1///////////////////////////////////////////////////////////////////////////////
2//
3/// \file       sha256.c
4/// \brief      SHA-256
5///
6/// \todo       Crypto++ has x86 ASM optimizations. They use SSE so if they
7///             are imported to liblzma, SSE instructions need to be used
8///             conditionally to keep the code working on older boxes.
9//
10//  This code is based on the code found from 7-Zip, which has a modified
11//  version of the SHA-256 found from Crypto++ <http://www.cryptopp.com/>.
12//  The code was modified a little to fit into liblzma.
13//
14//  Authors:    Kevin Springle
15//              Wei Dai
16//              Igor Pavlov
17//              Lasse Collin
18//
19//  This file has been put into the public domain.
20//  You can do whatever you want with this file.
21//
22///////////////////////////////////////////////////////////////////////////////
23
24// Avoid bogus warnings in transform().
25#if (__GNUC__ == 4 && __GNUC_MINOR__ >= 2) || __GNUC__ > 4
26#	pragma GCC diagnostic ignored "-Wuninitialized"
27#endif
28
29#include "check.h"
30
31// At least on x86, GCC is able to optimize this to a rotate instruction.
32#define rotr_32(num, amount) ((num) >> (amount) | (num) << (32 - (amount)))
33
34#define blk0(i) (W[i] = data[i])
35#define blk2(i) (W[i & 15] += s1(W[(i - 2) & 15]) + W[(i - 7) & 15] \
36		+ s0(W[(i - 15) & 15]))
37
38#define Ch(x, y, z) (z ^ (x & (y ^ z)))
39#define Maj(x, y, z) ((x & y) | (z & (x | y)))
40
41#define a(i) T[(0 - i) & 7]
42#define b(i) T[(1 - i) & 7]
43#define c(i) T[(2 - i) & 7]
44#define d(i) T[(3 - i) & 7]
45#define e(i) T[(4 - i) & 7]
46#define f(i) T[(5 - i) & 7]
47#define g(i) T[(6 - i) & 7]
48#define h(i) T[(7 - i) & 7]
49
50#define R(i) \
51	h(i) += S1(e(i)) + Ch(e(i), f(i), g(i)) + SHA256_K[i + j] \
52		+ (j ? blk2(i) : blk0(i)); \
53	d(i) += h(i); \
54	h(i) += S0(a(i)) + Maj(a(i), b(i), c(i))
55
56#define S0(x) (rotr_32(x, 2) ^ rotr_32(x, 13) ^ rotr_32(x, 22))
57#define S1(x) (rotr_32(x, 6) ^ rotr_32(x, 11) ^ rotr_32(x, 25))
58#define s0(x) (rotr_32(x, 7) ^ rotr_32(x, 18) ^ (x >> 3))
59#define s1(x) (rotr_32(x, 17) ^ rotr_32(x, 19) ^ (x >> 10))
60
61
62static const uint32_t SHA256_K[64] = {
63	0x428A2F98, 0x71374491, 0xB5C0FBCF, 0xE9B5DBA5,
64	0x3956C25B, 0x59F111F1, 0x923F82A4, 0xAB1C5ED5,
65	0xD807AA98, 0x12835B01, 0x243185BE, 0x550C7DC3,
66	0x72BE5D74, 0x80DEB1FE, 0x9BDC06A7, 0xC19BF174,
67	0xE49B69C1, 0xEFBE4786, 0x0FC19DC6, 0x240CA1CC,
68	0x2DE92C6F, 0x4A7484AA, 0x5CB0A9DC, 0x76F988DA,
69	0x983E5152, 0xA831C66D, 0xB00327C8, 0xBF597FC7,
70	0xC6E00BF3, 0xD5A79147, 0x06CA6351, 0x14292967,
71	0x27B70A85, 0x2E1B2138, 0x4D2C6DFC, 0x53380D13,
72	0x650A7354, 0x766A0ABB, 0x81C2C92E, 0x92722C85,
73	0xA2BFE8A1, 0xA81A664B, 0xC24B8B70, 0xC76C51A3,
74	0xD192E819, 0xD6990624, 0xF40E3585, 0x106AA070,
75	0x19A4C116, 0x1E376C08, 0x2748774C, 0x34B0BCB5,
76	0x391C0CB3, 0x4ED8AA4A, 0x5B9CCA4F, 0x682E6FF3,
77	0x748F82EE, 0x78A5636F, 0x84C87814, 0x8CC70208,
78	0x90BEFFFA, 0xA4506CEB, 0xBEF9A3F7, 0xC67178F2,
79};
80
81
82static void
83transform(uint32_t state[static 8], const uint32_t data[static 16])
84{
85	uint32_t W[16];
86	uint32_t T[8];
87
88	// Copy state[] to working vars.
89	memcpy(T, state, sizeof(T));
90
91	// 64 operations, partially loop unrolled
92	for (unsigned int j = 0; j < 64; j += 16) {
93		R( 0); R( 1); R( 2); R( 3);
94		R( 4); R( 5); R( 6); R( 7);
95		R( 8); R( 9); R(10); R(11);
96		R(12); R(13); R(14); R(15);
97	}
98
99	// Add the working vars back into state[].
100	state[0] += a(0);
101	state[1] += b(0);
102	state[2] += c(0);
103	state[3] += d(0);
104	state[4] += e(0);
105	state[5] += f(0);
106	state[6] += g(0);
107	state[7] += h(0);
108}
109
110
111static void
112process(lzma_check_state *check)
113{
114#ifdef WORDS_BIGENDIAN
115	transform(check->state.sha256.state, check->buffer.u32);
116
117#else
118	uint32_t data[16];
119
120	for (size_t i = 0; i < 16; ++i)
121		data[i] = bswap32(check->buffer.u32[i]);
122
123	transform(check->state.sha256.state, data);
124#endif
125
126	return;
127}
128
129
130extern void
131lzma_sha256_init(lzma_check_state *check)
132{
133	static const uint32_t s[8] = {
134		0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A,
135		0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19,
136	};
137
138	memcpy(check->state.sha256.state, s, sizeof(s));
139	check->state.sha256.size = 0;
140
141	return;
142}
143
144
145extern void
146lzma_sha256_update(const uint8_t *buf, size_t size, lzma_check_state *check)
147{
148	// Copy the input data into a properly aligned temporary buffer.
149	// This way we can be called with arbitrarily sized buffers
150	// (no need to be multiple of 64 bytes), and the code works also
151	// on architectures that don't allow unaligned memory access.
152	while (size > 0) {
153		const size_t copy_start = check->state.sha256.size & 0x3F;
154		size_t copy_size = 64 - copy_start;
155		if (copy_size > size)
156			copy_size = size;
157
158		memcpy(check->buffer.u8 + copy_start, buf, copy_size);
159
160		buf += copy_size;
161		size -= copy_size;
162		check->state.sha256.size += copy_size;
163
164		if ((check->state.sha256.size & 0x3F) == 0)
165			process(check);
166	}
167
168	return;
169}
170
171
172extern void
173lzma_sha256_finish(lzma_check_state *check)
174{
175	// Add padding as described in RFC 3174 (it describes SHA-1 but
176	// the same padding style is used for SHA-256 too).
177	size_t pos = check->state.sha256.size & 0x3F;
178	check->buffer.u8[pos++] = 0x80;
179
180	while (pos != 64 - 8) {
181		if (pos == 64) {
182			process(check);
183			pos = 0;
184		}
185
186		check->buffer.u8[pos++] = 0x00;
187	}
188
189	// Convert the message size from bytes to bits.
190	check->state.sha256.size *= 8;
191
192	check->buffer.u64[(64 - 8) / 8] = conv64be(check->state.sha256.size);
193
194	process(check);
195
196	for (size_t i = 0; i < 8; ++i)
197		check->buffer.u32[i] = conv32be(check->state.sha256.state[i]);
198
199	return;
200}
201