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