mirror of
https://github.com/cheat/cheat.git
synced 2024-11-22 22:11:35 +01:00
95a4e31b6c
Upgrade all dependencies to newest versions.
181 lines
4.2 KiB
Go
181 lines
4.2 KiB
Go
package ed25519
|
||
|
||
import (
|
||
"crypto/subtle"
|
||
"encoding/binary"
|
||
"math/bits"
|
||
|
||
"github.com/cloudflare/circl/internal/conv"
|
||
"github.com/cloudflare/circl/math"
|
||
fp "github.com/cloudflare/circl/math/fp25519"
|
||
)
|
||
|
||
var paramD = fp.Elt{
|
||
0xa3, 0x78, 0x59, 0x13, 0xca, 0x4d, 0xeb, 0x75,
|
||
0xab, 0xd8, 0x41, 0x41, 0x4d, 0x0a, 0x70, 0x00,
|
||
0x98, 0xe8, 0x79, 0x77, 0x79, 0x40, 0xc7, 0x8c,
|
||
0x73, 0xfe, 0x6f, 0x2b, 0xee, 0x6c, 0x03, 0x52,
|
||
}
|
||
|
||
// mLSBRecoding parameters.
|
||
const (
|
||
fxT = 257
|
||
fxV = 2
|
||
fxW = 3
|
||
fx2w1 = 1 << (uint(fxW) - 1)
|
||
numWords64 = (paramB * 8 / 64)
|
||
)
|
||
|
||
// mLSBRecoding is the odd-only modified LSB-set.
|
||
//
|
||
// Reference:
|
||
//
|
||
// "Efficient and secure algorithms for GLV-based scalar multiplication and
|
||
// their implementation on GLV–GLS curves" by (Faz-Hernandez et al.)
|
||
// http://doi.org/10.1007/s13389-014-0085-7.
|
||
func mLSBRecoding(L []int8, k []byte) {
|
||
const ee = (fxT + fxW*fxV - 1) / (fxW * fxV)
|
||
const dd = ee * fxV
|
||
const ll = dd * fxW
|
||
if len(L) == (ll + 1) {
|
||
var m [numWords64 + 1]uint64
|
||
for i := 0; i < numWords64; i++ {
|
||
m[i] = binary.LittleEndian.Uint64(k[8*i : 8*i+8])
|
||
}
|
||
condAddOrderN(&m)
|
||
L[dd-1] = 1
|
||
for i := 0; i < dd-1; i++ {
|
||
kip1 := (m[(i+1)/64] >> (uint(i+1) % 64)) & 0x1
|
||
L[i] = int8(kip1<<1) - 1
|
||
}
|
||
{ // right-shift by d
|
||
right := uint(dd % 64)
|
||
left := uint(64) - right
|
||
lim := ((numWords64+1)*64 - dd) / 64
|
||
j := dd / 64
|
||
for i := 0; i < lim; i++ {
|
||
m[i] = (m[i+j] >> right) | (m[i+j+1] << left)
|
||
}
|
||
m[lim] = m[lim+j] >> right
|
||
}
|
||
for i := dd; i < ll; i++ {
|
||
L[i] = L[i%dd] * int8(m[0]&0x1)
|
||
div2subY(m[:], int64(L[i]>>1), numWords64)
|
||
}
|
||
L[ll] = int8(m[0])
|
||
}
|
||
}
|
||
|
||
// absolute returns always a positive value.
|
||
func absolute(x int32) int32 {
|
||
mask := x >> 31
|
||
return (x + mask) ^ mask
|
||
}
|
||
|
||
// condAddOrderN updates x = x+order if x is even, otherwise x remains unchanged.
|
||
func condAddOrderN(x *[numWords64 + 1]uint64) {
|
||
isOdd := (x[0] & 0x1) - 1
|
||
c := uint64(0)
|
||
for i := 0; i < numWords64; i++ {
|
||
orderWord := binary.LittleEndian.Uint64(order[8*i : 8*i+8])
|
||
o := isOdd & orderWord
|
||
x0, c0 := bits.Add64(x[i], o, c)
|
||
x[i] = x0
|
||
c = c0
|
||
}
|
||
x[numWords64], _ = bits.Add64(x[numWords64], 0, c)
|
||
}
|
||
|
||
// div2subY update x = (x/2) - y.
|
||
func div2subY(x []uint64, y int64, l int) {
|
||
s := uint64(y >> 63)
|
||
for i := 0; i < l-1; i++ {
|
||
x[i] = (x[i] >> 1) | (x[i+1] << 63)
|
||
}
|
||
x[l-1] = (x[l-1] >> 1)
|
||
|
||
b := uint64(0)
|
||
x0, b0 := bits.Sub64(x[0], uint64(y), b)
|
||
x[0] = x0
|
||
b = b0
|
||
for i := 1; i < l-1; i++ {
|
||
x0, b0 := bits.Sub64(x[i], s, b)
|
||
x[i] = x0
|
||
b = b0
|
||
}
|
||
x[l-1], _ = bits.Sub64(x[l-1], s, b)
|
||
}
|
||
|
||
func (P *pointR1) fixedMult(scalar []byte) {
|
||
if len(scalar) != paramB {
|
||
panic("wrong scalar size")
|
||
}
|
||
const ee = (fxT + fxW*fxV - 1) / (fxW * fxV)
|
||
const dd = ee * fxV
|
||
const ll = dd * fxW
|
||
|
||
L := make([]int8, ll+1)
|
||
mLSBRecoding(L[:], scalar)
|
||
S := &pointR3{}
|
||
P.SetIdentity()
|
||
for ii := ee - 1; ii >= 0; ii-- {
|
||
P.double()
|
||
for j := 0; j < fxV; j++ {
|
||
dig := L[fxW*dd-j*ee+ii-ee]
|
||
for i := (fxW-1)*dd - j*ee + ii - ee; i >= (2*dd - j*ee + ii - ee); i = i - dd {
|
||
dig = 2*dig + L[i]
|
||
}
|
||
idx := absolute(int32(dig))
|
||
sig := L[dd-j*ee+ii-ee]
|
||
Tabj := &tabSign[fxV-j-1]
|
||
for k := 0; k < fx2w1; k++ {
|
||
S.cmov(&Tabj[k], subtle.ConstantTimeEq(int32(k), idx))
|
||
}
|
||
S.cneg(subtle.ConstantTimeEq(int32(sig), -1))
|
||
P.mixAdd(S)
|
||
}
|
||
}
|
||
}
|
||
|
||
const (
|
||
omegaFix = 7
|
||
omegaVar = 5
|
||
)
|
||
|
||
// doubleMult returns P=mG+nQ.
|
||
func (P *pointR1) doubleMult(Q *pointR1, m, n []byte) {
|
||
nafFix := math.OmegaNAF(conv.BytesLe2BigInt(m), omegaFix)
|
||
nafVar := math.OmegaNAF(conv.BytesLe2BigInt(n), omegaVar)
|
||
|
||
if len(nafFix) > len(nafVar) {
|
||
nafVar = append(nafVar, make([]int32, len(nafFix)-len(nafVar))...)
|
||
} else if len(nafFix) < len(nafVar) {
|
||
nafFix = append(nafFix, make([]int32, len(nafVar)-len(nafFix))...)
|
||
}
|
||
|
||
var TabQ [1 << (omegaVar - 2)]pointR2
|
||
Q.oddMultiples(TabQ[:])
|
||
P.SetIdentity()
|
||
for i := len(nafFix) - 1; i >= 0; i-- {
|
||
P.double()
|
||
// Generator point
|
||
if nafFix[i] != 0 {
|
||
idxM := absolute(nafFix[i]) >> 1
|
||
R := tabVerif[idxM]
|
||
if nafFix[i] < 0 {
|
||
R.neg()
|
||
}
|
||
P.mixAdd(&R)
|
||
}
|
||
// Variable input point
|
||
if nafVar[i] != 0 {
|
||
idxN := absolute(nafVar[i]) >> 1
|
||
S := TabQ[idxN]
|
||
if nafVar[i] < 0 {
|
||
S.neg()
|
||
}
|
||
P.add(&S)
|
||
}
|
||
}
|
||
}
|