| 1 |
/* |
| 2 |
This file is for public-key generation |
| 3 |
*/ |
| 4 |
// 20240822 djb: more cryptoint usage |
| 5 |
// 20240805 djb: more use of cryptoint |
| 6 |
// 20240715 djb: more use of crypto_*_mask |
| 7 |
// 20240611 djb: using crypto_uint64_bottomzeros_num |
| 8 |
// 20240608 djb: using crypto_*_mask |
| 9 |
// 20240508 djb: switch to crypto_sort_int64 |
| 10 |
// 20221231 djb: remove unused min definition |
| 11 |
// 20221231 djb: more 0 initialization to clarify data flow; tnx thom wiggers |
| 12 |
// 20221230 djb: add linker lines |
| 13 |
|
| 14 |
// linker define pk_gen |
| 15 |
// linker use fft vec_inv vec_mul |
| 16 |
|
| 17 |
#include "pk_gen.h" |
| 18 |
|
| 19 |
#include "controlbits.h" |
| 20 |
#include "crypto_sort_int64.h" |
| 21 |
#include "transpose.h" |
| 22 |
#include "params.h" |
| 23 |
#include "benes.h" |
| 24 |
#include "util.h" |
| 25 |
#include "fft.h" |
| 26 |
#include "crypto_declassify.h" |
| 27 |
#include "crypto_uint16.h" |
| 28 |
#include "crypto_uint64.h" |
| 29 |
#include "crypto_int64.h" |
| 30 |
|
| 31 |
static crypto_uint64 uint64_is_equal_declassify(uint64_t t,uint64_t u) |
| 32 |
{ |
| 33 |
crypto_uint64 mask = crypto_uint64_equal_mask(t,u); |
| 34 |
crypto_declassify(&mask,sizeof mask); |
| 35 |
return mask; |
| 36 |
} |
| 37 |
|
| 38 |
static crypto_uint64 uint64_is_zero_declassify(uint64_t t) |
| 39 |
{ |
| 40 |
crypto_uint64 mask = crypto_uint64_zero_mask(t); |
| 41 |
crypto_declassify(&mask,sizeof mask); |
| 42 |
return mask; |
| 43 |
} |
| 44 |
|
| 45 |
#include <stdint.h> |
| 46 |
|
| 47 |
static void de_bitslicing(uint64_t * out, const vec in[][GFBITS]) |
| 48 |
{ |
| 49 |
int i, j, r; |
| 50 |
|
| 51 |
for (i = 0; i < (1 << GFBITS); i++) |
| 52 |
out[i] = 0 ; |
| 53 |
|
| 54 |
for (i = 0; i < 128; i++) |
| 55 |
for (j = GFBITS-1; j >= 0; j--) |
| 56 |
for (r = 0; r < 64; r++) |
| 57 |
{ |
| 58 |
out[i*64 + r] <<= 1; |
| 59 |
out[i*64 + r] |= crypto_int64_bitmod_01(in[i][j], r); |
| 60 |
} |
| 61 |
} |
| 62 |
|
| 63 |
static void to_bitslicing_2x(vec out0[][GFBITS], vec out1[][GFBITS], const uint64_t * in) |
| 64 |
{ |
| 65 |
int i, j, r; |
| 66 |
|
| 67 |
for (i = 0; i < 128; i++) |
| 68 |
{ |
| 69 |
for (j = 0;j < GFBITS;++j) out0[i][j] = out1[i][j] = 0; |
| 70 |
|
| 71 |
for (j = GFBITS-1; j >= 0; j--) |
| 72 |
for (r = 63; r >= 0; r--) |
| 73 |
{ |
| 74 |
out1[i][j] <<= 1; |
| 75 |
out1[i][j] |= crypto_int64_bitmod_01(in[i*64 + r], j + GFBITS); |
| 76 |
} |
| 77 |
|
| 78 |
for (j = GFBITS-1; j >= 0; j--) |
| 79 |
for (r = 63; r >= 0; r--) |
| 80 |
{ |
| 81 |
out0[i][GFBITS-1-j] <<= 1; |
| 82 |
out0[i][GFBITS-1-j] |= crypto_int64_bitmod_01(in[i*64 + r], j); |
| 83 |
} |
| 84 |
} |
| 85 |
} |
| 86 |
|
| 87 |
static int mov_columns(uint64_t mat[][ (SYS_N + 63) / 64 ], int16_t * pi, uint64_t * pivots) |
| 88 |
{ |
| 89 |
int i, j, k, s, block_idx, row, tail; |
| 90 |
uint64_t buf[64], ctz_list[32], t, d, allone = -1, one = 1; |
| 91 |
|
| 92 |
row = PK_NROWS - 32; |
| 93 |
block_idx = row/64; |
| 94 |
tail = row % 64; |
| 95 |
|
| 96 |
// extract the 32x64 matrix |
| 97 |
|
| 98 |
for (i = 0; i < 32; i++) |
| 99 |
buf[i] = (mat[ row + i ][ block_idx + 0 ] >> tail) | |
| 100 |
(mat[ row + i ][ block_idx + 1 ] << (64-tail)); |
| 101 |
|
| 102 |
// compute the column indices of pivots by Gaussian elimination. |
| 103 |
// the indices are stored in ctz_list |
| 104 |
|
| 105 |
*pivots = 0; |
| 106 |
|
| 107 |
for (i = 0; i < 32; i++) |
| 108 |
{ |
| 109 |
t = buf[i]; |
| 110 |
for (j = i+1; j < 32; j++) |
| 111 |
t |= buf[j]; |
| 112 |
|
| 113 |
if (uint64_is_zero_declassify(t)) return -1; // return if buf is not full rank |
| 114 |
|
| 115 |
ctz_list[i] = s = crypto_uint64_bottomzeros_num(t); |
| 116 |
*pivots |= crypto_uint64_shlmod(one,ctz_list[i]); |
| 117 |
|
| 118 |
for (j = i+1; j < 32; j++) buf[i] ^= buf[j] & ~crypto_uint64_bitmod_mask(buf[i],s); |
| 119 |
for (j = 0; j < i; j++) buf[j] ^= buf[i] & crypto_uint64_bitmod_mask(buf[j],s); |
| 120 |
for (j = i+1; j < 32; j++) buf[j] ^= buf[i] & crypto_uint64_bitmod_mask(buf[j],s); |
| 121 |
} |
| 122 |
|
| 123 |
// updating permutation |
| 124 |
|
| 125 |
for (j = 0; j < 32; j++) |
| 126 |
for (k = j+1; k < 64; k++) |
| 127 |
{ |
| 128 |
d = pi[ row + j ] ^ pi[ row + k ]; |
| 129 |
d &= crypto_uint16_equal_mask(k, ctz_list[j]); |
| 130 |
pi[ row + j ] ^= d; |
| 131 |
pi[ row + k ] ^= d; |
| 132 |
} |
| 133 |
|
| 134 |
// moving columns of mat according to the column indices of pivots |
| 135 |
|
| 136 |
for (i = 0; i < PK_NROWS; i++) |
| 137 |
{ |
| 138 |
t = (mat[ i ][ block_idx + 0 ] >> tail) | |
| 139 |
(mat[ i ][ block_idx + 1 ] << (64-tail)); |
| 140 |
|
| 141 |
for (j = 0; j < 32; j++) |
| 142 |
{ |
| 143 |
d = t >> j; |
| 144 |
d ^= crypto_uint64_shrmod(t,ctz_list[j]); |
| 145 |
d = crypto_uint64_bottombit_01(d); |
| 146 |
|
| 147 |
t ^= crypto_uint64_shlmod(d,ctz_list[j]); |
| 148 |
t ^= d << j; |
| 149 |
} |
| 150 |
|
| 151 |
mat[ i ][ block_idx + 0 ] = (mat[ i ][ block_idx + 0 ] & (allone >> (64-tail))) | (t << tail); |
| 152 |
mat[ i ][ block_idx + 1 ] = (mat[ i ][ block_idx + 1 ] & (allone << tail)) | (t >> (64-tail)); |
| 153 |
} |
| 154 |
|
| 155 |
return 0; |
| 156 |
} |
| 157 |
|
| 158 |
int pk_gen(unsigned char * pk, const unsigned char * irr, uint32_t * perm, int16_t * pi, uint64_t * pivots) |
| 159 |
{ |
| 160 |
const int nblocks_H = (SYS_N + 63) / 64; |
| 161 |
const int nblocks_I = (PK_NROWS + 63) / 64; |
| 162 |
const int block_idx = nblocks_I - 1; |
| 163 |
int tail = PK_NROWS % 64; |
| 164 |
|
| 165 |
int i, j, k; |
| 166 |
int row, c; |
| 167 |
|
| 168 |
uint64_t mat[ PK_NROWS ][ nblocks_H ]; |
| 169 |
|
| 170 |
uint64_t mask; |
| 171 |
|
| 172 |
vec irr_int[2][ GFBITS ]; |
| 173 |
|
| 174 |
vec consts[ 128 ][ GFBITS ]; |
| 175 |
vec eval[ 128 ][ GFBITS ]; |
| 176 |
vec prod[ 128 ][ GFBITS ]; |
| 177 |
vec tmp[ GFBITS ]; |
| 178 |
|
| 179 |
uint64_t list[1 << GFBITS]; |
| 180 |
|
| 181 |
// compute the inverses |
| 182 |
|
| 183 |
irr_load(irr_int, irr); |
| 184 |
|
| 185 |
fft(eval, irr_int); |
| 186 |
|
| 187 |
vec_copy(prod[0], eval[0]); |
| 188 |
|
| 189 |
for (i = 1; i < 128; i++) |
| 190 |
vec_mul(prod[i], prod[i-1], eval[i]); |
| 191 |
|
| 192 |
vec_inv(tmp, prod[127]); |
| 193 |
|
| 194 |
for (i = 126; i >= 0; i--) |
| 195 |
{ |
| 196 |
vec_mul(prod[i+1], prod[i], tmp); |
| 197 |
vec_mul(tmp, tmp, eval[i+1]); |
| 198 |
} |
| 199 |
|
| 200 |
vec_copy(prod[0], tmp); |
| 201 |
|
| 202 |
// fill matrix |
| 203 |
|
| 204 |
de_bitslicing(list, prod); |
| 205 |
|
| 206 |
for (i = 0; i < (1 << GFBITS); i++) |
| 207 |
{ |
| 208 |
list[i] <<= GFBITS; |
| 209 |
list[i] |= i; |
| 210 |
list[i] |= ((uint64_t) perm[i]) << 31; |
| 211 |
} |
| 212 |
|
| 213 |
crypto_sort_int64(list, 1 << GFBITS); |
| 214 |
|
| 215 |
for (i = 1; i < (1 << GFBITS); i++) |
| 216 |
if (uint64_is_equal_declassify(list[i-1] >> 31,list[i] >> 31)) |
| 217 |
return -1; |
| 218 |
|
| 219 |
to_bitslicing_2x(consts, prod, list); |
| 220 |
|
| 221 |
for (i = 0; i < (1 << GFBITS); i++) |
| 222 |
pi[i] = list[i] & GFMASK; |
| 223 |
|
| 224 |
for (j = 0; j < nblocks_H; j++) |
| 225 |
for (k = 0; k < GFBITS; k++) |
| 226 |
mat[ k ][ j ] = prod[ j ][ k ]; |
| 227 |
|
| 228 |
for (i = 1; i < SYS_T; i++) |
| 229 |
for (j = 0; j < nblocks_H; j++) |
| 230 |
{ |
| 231 |
vec_mul(prod[j], prod[j], consts[j]); |
| 232 |
|
| 233 |
for (k = 0; k < GFBITS; k++) |
| 234 |
mat[ i*GFBITS + k ][ j ] = prod[ j ][ k ]; |
| 235 |
} |
| 236 |
|
| 237 |
// gaussian elimination |
| 238 |
|
| 239 |
for (row = 0; row < PK_NROWS; row++) |
| 240 |
{ |
| 241 |
i = row >> 6; |
| 242 |
j = row & 63; |
| 243 |
|
| 244 |
if (row == PK_NROWS - 32) |
| 245 |
{ |
| 246 |
if (mov_columns(mat, pi, pivots)) |
| 247 |
return -1; |
| 248 |
} |
| 249 |
|
| 250 |
for (k = row + 1; k < PK_NROWS; k++) |
| 251 |
{ |
| 252 |
mask = ~crypto_uint64_bitmod_mask(mat[ row ][ i ], j); |
| 253 |
|
| 254 |
for (c = 0; c < nblocks_H; c++) |
| 255 |
mat[ row ][ c ] ^= mat[ k ][ c ] & mask; |
| 256 |
} |
| 257 |
|
| 258 |
mask = crypto_uint64_bitmod_mask(mat[ row ][ i ], j); |
| 259 |
if ( uint64_is_zero_declassify(mask) ) // return if not systematic |
| 260 |
{ |
| 261 |
return -1; |
| 262 |
} |
| 263 |
|
| 264 |
for (k = 0; k < row; k++) |
| 265 |
{ |
| 266 |
mask = crypto_uint64_bitmod_mask(mat[ k ][ i ], j); |
| 267 |
|
| 268 |
for (c = 0; c < nblocks_H; c++) |
| 269 |
mat[ k ][ c ] ^= mat[ row ][ c ] & mask; |
| 270 |
} |
| 271 |
|
| 272 |
for (k = row+1; k < PK_NROWS; k++) |
| 273 |
{ |
| 274 |
mask = crypto_uint64_bitmod_mask(mat[ k ][ i ], j); |
| 275 |
|
| 276 |
for (c = 0; c < nblocks_H; c++) |
| 277 |
mat[ k ][ c ] ^= mat[ row ][ c ] & mask; |
| 278 |
} |
| 279 |
} |
| 280 |
|
| 281 |
for (row = 0; row < PK_NROWS; row++) |
| 282 |
{ |
| 283 |
for (k = block_idx; k < nblocks_H - 1; k++) |
| 284 |
{ |
| 285 |
mat[row][k] = (mat[row][k] >> tail) | (mat[row][k+1] << (64-tail)); |
| 286 |
store8(pk, mat[row][k]); |
| 287 |
pk += 8; |
| 288 |
} |
| 289 |
|
| 290 |
mat[row][k] >>= tail; |
| 291 |
store_i(pk, mat[row][k], PK_ROW_BYTES % 8); |
| 292 |
|
| 293 |
pk[ (PK_ROW_BYTES % 8)-1 ] &= (1 << (PK_NCOLS % 8)) - 1; // removing redundant bits |
| 294 |
|
| 295 |
pk += PK_ROW_BYTES % 8; |
| 296 |
} |
| 297 |
|
| 298 |
// |
| 299 |
|
| 300 |
return 0; |
| 301 |
} |
| 302 |
|