Commit | Line | Data |
---|---|---|
194e93f2 MW |
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 | /* The Salsa20 quarter-round. Read from the matrix @y@ at indices @a@, @b@, | |
47 | * @c@, and @d@; and write to the corresponding elements of @z@. | |
48 | */ | |
49 | #define SALSA20_QR(z, y, a, b, c, d) do { \ | |
50 | (z)[b] = (y)[b] ^ ROL32((y)[a] + (y)[d], 7); \ | |
51 | (z)[c] = (y)[c] ^ ROL32((z)[b] + (y)[a], 9); \ | |
52 | (z)[d] = (y)[d] ^ ROL32((z)[c] + (z)[b], 13); \ | |
53 | (z)[a] = (y)[a] ^ ROL32((z)[d] + (z)[c], 18); \ | |
54 | } while (0) | |
55 | ||
56 | /* The Salsa20 double-round. Read from matrix @y@, writing the result to | |
57 | * @z@. | |
58 | */ | |
59 | #define SALSA20_DR(z, y) do { \ | |
60 | SALSA20_QR(z, y, 0, 4, 8, 12); \ | |
61 | SALSA20_QR(z, y, 5, 9, 13, 1); \ | |
62 | SALSA20_QR(z, y, 10, 14, 2, 6); \ | |
63 | SALSA20_QR(z, y, 15, 3, 7, 11); \ | |
64 | SALSA20_QR(z, z, 0, 1, 2, 3); \ | |
65 | SALSA20_QR(z, z, 5, 6, 7, 4); \ | |
66 | SALSA20_QR(z, z, 10, 11, 8, 9); \ | |
67 | SALSA20_QR(z, z, 15, 12, 13, 14); \ | |
68 | } while (0) | |
69 | ||
70 | /* The Salsa20 feedforward step, used at the end of the core function. Here, | |
71 | * @y@ contains the original input matrix; @z@ contains the final one, and is | |
72 | * updated. | |
73 | */ | |
74 | #define SALSA20_FFWD(z, y) do { \ | |
75 | int _i; \ | |
76 | for (_i = 0; _i < 16; _i++) (z)[_i] += (y)[_i]; \ | |
77 | } while (0) | |
78 | ||
79 | /* Various numbers of rounds, unrolled. Read from @y@, and write to @z@. */ | |
80 | #define SALSA20_4R(z, y) \ | |
81 | do { SALSA20_DR(z, y); SALSA20_DR(z, z); } while (0) | |
82 | #define SALSA20_8R(z, y) \ | |
83 | do { SALSA20_4R(z, y); SALSA20_4R(z, z); } while (0) | |
84 | #define SALSA20_12R(z, y) \ | |
85 | do { SALSA20_8R(z, y); SALSA20_4R(z, z); } while (0) | |
86 | #define SALSA20_20R(z, y) \ | |
87 | do { SALSA20_12R(z, y); SALSA20_8R(z, z); } while (0) | |
88 | ||
89 | /* Apply @n@ (must be even) rounds, rolled. (This seems to be faster, | |
90 | * probably because it fits in cache better). Read from @y@, and write to | |
91 | * @z@. | |
92 | */ | |
93 | #define SALSA20_nR(z, y, n) do { \ | |
94 | int _i; \ | |
95 | SALSA20_DR(z, y); \ | |
96 | for (_i = 0; _i < (n)/2 - 1; _i++) SALSA20_DR(z, z); \ | |
97 | } while (0) | |
98 | ||
99 | /* Step the counter in the Salsa20 state matrix @a@. */ | |
100 | #define SALSA20_STEP(a) \ | |
101 | do { (a)[8] = U32((a)[8] + 1); (a)[9] += !(a)[8]; } while (0) | |
102 | ||
103 | /*----- Buffering and output ----------------------------------------------* | |
104 | * | |
105 | * These macros are also used by ChaCha. | |
106 | */ | |
107 | ||
108 | /* Copy the Salsa20 matrix @a@ to the output buffer at @d@, advancing @d@ | |
109 | * past the new material. | |
110 | */ | |
111 | #define SALSA20_GENFULL(a, d) do { \ | |
112 | int _i; \ | |
113 | \ | |
114 | for (_i = 0; _i < 16; _i++) { STORE32_L((d), (a)[_i]); (d) += 4; } \ | |
115 | } while (0) | |
116 | ||
117 | /* XOR the contents the input buffer at @s@ with the Salsa20 matrix @a@, | |
118 | * writing the result to @d@ and advance @s@ and @d@. | |
119 | */ | |
120 | #define SALSA20_MIXFULL(a, d, s) do { \ | |
121 | uint32 _x; \ | |
122 | int _i; \ | |
123 | \ | |
124 | for (_i = 0; _i < 16; _i++) { \ | |
125 | _x = LOAD32_L(s); (s) += 4; \ | |
126 | _x ^= (a)[_i]; \ | |
127 | STORE32_L((d), _x); (d) += 4; \ | |
128 | } \ | |
129 | } while (0) | |
130 | ||
131 | /* Fill the context @ctx@'s buffer from the matrix @a@ in preparation for | |
132 | * emitting partial blocks of output. | |
133 | */ | |
134 | #define SALSA20_PREPBUF(ctx, a) do { \ | |
135 | int _i; \ | |
136 | for (_i = 0; _i < 16; _i++) STORE32_L((ctx)->buf + 4*_i, (a)[_i]); \ | |
137 | (ctx)->bufi = 0; \ | |
138 | } while (0) | |
139 | ||
140 | /* Write at most @n@ bytes of buffered output from the context @ctx@ to the | |
141 | * output buffer @d@ (if it's not null), XORing it with the input buffer @s@ | |
142 | * (if that's not null). Both @s@ and @d@ are advanced if they aren't null; | |
143 | * @n@ is decreased appropriately. | |
144 | */ | |
145 | #define SALSA20_OUTBUF(ctx, d, s, n) do { \ | |
146 | size_t _n = (n), _left = SALSA20_OUTSZ - (ctx)->bufi; \ | |
147 | if (_n > _left) _n = _left; \ | |
148 | (n) -= _n; \ | |
149 | if (!(d)) (ctx)->bufi += _n; \ | |
150 | else if (s) while (_n--) *(d)++ = (ctx)->buf[(ctx)->bufi++] ^ *(s)++; \ | |
151 | else while (_n--) *(d)++ = (ctx)->buf[(ctx)->bufi++]; \ | |
152 | } while (0) | |
153 | ||
154 | /*----- Variants and naming -----------------------------------------------*/ | |
155 | ||
156 | /* Common numbers of rounds, for which we generate definitions. */ | |
157 | #define SALSA20_VARS(_) _(8) _(12) _(20) | |
158 | ||
159 | /* Constructing externally-facing names. */ | |
160 | #define SALSA20_DECOR(base, r, suff) SALSA20__DECOR_##r(base, suff) | |
161 | #define SALSA20__DECOR_20(base, suff) GLUE(base, suff) | |
162 | #define SALSA20__DECOR_12(base, suff) GLUE(base##12, suff) | |
163 | #define SALSA20__DECOR_8(base, suff) GLUE(base##8, suff) | |
164 | ||
165 | /* Preprocessor-time table of the standard names. */ | |
166 | #define SALSA20_NAME_20 "salsa20" | |
167 | #define SALSA20_NAME_12 "salsa20/12" | |
168 | #define SALSA20_NAME_8 "salsa20/8" | |
169 | ||
170 | /*----- That's all, folks -------------------------------------------------*/ | |
171 | ||
172 | #ifdef __cplusplus | |
173 | } | |
174 | #endif | |
175 | ||
176 | #endif |