pqc/crypto_sign/qtesla-p-III/clean/poly.c
Thom Wiggers 74bc3e74e1
Fix UB in qTESLA
Lots of shifts left of signed integers
2020-03-13 15:02:24 -04:00

247 lines
8.4 KiB
C

/*************************************************************************************
* qTESLA: an efficient post-quantum signature scheme based on the R-LWE problem
*
* Abstract: NTT, modular reduction and polynomial functions
**************************************************************************************/
#include "api.h"
#include "poly.h"
#include "sp800-185.h"
extern const poly PQCLEAN_QTESLAPIII_CLEAN_zeta;
extern const poly PQCLEAN_QTESLAPIII_CLEAN_zetainv;
static int64_t reduce(int64_t a) {
// Montgomery reduction
int64_t u;
u = ((uint64_t)a * PARAM_QINV) & 0xFFFFFFFF;
u *= PARAM_Q;
a += u;
return a >> 32;
}
static int64_t barr_reduce(int64_t a) {
// Barrett reduction
int64_t u = (int64_t)((uint64_t)a * PARAM_BARR_MULT) >> PARAM_BARR_DIV;
return a - u * PARAM_Q;
}
static void ntt(poly a, const poly w) {
// Forward NTT transform
size_t NumoProblems = PARAM_N >> 1, jTwiddle = 0;
for (; NumoProblems > 0; NumoProblems >>= 1) {
size_t jFirst, j = 0;
for (jFirst = 0; jFirst < PARAM_N; jFirst = j + NumoProblems) {
sdigit_t W = (sdigit_t)w[jTwiddle++];
for (j = jFirst; j < jFirst + NumoProblems; j++) {
int64_t temp = barr_reduce(reduce((int64_t)W * a[j + NumoProblems]));
a[j + NumoProblems] = barr_reduce(a[j] + (2LL * PARAM_Q - temp));
a[j] = barr_reduce(temp + a[j]);
}
}
}
}
static void nttinv(poly a, const poly w) {
// Inverse NTT transform
size_t NumoProblems = 1, jTwiddle = 0;
for (; NumoProblems < PARAM_N; NumoProblems *= 2) {
size_t jFirst, j = 0;
for (jFirst = 0; jFirst < PARAM_N; jFirst = j + NumoProblems) {
sdigit_t W = (sdigit_t)w[jTwiddle++];
for (j = jFirst; j < jFirst + NumoProblems; j++) {
int64_t temp = a[j];
a[j] = barr_reduce((temp + a[j + NumoProblems]));
a[j + NumoProblems] = barr_reduce(reduce((int64_t)W * (temp + (2LL * PARAM_Q - a[j + NumoProblems]))));
}
}
}
}
static void poly_pointwise(poly result, const poly x, const poly y) {
// Pointwise polynomial multiplication result = x.y
for (size_t i = 0; i < PARAM_N; i++) {
result[i] = reduce(x[i] * y[i]);
}
}
void PQCLEAN_QTESLAPIII_CLEAN_poly_ntt(poly x_ntt, const poly x) {
// Call to NTT function. Avoids input destruction
for (size_t i = 0; i < PARAM_N; i++) {
x_ntt[i] = x[i];
}
ntt(x_ntt, PQCLEAN_QTESLAPIII_CLEAN_zeta);
}
void PQCLEAN_QTESLAPIII_CLEAN_poly_mul(poly result, const poly x, const poly y) {
// Polynomial multiplication result = x*y, with in place reduction for (X^N+1)
// The inputs x and y are assumed to be in NTT form
poly_pointwise(result, x, y);
nttinv(result, PQCLEAN_QTESLAPIII_CLEAN_zetainv);
}
void PQCLEAN_QTESLAPIII_CLEAN_poly_add(poly result, const poly x, const poly y) {
// Polynomial addition result = x+y
for (size_t i = 0; i < PARAM_N; i++) {
result[i] = x[i] + y[i];
}
}
void PQCLEAN_QTESLAPIII_CLEAN_poly_add_correct(poly result, const poly x, const poly y) {
// Polynomial addition result = x+y with correction
for (size_t i = 0; i < PARAM_N; i++) {
result[i] = x[i] + y[i];
result[i] -= PARAM_Q;
result[i] += (result[i] >> (RADIX32 - 1)) & PARAM_Q; // If result[i] >= q then subtract q
}
}
void PQCLEAN_QTESLAPIII_CLEAN_poly_sub(poly result, const poly x, const poly y) {
// Polynomial subtraction result = x-y
for (size_t i = 0; i < PARAM_N; i++) {
result[i] = barr_reduce(x[i] - y[i]);
}
}
/********************************************************************************************
* Name: sparse_mul8
* Description: performs sparse polynomial multiplication
* Parameters: inputs:
* - const uint8_t *s: part of the secret key
* - const uint32_t pos_list[PARAM_H]: list of indices of nonzero elements in c
* - const int16_t sign_list[PARAM_H]: list of signs of nonzero elements in c
* outputs:
* - poly prod: product of 2 polynomials
*
* Note: pos_list[] and sign_list[] contain public information since c is public
*********************************************************************************************/
void PQCLEAN_QTESLAPIII_CLEAN_sparse_mul8(poly prod, const uint8_t *s, const uint32_t pos_list[PARAM_H], const int16_t sign_list[PARAM_H]) {
size_t i, j, pos;
int8_t *t = (int8_t *)s;
for (i = 0; i < PARAM_N; i++) {
prod[i] = 0;
}
for (i = 0; i < PARAM_H; i++) {
pos = pos_list[i];
for (j = 0; j < pos; j++) {
prod[j] = prod[j] - sign_list[i] * t[j + PARAM_N - pos];
}
for (j = pos; j < PARAM_N; j++) {
prod[j] = prod[j] + sign_list[i] * t[j - pos];
}
}
}
/********************************************************************************************
* Name: sparse_mul32
* Description: performs sparse polynomial multiplication
* Parameters: inputs:
* - const int32_t* pk: part of the public key
* - const uint32_t pos_list[PARAM_H]: list of indices of nonzero elements in c
* - const int16_t sign_list[PARAM_H]: list of signs of nonzero elements in c
* outputs:
* - poly prod: product of 2 polynomials
*********************************************************************************************/
void PQCLEAN_QTESLAPIII_CLEAN_sparse_mul32(poly prod, const int32_t *pk, const uint32_t pos_list[PARAM_H], const int16_t sign_list[PARAM_H]) {
size_t i, j, pos;
for (i = 0; i < PARAM_N; i++) {
prod[i] = 0;
}
for (i = 0; i < PARAM_H; i++) {
pos = pos_list[i];
for (j = 0; j < pos; j++) {
prod[j] = prod[j] - sign_list[i] * pk[j + PARAM_N - pos];
}
for (j = pos; j < PARAM_N; j++) {
prod[j] = prod[j] + sign_list[i] * pk[j - pos];
}
}
for (i = 0; i < PARAM_N; i++) {
prod[i] = barr_reduce(prod[i]);
}
}
void PQCLEAN_QTESLAPIII_CLEAN_poly_uniform(poly_k a, const uint8_t *seed) {
// Generation of polynomials "a_i"
size_t pos = 0, i = 0, nbytes = (PARAM_Q_LOG + 7) / 8;
size_t nblocks = PARAM_GEN_A;
uint32_t val1, val2, val3, val4, mask = (uint32_t)(1 << PARAM_Q_LOG) - 1;
uint8_t buf[SHAKE128_RATE * PARAM_GEN_A];
uint16_t dmsp = 0;
uint8_t dmsp_bytes[2];
dmsp_bytes[0] = (uint8_t)(dmsp & 0xff);
dmsp_bytes[1] = (uint8_t)(dmsp >> 8);
cshake128(buf, SHAKE128_RATE * PARAM_GEN_A, (uint8_t *)NULL, 0, dmsp_bytes, 2, seed, CRYPTO_RANDOMBYTES);
++dmsp;
while (i < PARAM_K * PARAM_N) {
if (pos > SHAKE128_RATE * nblocks - 4 * nbytes) {
nblocks = 1;
dmsp_bytes[0] = (uint8_t)(dmsp & 0xff);
dmsp_bytes[1] = (uint8_t)(dmsp >> 8);
cshake128(buf, SHAKE128_RATE * nblocks, (uint8_t *)NULL, 0, dmsp_bytes, 2, seed, CRYPTO_RANDOMBYTES);
++dmsp;
pos = 0;
}
val1 = ((uint32_t)(buf[pos])
| ((uint32_t)buf[pos + 1] << 8)
| ((uint32_t)buf[pos + 2] << 16)
| ((uint32_t)buf[pos + 3] << 24))
& mask;
pos += nbytes;
val2 = ((uint32_t)(buf[pos])
| ((uint32_t)buf[pos + 1] << 8)
| ((uint32_t)buf[pos + 2] << 16)
| ((uint32_t)buf[pos + 3] << 24))
& mask;
pos += nbytes;
val3 = ((uint32_t)(buf[pos])
| ((uint32_t)buf[pos + 1] << 8)
| ((uint32_t)buf[pos + 2] << 16)
| ((uint32_t)buf[pos + 3] << 24))
& mask;
pos += nbytes;
val4 = ((uint32_t)(buf[pos])
| ((uint32_t)buf[pos + 1] << 8)
| ((uint32_t)buf[pos + 2] << 16)
| ((uint32_t)buf[pos + 3] << 24))
& mask;
pos += nbytes;
if (val1 < PARAM_Q && i < PARAM_K * PARAM_N) {
a[i++] = reduce((int64_t)val1 * PARAM_R2_INVN);
}
if (val2 < PARAM_Q && i < PARAM_K * PARAM_N) {
a[i++] = reduce((int64_t)val2 * PARAM_R2_INVN);
}
if (val3 < PARAM_Q && i < PARAM_K * PARAM_N) {
a[i++] = reduce((int64_t)val3 * PARAM_R2_INVN);
}
if (val4 < PARAM_Q && i < PARAM_K * PARAM_N) {
a[i++] = reduce((int64_t)val4 * PARAM_R2_INVN);
}
}
}