2 * Copyright 2015-2016 The OpenSSL Project Authors. All Rights Reserved.
4 * Licensed under the Apache License 2.0 (the "License"). You may not use
5 * this file except in compliance with the License. You can obtain a copy
6 * in the file LICENSE in the source distribution or at
7 * https://www.openssl.org/source/license.html
10 /* Adapted from the public domain code by D. Bernstein from SUPERCOP. */
14 #include "crypto/chacha.h"
15 #include "crypto/ctype.h"
17 typedef unsigned int u32;
18 typedef unsigned char u8;
24 # define ROTATE(v, n) (((v) << (n)) | ((v) >> (32 - (n))))
26 # define U32TO8_LITTLE(p, v) do { \
27 (p)[0] = (u8)(v >> 0); \
28 (p)[1] = (u8)(v >> 8); \
29 (p)[2] = (u8)(v >> 16); \
30 (p)[3] = (u8)(v >> 24); \
33 /* QUARTERROUND updates a, b, c, d with a ChaCha "quarter" round. */
34 # define QUARTERROUND(a,b,c,d) ( \
35 x[a] += x[b], x[d] = ROTATE((x[d] ^ x[a]),16), \
36 x[c] += x[d], x[b] = ROTATE((x[b] ^ x[c]),12), \
37 x[a] += x[b], x[d] = ROTATE((x[d] ^ x[a]), 8), \
38 x[c] += x[d], x[b] = ROTATE((x[b] ^ x[c]), 7) )
40 /* chacha_core performs 20 rounds of ChaCha on the input words in
41 * |input| and writes the 64 output bytes to |output|. */
42 static void chacha20_core(chacha_buf *output, const u32 input[16])
51 memcpy(x, input, sizeof(x));
53 for (i = 20; i > 0; i -= 2) {
54 QUARTERROUND(0, 4, 8, 12);
55 QUARTERROUND(1, 5, 9, 13);
56 QUARTERROUND(2, 6, 10, 14);
57 QUARTERROUND(3, 7, 11, 15);
58 QUARTERROUND(0, 5, 10, 15);
59 QUARTERROUND(1, 6, 11, 12);
60 QUARTERROUND(2, 7, 8, 13);
61 QUARTERROUND(3, 4, 9, 14);
64 if (is_endian.little) {
65 for (i = 0; i < 16; ++i)
66 output->u[i] = x[i] + input[i];
68 for (i = 0; i < 16; ++i)
69 U32TO8_LITTLE(output->c + 4 * i, (x[i] + input[i]));
73 void ChaCha20_ctr32(unsigned char *out, const unsigned char *inp,
74 size_t len, const unsigned int key[8],
75 const unsigned int counter[4])
81 /* sigma constant "expand 32-byte k" in little-endian encoding */
82 input[0] = ((u32)ossl_toascii('e')) | ((u32)ossl_toascii('x') << 8)
83 | ((u32)ossl_toascii('p') << 16)
84 | ((u32)ossl_toascii('a') << 24);
85 input[1] = ((u32)ossl_toascii('n')) | ((u32)ossl_toascii('d') << 8)
86 | ((u32)ossl_toascii(' ') << 16)
87 | ((u32)ossl_toascii('3') << 24);
88 input[2] = ((u32)ossl_toascii('2')) | ((u32)ossl_toascii('-') << 8)
89 | ((u32)ossl_toascii('b') << 16)
90 | ((u32)ossl_toascii('y') << 24);
91 input[3] = ((u32)ossl_toascii('t')) | ((u32)ossl_toascii('e') << 8)
92 | ((u32)ossl_toascii(' ') << 16)
93 | ((u32)ossl_toascii('k') << 24);
104 input[12] = counter[0];
105 input[13] = counter[1];
106 input[14] = counter[2];
107 input[15] = counter[3];
114 chacha20_core(&buf, input);
116 for (i = 0; i < todo; i++)
117 out[i] = inp[i] ^ buf.c[i];
123 * Advance 32-bit counter. Note that as subroutine is so to
124 * say nonce-agnostic, this limited counter width doesn't
125 * prevent caller from implementing wider counter. It would
126 * simply take two calls split on counter overflow...