| 1 | /* -*-c-*- |
| 2 | * |
| 3 | * Salsa20 core definitions |
| 4 | * |
| 5 | * (c) 2015 Straylight/Edgeware |
| 6 | */ |
| 7 | |
| 8 | #ifndef CATACOMB_SALSA20_CORE_H |
| 9 | #define CATACOMB_SALSA20_CORE_H |
| 10 | |
| 11 | #ifdef __cplusplus |
| 12 | extern "C" { |
| 13 | #endif |
| 14 | |
| 15 | /*----- Header files ------------------------------------------------------*/ |
| 16 | |
| 17 | #include <mLib/bits.h> |
| 18 | #include <mLib/macros.h> |
| 19 | |
| 20 | #ifndef CATACOMB_SALSA20_H |
| 21 | # include "salsa20.h" |
| 22 | #endif |
| 23 | |
| 24 | /*----- Magic constants ---------------------------------------------------*/ |
| 25 | |
| 26 | /* The magic Salsa20 constants, for 256-bit keys... */ |
| 27 | #define SALSA20_A256 0x61707865 /* e x p a */ |
| 28 | #define SALSA20_B256 0x3320646e /* n d 3 */ |
| 29 | #define SALSA20_C256 0x79622d32 /* 2 - b y */ |
| 30 | #define SALSA20_D256 0x6b206574 /* t e k */ |
| 31 | |
| 32 | /* ... and for 128-bit keys ... */ |
| 33 | #define SALSA20_A128 SALSA20_A256 /* e x p a */ |
| 34 | #define SALSA20_B128 0x3120646e /* n d 1 */ |
| 35 | #define SALSA20_C128 0x79622d36 /* 6 - b y */ |
| 36 | #define SALSA20_D128 SALSA20_D256 /* t e k */ |
| 37 | |
| 38 | /* ... and for 80-bit keys, for completeness's sake. */ |
| 39 | #define SALSA20_A80 SALSA20_A128 /* e x p a */ |
| 40 | #define SALSA20_B80 SALSA20_B128 /* n d 1 */ |
| 41 | #define SALSA20_C80 0x79622d30 /* 0 - b y */ |
| 42 | #define SALSA20_D80 SALSA20_D128 /* t e k */ |
| 43 | |
| 44 | /*----- The Salsa20 core function -----------------------------------------*/ |
| 45 | |
| 46 | /* It makes life somewhat easier if we don't actually store and maintain the |
| 47 | * input matrix in the textbook order. Instead, we rotate the columns other |
| 48 | * than the leftmost one upwards, so that the constants which were originally |
| 49 | * along the diagonal end up on the top row. We'll need to undo this |
| 50 | * permutation on output, but that's not too terrible an imposition. |
| 51 | * |
| 52 | * The permutation we're applying to the matrix elements is this: |
| 53 | * |
| 54 | * [ 0 1 2 3 ] [ 0 5 10 15 ] |
| 55 | * [ 4 5 6 7 ] --> [ 4 9 14 3 ] |
| 56 | * [ 8 9 10 11 ] [ 8 13 2 7 ] |
| 57 | * [ 12 13 14 15 ] [ 12 1 6 11 ] |
| 58 | * |
| 59 | * and as a result, we need to apply this inverse permutation to figure out |
| 60 | * which indices to use in the doublerow function and elsewhere. |
| 61 | * |
| 62 | * [ 0 13 10 7 ] |
| 63 | * [ 4 1 14 11 ] |
| 64 | * [ 8 5 2 15 ] |
| 65 | * [ 12 9 6 3 ] |
| 66 | */ |
| 67 | |
| 68 | /* The Salsa20 quarter-round. Read from the matrix @y@ at indices @a@, @b@, |
| 69 | * @c@, and @d@; and write to the corresponding elements of @z@. |
| 70 | */ |
| 71 | #define SALSA20_QR(z, y, a, b, c, d) do { \ |
| 72 | (z)[b] = (y)[b] ^ ROL32((y)[a] + (y)[d], 7); \ |
| 73 | (z)[c] = (y)[c] ^ ROL32((z)[b] + (y)[a], 9); \ |
| 74 | (z)[d] = (y)[d] ^ ROL32((z)[c] + (z)[b], 13); \ |
| 75 | (z)[a] = (y)[a] ^ ROL32((z)[d] + (z)[c], 18); \ |
| 76 | } while (0) |
| 77 | |
| 78 | /* The Salsa20 double-round. Read from matrix @y@, writing the result to |
| 79 | * @z@. |
| 80 | */ |
| 81 | #define SALSA20_DR(z, y) do { \ |
| 82 | SALSA20_QR(z, y, 0, 4, 8, 12); \ |
| 83 | SALSA20_QR(z, y, 1, 5, 9, 13); \ |
| 84 | SALSA20_QR(z, y, 2, 6, 10, 14); \ |
| 85 | SALSA20_QR(z, y, 3, 7, 11, 15); \ |
| 86 | SALSA20_QR(z, z, 0, 13, 10, 7); \ |
| 87 | SALSA20_QR(z, z, 1, 14, 11, 4); \ |
| 88 | SALSA20_QR(z, z, 2, 15, 8, 5); \ |
| 89 | SALSA20_QR(z, z, 3, 12, 9, 6); \ |
| 90 | } while (0) |
| 91 | |
| 92 | /* The Salsa20 feedforward step, used at the end of the core function. Here, |
| 93 | * @y@ contains the original input matrix; @z@ contains the final one, and is |
| 94 | * updated. The output is rendered in canonical order, ready for output. |
| 95 | */ |
| 96 | #define SALSA20_FFWD(z, y) do { \ |
| 97 | const uint32 *_y = (y); \ |
| 98 | uint32 *_z = (z); \ |
| 99 | int _t; \ |
| 100 | _z[ 0] = _z[ 0] + _y[ 0]; _z[ 4] = _z[ 4] + _y[ 4]; \ |
| 101 | _z[ 8] = _z[ 8] + _y[ 8]; _z[12] = _z[12] + _y[12]; \ |
| 102 | _t = _z[ 1] + _y[ 1]; _z[ 1] = _z[13] + _y[13]; \ |
| 103 | _z[13] = _z[ 9] + _y[ 9]; _z[ 9] = _z[ 5] + _y[ 5]; _z[ 5] = _t; \ |
| 104 | _t = _z[ 2] + _y[ 2]; _z[ 2] = _z[10] + _y[10]; _z[10] = _t; \ |
| 105 | _t = _z[ 6] + _y[ 6]; _z[ 6] = _z[14] + _y[14]; _z[14] = _t; \ |
| 106 | _t = _z[ 3] + _y[ 3]; _z[ 3] = _z[ 7] + _y[ 7]; \ |
| 107 | _z[ 7] = _z[11] + _y[11]; _z[11] = _z[15] + _y[15]; _z[15] = _t; \ |
| 108 | } while (0) |
| 109 | |
| 110 | /* Various numbers of rounds, unrolled. Read from @y@, and write to @z@. */ |
| 111 | #define SALSA20_4R(z, y) \ |
| 112 | do { SALSA20_DR(z, y); SALSA20_DR(z, z); } while (0) |
| 113 | #define SALSA20_8R(z, y) \ |
| 114 | do { SALSA20_4R(z, y); SALSA20_4R(z, z); } while (0) |
| 115 | #define SALSA20_12R(z, y) \ |
| 116 | do { SALSA20_8R(z, y); SALSA20_4R(z, z); } while (0) |
| 117 | #define SALSA20_20R(z, y) \ |
| 118 | do { SALSA20_12R(z, y); SALSA20_8R(z, z); } while (0) |
| 119 | |
| 120 | /* Apply @n@ (must be even) rounds, rolled. (This seems to be faster, |
| 121 | * probably because it fits in cache better). Read from @y@, and write to |
| 122 | * @z@. |
| 123 | */ |
| 124 | #define SALSA20_nR(z, y, n) do { \ |
| 125 | int _i; \ |
| 126 | SALSA20_DR(z, y); \ |
| 127 | for (_i = 0; _i < (n)/2 - 1; _i++) SALSA20_DR(z, z); \ |
| 128 | } while (0) |
| 129 | |
| 130 | /* Step the counter in the Salsa20 state matrix @a@. */ |
| 131 | #define SALSA20_STEP(a) \ |
| 132 | do { (a)[8] = U32((a)[8] + 1); (a)[5] += !(a)[8]; } while (0) |
| 133 | |
| 134 | /*----- Buffering and output ----------------------------------------------* |
| 135 | * |
| 136 | * These macros are also used by ChaCha. |
| 137 | */ |
| 138 | |
| 139 | /* Copy the Salsa20 matrix @a@ to the output buffer at @d@, advancing @d@ |
| 140 | * past the new material. |
| 141 | */ |
| 142 | #define SALSA20_GENFULL(a, d) do { \ |
| 143 | int _i; \ |
| 144 | \ |
| 145 | for (_i = 0; _i < 16; _i++) { STORE32_L((d), (a)[_i]); (d) += 4; } \ |
| 146 | } while (0) |
| 147 | |
| 148 | /* XOR the contents the input buffer at @s@ with the Salsa20 matrix @a@, |
| 149 | * writing the result to @d@ and advance @s@ and @d@. |
| 150 | */ |
| 151 | #define SALSA20_MIXFULL(a, d, s) do { \ |
| 152 | uint32 _x; \ |
| 153 | int _i; \ |
| 154 | \ |
| 155 | for (_i = 0; _i < 16; _i++) { \ |
| 156 | _x = LOAD32_L(s); (s) += 4; \ |
| 157 | _x ^= (a)[_i]; \ |
| 158 | STORE32_L((d), _x); (d) += 4; \ |
| 159 | } \ |
| 160 | } while (0) |
| 161 | |
| 162 | /* Fill the context @ctx@'s buffer from the matrix @a@ in preparation for |
| 163 | * emitting partial blocks of output. |
| 164 | */ |
| 165 | #define SALSA20_PREPBUF(ctx, a) do { \ |
| 166 | int _i; \ |
| 167 | for (_i = 0; _i < 16; _i++) STORE32_L((ctx)->buf + 4*_i, (a)[_i]); \ |
| 168 | (ctx)->bufi = 0; \ |
| 169 | } while (0) |
| 170 | |
| 171 | /* Write at most @n@ bytes of buffered output from the context @ctx@ to the |
| 172 | * output buffer @d@ (if it's not null), XORing it with the input buffer @s@ |
| 173 | * (if that's not null). Both @s@ and @d@ are advanced if they aren't null; |
| 174 | * @n@ is decreased appropriately. |
| 175 | */ |
| 176 | #define SALSA20_OUTBUF(ctx, d, s, n) do { \ |
| 177 | size_t _n = (n), _left = SALSA20_OUTSZ - (ctx)->bufi; \ |
| 178 | if (_n > _left) _n = _left; \ |
| 179 | (n) -= _n; \ |
| 180 | if (!(d)) (ctx)->bufi += _n; \ |
| 181 | else if (s) while (_n--) *(d)++ = (ctx)->buf[(ctx)->bufi++] ^ *(s)++; \ |
| 182 | else while (_n--) *(d)++ = (ctx)->buf[(ctx)->bufi++]; \ |
| 183 | } while (0) |
| 184 | |
| 185 | /*----- Variants and naming -----------------------------------------------*/ |
| 186 | |
| 187 | /* Common numbers of rounds, for which we generate definitions. */ |
| 188 | #define SALSA20_VARS(_) _(8) _(12) _(20) |
| 189 | |
| 190 | /* Constructing externally-facing names. */ |
| 191 | #define SALSA20_DECOR(base, r, suff) SALSA20__DECOR_##r(base, suff) |
| 192 | #define SALSA20__DECOR_20(base, suff) GLUE(base, suff) |
| 193 | #define SALSA20__DECOR_12(base, suff) GLUE(base##12, suff) |
| 194 | #define SALSA20__DECOR_8(base, suff) GLUE(base##8, suff) |
| 195 | |
| 196 | /* Preprocessor-time table of the standard names. */ |
| 197 | #define SALSA20_NAME_20 "salsa20" |
| 198 | #define SALSA20_NAME_12 "salsa20/12" |
| 199 | #define SALSA20_NAME_8 "salsa20/8" |
| 200 | |
| 201 | /*----- That's all, folks -------------------------------------------------*/ |
| 202 | |
| 203 | #ifdef __cplusplus |
| 204 | } |
| 205 | #endif |
| 206 | |
| 207 | #endif |