e4);
|
|
|
|
pol_r = new ffjavascript.BigBuffer((zkey.domainSize+3)*n8r);
|
|
|
|
for (let i = 0; i<zkey.domainSize+3; i++) {
|
|
let v = Fr.mul(coefz, pol_z.slice(i*n8r,(i+1)*n8r));
|
|
if (i<zkey.domainSize) {
|
|
v = Fr.add(v, Fr.mul(coef_ab, pol_qm.slice(i*n8r,(i+1)*n8r)));
|
|
v = Fr.add(v, Fr.mul(proof.eval_a, pol_ql.slice(i*n8r,(i+1)*n8r)));
|
|
v = Fr.add(v, Fr.mul(proof.eval_b, pol_qr.slice(i*n8r,(i+1)*n8r)));
|
|
v = Fr.add(v, Fr.mul(proof.eval_c, pol_qo.slice(i*n8r,(i+1)*n8r)));
|
|
v = Fr.add(v, pol_qc.slice(i*n8r,(i+1)*n8r));
|
|
v = Fr.sub(v, Fr.mul(coefs3, pol_s3.slice(i*n8r,(i+1)*n8r)));
|
|
}
|
|
pol_r.set(v, i*n8r);
|
|
}
|
|
|
|
proof.eval_r = evalPol(pol_r, ch.xi);
|
|
}
|
|
|
|
async function round5() {
|
|
const transcript5 = new Uint8Array(n8r*7);
|
|
Fr.toRprBE(transcript5, 0, proof.eval_a);
|
|
Fr.toRprBE(transcript5, n8r, proof.eval_b);
|
|
Fr.toRprBE(transcript5, n8r*2, proof.eval_c);
|
|
Fr.toRprBE(transcript5, n8r*3, proof.eval_s1);
|
|
Fr.toRprBE(transcript5, n8r*4, proof.eval_s2);
|
|
Fr.toRprBE(transcript5, n8r*5, proof.eval_zw);
|
|
Fr.toRprBE(transcript5, n8r*6, proof.eval_r);
|
|
ch.v = [];
|
|
ch.v[1] = hashToFr(transcript5);
|
|
if (logger) logger.debug("v: " + Fr.toString(ch.v[1]));
|
|
|
|
for (let i=2; i<=6; i++ ) ch.v[i] = Fr.mul(ch.v[i-1], ch.v[1]);
|
|
|
|
let pol_wxi = new ffjavascript.BigBuffer((zkey.domainSize+6)*n8r);
|
|
|
|
const xi2m = Fr.mul(ch.xim, ch.xim);
|
|
|
|
for (let i=0; i<zkey.domainSize+6; i++) {
|
|
let w = Fr.zero;
|
|
w = Fr.add(w, Fr.mul(xi2m, pol_t.slice( (zkey.domainSize*2+i)*n8r, (zkey.domainSize*2+i+1)*n8r )));
|
|
|
|
if (i<zkey.domainSize+3) {
|
|
w = Fr.add(w, Fr.mul(ch.v[1], pol_r.slice(i*n8r, (i+1)*n8r)));
|
|
}
|
|
|
|
if (i<zkey.domainSize+2) {
|
|
w = Fr.add(w, Fr.mul(ch.v[2], pol_a.slice(i*n8r, (i+1)*n8r)));
|
|
w = Fr.add(w, Fr.mul(ch.v[3], pol_b.slice(i*n8r, (i+1)*n8r)));
|
|
w = Fr.add(w, Fr.mul(ch.v[4], pol_c.slice(i*n8r, (i+1)*n8r)));
|
|
}
|
|
|
|
if (i<zkey.domainSize) {
|
|
w = Fr.add(w, pol_t.slice(i*n8r, (i+1)*n8r));
|
|
w = Fr.add(w, Fr.mul(ch.xim, pol_t.slice( (zkey.domainSize+i)*n8r, (zkey.domainSize+i+1)*n8r )));
|
|
w = Fr.add(w, Fr.mul(ch.v[5], pol_s1.slice(i*n8r, (i+1)*n8r)));
|
|
w = Fr.add(w, Fr.mul(ch.v[6], pol_s2.slice(i*n8r, (i+1)*n8r)));
|
|
}
|
|
|
|
pol_wxi.set(w, i*n8r);
|
|
}
|
|
|
|
let w0 = pol_wxi.slice(0, n8r);
|
|
w0 = Fr.sub(w0, proof.eval_t);
|
|
w0 = Fr.sub(w0, Fr.mul(ch.v[1], proof.eval_r));
|
|
w0 = Fr.sub(w0, Fr.mul(ch.v[2], proof.eval_a));
|
|
w0 = Fr.sub(w0, Fr.mul(ch.v[3], proof.eval_b));
|
|
w0 = Fr.sub(w0, Fr.mul(ch.v[4], proof.eval_c));
|
|
w0 = Fr.sub(w0, Fr.mul(ch.v[5], proof.eval_s1));
|
|
w0 = Fr.sub(w0, Fr.mul(ch.v[6], proof.eval_s2));
|
|
pol_wxi.set(w0, 0);
|
|
|
|
pol_wxi= divPol1(pol_wxi, ch.xi);
|
|
|
|
proof.Wxi = await expTau(pol_wxi, "multiexp Wxi");
|
|
|
|
let pol_wxiw = new ffjavascript.BigBuffer((zkey.domainSize+3)*n8r);
|
|
for (let i=0; i<zkey.domainSize+3; i++) {
|
|
const w = pol_z.slice(i*n8r, (i+1)*n8r);
|
|
pol_wxiw.set(w, i*n8r);
|
|
}
|
|
w0 = pol_wxiw.slice(0, n8r);
|
|
w0 = Fr.sub(w0, proof.eval_zw);
|
|
pol_wxiw.set(w0, 0);
|
|
|
|
pol_wxiw= divPol1(pol_wxiw, Fr.mul(ch.xi, Fr.w[zkey.power]));
|
|
proof.Wxiw = await expTau(pol_wxiw, "multiexp Wxiw");
|
|
}
|
|
|
|
function hashToFr(transcript) {
|
|
const v = ffjavascript.Scalar.fromRprBE(new Uint8Array(keccak256$1.arrayBuffer(transcript)));
|
|
return Fr.e(v);
|
|
}
|
|
|
|
|
|
function evalPol(P, x) {
|
|
const n = P.byteLength / n8r;
|
|
if (n == 0) return Fr.zero;
|
|
let res = P.slice((n-1)*n8r, n*n8r);
|
|
for (let i=n-2; i>=0; i--) {
|
|
res = Fr.add(Fr.mul(res, x), P.slice(i*n8r, (i+1)*n8r));
|
|
}
|
|
return res;
|
|
}
|
|
|
|
function divPol1(P, d) {
|
|
const n = P.byteLength/n8r;
|
|
const res = new ffjavascript.BigBuffer(n*n8r);
|
|
res.set(Fr.zero, (n-1) *n8r);
|
|
res.set(P.slice((n-1)*n8r, n*n8r), (n-2)*n8r);
|
|
for (let i=n-3; i>=0; i--) {
|
|
res.set(
|
|
Fr.add(
|
|
P.slice((i+1)*n8r, (i+2)*n8r),
|
|
Fr.mul(
|
|
d,
|
|
res.slice((i+1)*n8r, (i+2)*n8r)
|
|
)
|
|
),
|
|
i*n8r
|
|
);
|
|
}
|
|
if (!Fr.eq(
|
|
P.slice(0, n8r),
|
|
Fr.mul(
|
|
Fr.neg(d),
|
|
res.slice(0, n8r)
|
|
)
|
|
)) {
|
|
throw new Error("Polinomial does not divide");
|
|
}
|
|
return res;
|
|
}
|
|
|
|
async function expTau(b, name) {
|
|
const n = b.byteLength/n8r;
|
|
const PTauN = PTau.slice(0, n*curve.G1.F.n8*2);
|
|
const bm = await curve.Fr.batchFromMontgomery(b);
|
|
let res = await curve.G1.multiExpAffine(PTauN, bm, logger, name);
|
|
res = curve.G1.toAffine(res);
|
|
return res;
|
|
}
|
|
|
|
|
|
async function to4T(A, pz) {
|
|
pz = pz || [];
|
|
let a = await Fr.ifft(A);
|
|
const a4 = new ffjavascript.BigBuffer(n8r*zkey.domainSize*4);
|
|
a4.set(a, 0);
|
|
|
|
const a1 = new ffjavascript.BigBuffer(n8r*(zkey.domainSize + pz.length));
|
|
a1.set(a, 0);
|
|
for (let i= 0; i<pz.length; i++) {
|
|
a1.set(
|
|
Fr.add(
|
|
a1.slice((zkey.domainSize+i)*n8r, (zkey.domainSize+i+1)*n8r),
|
|
pz[i]
|
|
),
|
|
(zkey.domainSize+i)*n8r
|
|
);
|
|
a1.set(
|
|
Fr.sub(
|
|
a1.slice(i*n8r, (i+1)*n8r),
|
|
pz[i]
|
|
),
|
|
i*n8r
|
|
);
|
|
}
|
|
const A4 = await Fr.fft(a4);
|
|
return [a1, A4];
|
|
}
|
|
|
|
|
|
}
|
|
|
|
/*
|
|
Copyright 2021 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
async function plonkFullProve(input, wasmFile, zkeyFileName, logger) {
|
|
const wtns= {
|
|
type: "mem"
|
|
};
|
|
await wtnsCalculate(input, wasmFile, wtns);
|
|
return await plonk16Prove(zkeyFileName, wtns, logger);
|
|
}
|
|
|
|
/*
|
|
Copyright 2021 0kims association.
|
|
|
|
This file is part of snarkjs.
|
|
|
|
snarkjs is a free software: you can redistribute it and/or
|
|
modify it under the terms of the GNU General Public License as published by the
|
|
Free Software Foundation, either version 3 of the License, or (at your option)
|
|
any later version.
|
|
|
|
snarkjs is distributed in the hope that it will be useful,
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
|
|
more details.
|
|
|
|
You should have received a copy of the GNU General Public License along with
|
|
snarkjs. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
const {unstringifyBigInts} = ffjavascript.utils;
|
|
const { keccak256 } = jsSha3__default["default"];
|
|
|
|
|
|
async function plonkVerify(vk_verifier, publicSignals, proof, logger) {
|
|
vk_verifier = unstringifyBigInts(vk_verifier);
|
|
proof = unstringifyBigInts(proof);
|
|
publicSignals = unstringifyBigInts(publicSignals);
|
|
|
|
const curve = await getCurveFromName(vk_verifier.curve);
|
|
|
|
const Fr = curve.Fr;
|
|
const G1 = curve.G1;
|
|
|
|
proof = fromObjectProof(curve,proof);
|
|
vk_verifier = fromObjectVk(curve, vk_verifier);
|
|
if (!isWellConstructed(curve, proof)) {
|
|
logger.error("Proof is not well constructed");
|
|
return false;
|
|
}
|
|
const challanges = calculateChallanges(curve, proof);
|
|
if (logger) {
|
|
logger.debug("beta: " + Fr.toString(challanges.beta, 16));
|
|
logger.debug("gamma: " + Fr.toString(challanges.gamma, 16));
|
|
logger.debug("alpha: " + Fr.toString(challanges.alpha, 16));
|
|
logger.debug("xi: " + Fr.toString(challanges.xi, 16));
|
|
logger.debug("v1: " + Fr.toString(challanges.v[1], 16));
|
|
logger.debug("v6: " + Fr.toString(challanges.v[6], 16));
|
|
logger.debug("u: " + Fr.toString(challanges.u, 16));
|
|
}
|
|
const L = calculateLagrangeEvaluations(curve, challanges, vk_verifier);
|
|
if (logger) {
|
|
logger.debug("Lagrange Evaluations: ");
|
|
for (let i=1; i<L.length; i++) {
|
|
logger.debug(`L${i}(xi)=` + Fr.toString(L[i], 16));
|
|
}
|
|
}
|
|
|
|
if (publicSignals.length != vk_verifier.nPublic) {
|
|
logger.error("Number of public signals does not match with vk");
|
|
return false;
|
|
}
|
|
|
|
const pl = calculatePl(curve, publicSignals, L);
|
|
if (logger) {
|
|
logger.debug("Pl: " + Fr.toString(pl, 16));
|
|
}
|
|
|
|
const t = calculateT(curve, proof, challanges, pl, L[1]);
|
|
if (logger) {
|
|
logger.debug("t: " + Fr.toString(t, 16));
|
|
}
|
|
|
|
const D = calculateD(curve, proof, challanges, vk_verifier, L[1]);
|
|
if (logger) {
|
|
logger.debug("D: " + G1.toString(G1.toAffine(D), 16));
|
|
}
|
|
|
|
const F = calculateF(curve, proof, challanges, vk_verifier, D);
|
|
if (logger) {
|
|
logger.debug("F: " + G1.toString(G1.toAffine(F), 16));
|
|
}
|
|
|
|
const E = calculateE(curve, proof, challanges, vk_verifier, t);
|
|
if (logger) {
|
|
logger.debug("E: " + G1.toString(G1.toAffine(E), 16));
|
|
}
|
|
|
|
const res = await isValidPairing(curve, proof, challanges, vk_verifier, E, F);
|
|
|
|
if (logger) {
|
|
if (res) {
|
|
logger.info("OK!");
|
|
} else {
|
|
logger.warn("Invalid Proof");
|
|
}
|
|
}
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
function fromObjectProof(curve, proof) {
|
|
const G1 = curve.G1;
|
|
const Fr = curve.Fr;
|
|
const res = {};
|
|
res.A = G1.fromObject(proof.A);
|
|
res.B = G1.fromObject(proof.B);
|
|
res.C = G1.fromObject(proof.C);
|
|
res.Z = G1.fromObject(proof.Z);
|
|
res.T1 = G1.fromObject(proof.T1);
|
|
res.T2 = G1.fromObject(proof.T2);
|
|
res.T3 = G1.fromObject(proof.T3);
|
|
res.eval_a = Fr.fromObject(proof.eval_a);
|
|
res.eval_b = Fr.fromObject(proof.eval_b);
|
|
res.eval_c = Fr.fromObject(proof.eval_c);
|
|
res.eval_zw = Fr.fromObject(proof.eval_zw);
|
|
res.eval_s1 = Fr.fromObject(proof.eval_s1);
|
|
res.eval_s2 = Fr.fromObject(proof.eval_s2);
|
|
res.eval_r = Fr.fromObject(proof.eval_r);
|
|
res.Wxi = G1.fromObject(proof.Wxi);
|
|
res.Wxiw = G1.fromObject(proof.Wxiw);
|
|
return res;
|
|
}
|
|
|
|
function fromObjectVk(curve, vk) {
|
|
const G1 = curve.G1;
|
|
const G2 = curve.G2;
|
|
const Fr = curve.Fr;
|
|
const res = vk;
|
|
res.Qm = G1.fromObject(vk.Qm);
|
|
res.Ql = G1.fromObject(vk.Ql);
|
|
res.Qr = G1.fromObject(vk.Qr);
|
|
res.Qo = G1.fromObject(vk.Qo);
|
|
res.Qc = G1.fromObject(vk.Qc);
|
|
res.S1 = G1.fromObject(vk.S1);
|
|
res.S2 = G1.fromObject(vk.S2);
|
|
res.S3 = G1.fromObject(vk.S3);
|
|
res.k1 = Fr.fromObject(vk.k1);
|
|
res.k2 = Fr.fromObject(vk.k2);
|
|
res.X_2 = G2.fromObject(vk.X_2);
|
|
|
|
return res;
|
|
}
|
|
|
|
function isWellConstructed(curve, proof) {
|
|
const G1 = curve.G1;
|
|
if (!G1.isValid(proof.A)) return false;
|
|
if (!G1.isValid(proof.B)) return false;
|
|
if (!G1.isValid(proof.C)) return false;
|
|
if (!G1.isValid(proof.Z)) return false;
|
|
if (!G1.isValid(proof.T1)) return false;
|
|
if (!G1.isValid(proof.T2)) return false;
|
|
if (!G1.isValid(proof.T3)) return false;
|
|
if (!G1.isValid(proof.Wxi)) return false;
|
|
if (!G1.isValid(proof.Wxiw)) return false;
|
|
return true;
|
|
}
|
|
|
|
function calculateChallanges(curve, proof) {
|
|
const G1 = curve.G1;
|
|
const Fr = curve.Fr;
|
|
const n8r = curve.Fr.n8;
|
|
const res = {};
|
|
|
|
const transcript1 = new Uint8Array(G1.F.n8*2*3);
|
|
G1.toRprUncompressed(transcript1, 0, proof.A);
|
|
G1.toRprUncompressed(transcript1, G1.F.n8*2, proof.B);
|
|
G1.toRprUncompressed(transcript1, G1.F.n8*4, proof.C);
|
|
res.beta = hashToFr(curve, transcript1);
|
|
|
|
const transcript2 = new Uint8Array(n8r);
|
|
Fr.toRprBE(transcript2, 0, res.beta);
|
|
res.gamma = hashToFr(curve, transcript2);
|
|
|
|
const transcript3 = new Uint8Array(G1.F.n8*2);
|
|
G1.toRprUncompressed(transcript3, 0, proof.Z);
|
|
res.alpha = hashToFr(curve, transcript3);
|
|
|
|
const transcript4 = new Uint8Array(G1.F.n8*2*3);
|
|
G1.toRprUncompressed(transcript4, 0, proof.T1);
|
|
G1.toRprUncompressed(transcript4, G1.F.n8*2, proof.T2);
|
|
G1.toRprUncompressed(transcript4, G1.F.n8*4, proof.T3);
|
|
res.xi = hashToFr(curve, transcript4);
|
|
|
|
const transcript5 = new Uint8Array(n8r*7);
|
|
Fr.toRprBE(transcript5, 0, proof.eval_a);
|
|
Fr.toRprBE(transcript5, n8r, proof.eval_b);
|
|
Fr.toRprBE(transcript5, n8r*2, proof.eval_c);
|
|
Fr.toRprBE(transcript5, n8r*3, proof.eval_s1);
|
|
Fr.toRprBE(transcript5, n8r*4, proof.eval_s2);
|
|
Fr.toRprBE(transcript5, n8r*5, proof.eval_zw);
|
|
Fr.toRprBE(transcript5, n8r*6, proof.eval_r);
|
|
res.v = [];
|
|
res.v[1] = hashToFr(curve, transcript5);
|
|
|
|
for (let i=2; i<=6; i++ ) res.v[i] = Fr.mul(res.v[i-1], res.v[1]);
|
|
|
|
const transcript6 = new Uint8Array(G1.F.n8*2*2);
|
|
G1.toRprUncompressed(transcript6, 0, proof.Wxi);
|
|
G1.toRprUncompressed(transcript6, G1.F.n8*2, proof.Wxiw);
|
|
res.u = hashToFr(curve, transcript6);
|
|
|
|
return res;
|
|
}
|
|
|
|
function calculateLagrangeEvaluations(curve, challanges, vk) {
|
|
const Fr = curve.Fr;
|
|
|
|
let xin = challanges.xi;
|
|
let domainSize = 1;
|
|
for (let i=0; i<vk.power; i++) {
|
|
xin = Fr.square(xin);
|
|
domainSize *= 2;
|
|
}
|
|
challanges.xin = xin;
|
|
|
|
challanges.zh = Fr.sub(xin, Fr.one);
|
|
const L = [];
|
|
|
|
const n = Fr.e(domainSize);
|
|
let w = Fr.one;
|
|
for (let i=1; i<=Math.max(1, vk.nPublic); i++) {
|
|
L[i] = Fr.div(Fr.mul(w, challanges.zh), Fr.mul(n, Fr.sub(challanges.xi, w)));
|
|
w = Fr.mul(w, Fr.w[vk.power]);
|
|
}
|
|
|
|
return L;
|
|
}
|
|
|
|
function hashToFr(curve, transcript) {
|
|
const v = ffjavascript.Scalar.fromRprBE(new Uint8Array(keccak256.arrayBuffer(transcript)));
|
|
return curve.Fr.e(v);
|
|
}
|
|
|
|
function calculatePl(curve, publicSignals, L) {
|
|
const Fr = curve.Fr;
|
|
|
|
let pl = Fr.zero;
|
|
for (let i=0; i<publicSignals.length; i++) {
|
|
const w = Fr.e(publicSignals[i]);
|
|
pl = Fr.sub(pl, Fr.mul(w, L[i+1]));
|
|
}
|
|
return pl;
|
|
}
|
|
|
|
function calculateT(curve, proof, challanges, pl, l1) {
|
|
const Fr = curve.Fr;
|
|
let num = proof.eval_r;
|
|
num = Fr.add(num, pl);
|
|
|
|
let e1 = proof.eval_a;
|
|
e1 = Fr.add(e1, Fr.mul(challanges.beta, proof.eval_s1));
|
|
e1 = Fr.add(e1, challanges.gamma);
|
|
|
|
let e2 = proof.eval_b;
|
|
e2 = Fr.add(e2, Fr.mul(challanges.beta, proof.eval_s2));
|
|
e2 = Fr.add(e2, challanges.gamma);
|
|
|
|
let e3 = proof.eval_c;
|
|
e3 = Fr.add(e3, challanges.gamma);
|
|
|
|
let e = Fr.mul(Fr.mul(e1, e2), e3);
|
|
e = Fr.mul(e, proof.eval_zw);
|
|
e = Fr.mul(e, challanges.alpha);
|
|
|
|
num = Fr.sub(num, e);
|
|
|
|
num = Fr.sub(num, Fr.mul(l1, Fr.square(challanges.alpha)));
|
|
|
|
const t = Fr.div(num, challanges.zh);
|
|
|
|
return t;
|
|
}
|
|
|
|
function calculateD(curve, proof, challanges, vk, l1) {
|
|
const G1 = curve.G1;
|
|
const Fr = curve.Fr;
|
|
|
|
let s1 = Fr.mul(Fr.mul(proof.eval_a, proof.eval_b), challanges.v[1]);
|
|
let res = G1.timesFr(vk.Qm, s1);
|
|
|
|
let s2 = Fr.mul(proof.eval_a, challanges.v[1]);
|
|
res = G1.add(res, G1.timesFr(vk.Ql, s2));
|
|
|
|
let s3 = Fr.mul(proof.eval_b, challanges.v[1]);
|
|
res = G1.add(res, G1.timesFr(vk.Qr, s3));
|
|
|
|
let s4 = Fr.mul(proof.eval_c, challanges.v[1]);
|
|
res = G1.add(res, G1.timesFr(vk.Qo, s4));
|
|
|
|
res = G1.add(res, G1.timesFr(vk.Qc, challanges.v[1]));
|
|
|
|
const betaxi = Fr.mul(challanges.beta, challanges.xi);
|
|
let s6a = proof.eval_a;
|
|
s6a = Fr.add(s6a, betaxi);
|
|
s6a = Fr.add(s6a, challanges.gamma);
|
|
|
|
let s6b = proof.eval_b;
|
|
s6b = Fr.add(s6b, Fr.mul(betaxi, vk.k1));
|
|
s6b = Fr.add(s6b, challanges.gamma);
|
|
|
|
let s6c = proof.eval_c;
|
|
s6c = Fr.add(s6c, Fr.mul(betaxi, vk.k2));
|
|
s6c = Fr.add(s6c, challanges.gamma);
|
|
|
|
let s6 = Fr.mul(Fr.mul(s6a, s6b), s6c);
|
|
s6 = Fr.mul(s6, Fr.mul(challanges.alpha, challanges.v[1]));
|
|
|
|
let s6d = Fr.mul(Fr.mul(l1, Fr.square(challanges.alpha)), challanges.v[1]);
|
|
s6 = Fr.add(s6, s6d);
|
|
|
|
s6 = Fr.add(s6, challanges.u);
|
|
res = G1.add(res, G1.timesFr(proof.Z, s6));
|
|
|
|
|
|
let s7a = proof.eval_a;
|
|
s7a = Fr.add(s7a, Fr.mul(challanges.beta, proof.eval_s1));
|
|
s7a = Fr.add(s7a, challanges.gamma);
|
|
|
|
let s7b = proof.eval_b;
|
|
s7b = Fr.add(s7b, Fr.mul(challanges.beta, proof.eval_s2));
|
|
s7b = Fr.add(s7b, challanges.gamma);
|
|
|
|
let s7 = Fr.mul(s7a, s7b);
|
|
s7 = Fr.mul(s7, challanges.alpha);
|
|
s7 = Fr.mul(s7, challanges.v[1]);
|
|
s7 = Fr.mul(s7, challanges.beta);
|
|
s7 = Fr.mul(s7, proof.eval_zw);
|
|
res = G1.sub(res, G1.timesFr(vk.S3, s7));
|
|
|
|
return res;
|
|
}
|
|
|
|
function calculateF(curve, proof, challanges, vk, D) {
|
|
const G1 = curve.G1;
|
|
const Fr = curve.Fr;
|
|
|
|
let res = proof.T1;
|
|
|
|
res = G1.add(res, G1.timesFr(proof.T2, challanges.xin));
|
|
res = G1.add(res, G1.timesFr(proof.T3, Fr.square(challanges.xin)));
|
|
res = G1.add(res, D);
|
|
res = G1.add(res, G1.timesFr(proof.A, challanges.v[2]));
|
|
res = G1.add(res, G1.timesFr(proof.B, challanges.v[3]));
|
|
res = G1.add(res, G1.timesFr(proof.C, challanges.v[4]));
|
|
res = G1.add(res, G1.timesFr(vk.S1, challanges.v[5]));
|
|
res = G1.add(res, G1.timesFr(vk.S2, challanges.v[6]));
|
|
|
|
return res;
|
|
}
|
|
|
|
|
|
function calculateE(curve, proof, challanges, vk, t) {
|
|
const G1 = curve.G1;
|
|
const Fr = curve.Fr;
|
|
|
|
let s = t;
|
|
|
|
s = Fr.add(s, Fr.mul(challanges.v[1], proof.eval_r));
|
|
s = Fr.add(s, Fr.mul(challanges.v[2], proof.eval_a));
|
|
s = Fr.add(s, Fr.mul(challanges.v[3], proof.eval_b));
|
|
s = Fr.add(s, Fr.mul(challanges.v[4], proof.eval_c));
|
|
s = Fr.add(s, Fr.mul(challanges.v[5], proof.eval_s1));
|
|
s = Fr.add(s, Fr.mul(challanges.v[6], proof.eval_s2));
|
|
s = Fr.add(s, Fr.mul(challanges.u, proof.eval_zw));
|
|
|
|
const res = G1.timesFr(G1.one, s);
|
|
|
|
return res;
|
|
}
|
|
|
|
async function isValidPairing(curve, proof, challanges, vk, E, F) {
|
|
const G1 = curve.G1;
|
|
const Fr = curve.Fr;
|
|
|
|
let A1 = proof.Wxi;
|
|
A1 = G1.add(A1, G1.timesFr(proof.Wxiw, challanges.u));
|
|
|
|
let B1 = G1.timesFr(proof.Wxi, challanges.xi);
|
|
const s = Fr.mul(Fr.mul(challanges.u, challanges.xi), Fr.w[vk.power]);
|
|
B1 = G1.add(B1, G1.timesFr(proof.Wxiw, s));
|
|
B1 = G1.add(B1, F);
|
|
B1 = G1.sub(B1, E);
|
|
|
|
const res = await curve.pairingEq(
|
|
G1.neg(A1) , vk.X_2,
|
|
B1 , curve.G2.one
|
|
);
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
/*
|
|
Copyright 2021 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
function i2hex(i) {
|
|
return ("0" + i.toString(16)).slice(-2);
|
|
}
|
|
|
|
function p256(n) {
|
|
let nstr = n.toString(16);
|
|
while (nstr.length < 64) nstr = "0"+nstr;
|
|
nstr = `"0x${nstr}"`;
|
|
return nstr;
|
|
}
|
|
|
|
async function plonkExportSolidityCallData(proof, pub) {
|
|
|
|
const curve = await getCurveFromName(proof.curve);
|
|
const G1 = curve.G1;
|
|
const Fr = curve.Fr;
|
|
|
|
let inputs = "";
|
|
for (let i=0; i<pub.length; i++) {
|
|
if (inputs != "") inputs = inputs + ",";
|
|
inputs = inputs + p256(pub[i]);
|
|
}
|
|
|
|
const proofBuff = new Uint8Array(G1.F.n8*2*9 + Fr.n8*7);
|
|
G1.toRprUncompressed(proofBuff, 0, G1.e(proof.A));
|
|
G1.toRprUncompressed(proofBuff, G1.F.n8*2, G1.e(proof.B));
|
|
G1.toRprUncompressed(proofBuff, G1.F.n8*4, G1.e(proof.C));
|
|
G1.toRprUncompressed(proofBuff, G1.F.n8*6, G1.e(proof.Z));
|
|
G1.toRprUncompressed(proofBuff, G1.F.n8*8, G1.e(proof.T1));
|
|
G1.toRprUncompressed(proofBuff, G1.F.n8*10, G1.e(proof.T2));
|
|
G1.toRprUncompressed(proofBuff, G1.F.n8*12, G1.e(proof.T3));
|
|
G1.toRprUncompressed(proofBuff, G1.F.n8*14, G1.e(proof.Wxi));
|
|
G1.toRprUncompressed(proofBuff, G1.F.n8*16, G1.e(proof.Wxiw));
|
|
Fr.toRprBE(proofBuff, G1.F.n8*18 , Fr.e(proof.eval_a));
|
|
Fr.toRprBE(proofBuff, G1.F.n8*18 + Fr.n8, Fr.e(proof.eval_b));
|
|
Fr.toRprBE(proofBuff, G1.F.n8*18 + Fr.n8*2, Fr.e(proof.eval_c));
|
|
Fr.toRprBE(proofBuff, G1.F.n8*18 + Fr.n8*3, Fr.e(proof.eval_s1));
|
|
Fr.toRprBE(proofBuff, G1.F.n8*18 + Fr.n8*4, Fr.e(proof.eval_s2));
|
|
Fr.toRprBE(proofBuff, G1.F.n8*18 + Fr.n8*5, Fr.e(proof.eval_zw));
|
|
Fr.toRprBE(proofBuff, G1.F.n8*18 + Fr.n8*6, Fr.e(proof.eval_r));
|
|
|
|
const proofHex = Array.from(proofBuff).map(i2hex).join("");
|
|
|
|
const S="0x"+proofHex+",["+inputs+"]";
|
|
|
|
return S;
|
|
}
|
|
|
|
/*
|
|
Copyright 2018 0KIMS association.
|
|
|
|
This file is part of snarkJS.
|
|
|
|
snarkJS is a free software: you can redistribute it and/or modify it
|
|
under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
(at your option) any later version.
|
|
|
|
snarkJS is distributed in the hope that it will be useful, but WITHOUT
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with snarkJS. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
var plonk = /*#__PURE__*/Object.freeze({
|
|
__proto__: null,
|
|
setup: plonkSetup,
|
|
fullProve: plonkFullProve,
|
|
prove: plonk16Prove,
|
|
verify: plonkVerify,
|
|
exportSolidityCallData: plonkExportSolidityCallData
|
|
});
|
|
|
|
exports.groth16 = groth16;
|
|
exports.plonk = plonk;
|
|
exports.powersOfTau = powersoftau;
|
|
exports.r1cs = r1cs;
|
|
exports.wtns = wtns;
|
|
exports.zKey = zkey;
|