1 /* Copyright (c) 2014, Google Inc.
2 *
3 * Permission to use, copy, modify, and/or distribute this software for any
4 * purpose with or without fee is hereby granted, provided that the above
5 * copyright notice and this permission notice appear in all copies.
6 *
7 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
8 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
9 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
10 * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
11 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
12 * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
13 * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */
14
15 // This implementation of poly1305 is by Andrew Moon
16 // (https://github.com/floodyberry/poly1305-donna) and released as public
17 // domain.
18
19 #include <openssl/poly1305.h>
20
21 #include <assert.h>
22 #include <string.h>
23
24 #include "internal.h"
25 #include "../internal.h"
26
27
28 #if !defined(BORINGSSL_HAS_UINT128) || !defined(OPENSSL_X86_64)
29
mul32x32_64(uint32_t a,uint32_t b)30 static uint64_t mul32x32_64(uint32_t a, uint32_t b) { return (uint64_t)a * b; }
31
32 struct poly1305_state_st {
33 uint32_t r0, r1, r2, r3, r4;
34 uint32_t s1, s2, s3, s4;
35 uint32_t h0, h1, h2, h3, h4;
36 uint8_t buf[16];
37 size_t buf_used;
38 uint8_t key[16];
39 };
40
41 static_assert(
42 sizeof(struct poly1305_state_st) + 63 <= sizeof(poly1305_state),
43 "poly1305_state isn't large enough to hold aligned poly1305_state_st");
44
poly1305_aligned_state(poly1305_state * state)45 static inline struct poly1305_state_st *poly1305_aligned_state(
46 poly1305_state *state) {
47 return align_pointer(state, 64);
48 }
49
50 // poly1305_blocks updates |state| given some amount of input data. This
51 // function may only be called with a |len| that is not a multiple of 16 at the
52 // end of the data. Otherwise the input must be buffered into 16 byte blocks.
poly1305_update(struct poly1305_state_st * state,const uint8_t * in,size_t len)53 static void poly1305_update(struct poly1305_state_st *state, const uint8_t *in,
54 size_t len) {
55 uint32_t t0, t1, t2, t3;
56 uint64_t t[5];
57 uint32_t b;
58 uint64_t c;
59 size_t j;
60 uint8_t mp[16];
61
62 if (len < 16) {
63 goto poly1305_donna_atmost15bytes;
64 }
65
66 poly1305_donna_16bytes:
67 t0 = CRYPTO_load_u32_le(in);
68 t1 = CRYPTO_load_u32_le(in + 4);
69 t2 = CRYPTO_load_u32_le(in + 8);
70 t3 = CRYPTO_load_u32_le(in + 12);
71
72 in += 16;
73 len -= 16;
74
75 state->h0 += t0 & 0x3ffffff;
76 state->h1 += ((((uint64_t)t1 << 32) | t0) >> 26) & 0x3ffffff;
77 state->h2 += ((((uint64_t)t2 << 32) | t1) >> 20) & 0x3ffffff;
78 state->h3 += ((((uint64_t)t3 << 32) | t2) >> 14) & 0x3ffffff;
79 state->h4 += (t3 >> 8) | (1 << 24);
80
81 poly1305_donna_mul:
82 t[0] = mul32x32_64(state->h0, state->r0) + mul32x32_64(state->h1, state->s4) +
83 mul32x32_64(state->h2, state->s3) + mul32x32_64(state->h3, state->s2) +
84 mul32x32_64(state->h4, state->s1);
85 t[1] = mul32x32_64(state->h0, state->r1) + mul32x32_64(state->h1, state->r0) +
86 mul32x32_64(state->h2, state->s4) + mul32x32_64(state->h3, state->s3) +
87 mul32x32_64(state->h4, state->s2);
88 t[2] = mul32x32_64(state->h0, state->r2) + mul32x32_64(state->h1, state->r1) +
89 mul32x32_64(state->h2, state->r0) + mul32x32_64(state->h3, state->s4) +
90 mul32x32_64(state->h4, state->s3);
91 t[3] = mul32x32_64(state->h0, state->r3) + mul32x32_64(state->h1, state->r2) +
92 mul32x32_64(state->h2, state->r1) + mul32x32_64(state->h3, state->r0) +
93 mul32x32_64(state->h4, state->s4);
94 t[4] = mul32x32_64(state->h0, state->r4) + mul32x32_64(state->h1, state->r3) +
95 mul32x32_64(state->h2, state->r2) + mul32x32_64(state->h3, state->r1) +
96 mul32x32_64(state->h4, state->r0);
97
98 state->h0 = (uint32_t)t[0] & 0x3ffffff;
99 c = (t[0] >> 26);
100 t[1] += c;
101 state->h1 = (uint32_t)t[1] & 0x3ffffff;
102 b = (uint32_t)(t[1] >> 26);
103 t[2] += b;
104 state->h2 = (uint32_t)t[2] & 0x3ffffff;
105 b = (uint32_t)(t[2] >> 26);
106 t[3] += b;
107 state->h3 = (uint32_t)t[3] & 0x3ffffff;
108 b = (uint32_t)(t[3] >> 26);
109 t[4] += b;
110 state->h4 = (uint32_t)t[4] & 0x3ffffff;
111 b = (uint32_t)(t[4] >> 26);
112 state->h0 += b * 5;
113
114 if (len >= 16) {
115 goto poly1305_donna_16bytes;
116 }
117
118 // final bytes
119 poly1305_donna_atmost15bytes:
120 if (!len) {
121 return;
122 }
123
124 for (j = 0; j < len; j++) {
125 mp[j] = in[j];
126 }
127 mp[j++] = 1;
128 for (; j < 16; j++) {
129 mp[j] = 0;
130 }
131 len = 0;
132
133 t0 = CRYPTO_load_u32_le(mp + 0);
134 t1 = CRYPTO_load_u32_le(mp + 4);
135 t2 = CRYPTO_load_u32_le(mp + 8);
136 t3 = CRYPTO_load_u32_le(mp + 12);
137
138 state->h0 += t0 & 0x3ffffff;
139 state->h1 += ((((uint64_t)t1 << 32) | t0) >> 26) & 0x3ffffff;
140 state->h2 += ((((uint64_t)t2 << 32) | t1) >> 20) & 0x3ffffff;
141 state->h3 += ((((uint64_t)t3 << 32) | t2) >> 14) & 0x3ffffff;
142 state->h4 += (t3 >> 8);
143
144 goto poly1305_donna_mul;
145 }
146
CRYPTO_poly1305_init(poly1305_state * statep,const uint8_t key[32])147 void CRYPTO_poly1305_init(poly1305_state *statep, const uint8_t key[32]) {
148 struct poly1305_state_st *state = poly1305_aligned_state(statep);
149 uint32_t t0, t1, t2, t3;
150
151 #if defined(OPENSSL_POLY1305_NEON)
152 if (CRYPTO_is_NEON_capable()) {
153 CRYPTO_poly1305_init_neon(statep, key);
154 return;
155 }
156 #endif
157
158 t0 = CRYPTO_load_u32_le(key + 0);
159 t1 = CRYPTO_load_u32_le(key + 4);
160 t2 = CRYPTO_load_u32_le(key + 8);
161 t3 = CRYPTO_load_u32_le(key + 12);
162
163 // precompute multipliers
164 state->r0 = t0 & 0x3ffffff;
165 t0 >>= 26;
166 t0 |= t1 << 6;
167 state->r1 = t0 & 0x3ffff03;
168 t1 >>= 20;
169 t1 |= t2 << 12;
170 state->r2 = t1 & 0x3ffc0ff;
171 t2 >>= 14;
172 t2 |= t3 << 18;
173 state->r3 = t2 & 0x3f03fff;
174 t3 >>= 8;
175 state->r4 = t3 & 0x00fffff;
176
177 state->s1 = state->r1 * 5;
178 state->s2 = state->r2 * 5;
179 state->s3 = state->r3 * 5;
180 state->s4 = state->r4 * 5;
181
182 // init state
183 state->h0 = 0;
184 state->h1 = 0;
185 state->h2 = 0;
186 state->h3 = 0;
187 state->h4 = 0;
188
189 state->buf_used = 0;
190 OPENSSL_memcpy(state->key, key + 16, sizeof(state->key));
191 }
192
CRYPTO_poly1305_update(poly1305_state * statep,const uint8_t * in,size_t in_len)193 void CRYPTO_poly1305_update(poly1305_state *statep, const uint8_t *in,
194 size_t in_len) {
195 struct poly1305_state_st *state = poly1305_aligned_state(statep);
196
197 // Work around a C language bug. See https://crbug.com/1019588.
198 if (in_len == 0) {
199 return;
200 }
201
202 #if defined(OPENSSL_POLY1305_NEON)
203 if (CRYPTO_is_NEON_capable()) {
204 CRYPTO_poly1305_update_neon(statep, in, in_len);
205 return;
206 }
207 #endif
208
209 if (state->buf_used) {
210 size_t todo = 16 - state->buf_used;
211 if (todo > in_len) {
212 todo = in_len;
213 }
214 for (size_t i = 0; i < todo; i++) {
215 state->buf[state->buf_used + i] = in[i];
216 }
217 state->buf_used += todo;
218 in_len -= todo;
219 in += todo;
220
221 if (state->buf_used == 16) {
222 poly1305_update(state, state->buf, 16);
223 state->buf_used = 0;
224 }
225 }
226
227 if (in_len >= 16) {
228 size_t todo = in_len & ~0xf;
229 poly1305_update(state, in, todo);
230 in += todo;
231 in_len &= 0xf;
232 }
233
234 if (in_len) {
235 for (size_t i = 0; i < in_len; i++) {
236 state->buf[i] = in[i];
237 }
238 state->buf_used = in_len;
239 }
240 }
241
CRYPTO_poly1305_finish(poly1305_state * statep,uint8_t mac[16])242 void CRYPTO_poly1305_finish(poly1305_state *statep, uint8_t mac[16]) {
243 struct poly1305_state_st *state = poly1305_aligned_state(statep);
244 uint32_t g0, g1, g2, g3, g4;
245 uint32_t b, nb;
246
247 #if defined(OPENSSL_POLY1305_NEON)
248 if (CRYPTO_is_NEON_capable()) {
249 CRYPTO_poly1305_finish_neon(statep, mac);
250 return;
251 }
252 #endif
253
254 if (state->buf_used) {
255 poly1305_update(state, state->buf, state->buf_used);
256 }
257
258 b = state->h0 >> 26;
259 state->h0 = state->h0 & 0x3ffffff;
260 state->h1 += b;
261 b = state->h1 >> 26;
262 state->h1 = state->h1 & 0x3ffffff;
263 state->h2 += b;
264 b = state->h2 >> 26;
265 state->h2 = state->h2 & 0x3ffffff;
266 state->h3 += b;
267 b = state->h3 >> 26;
268 state->h3 = state->h3 & 0x3ffffff;
269 state->h4 += b;
270 b = state->h4 >> 26;
271 state->h4 = state->h4 & 0x3ffffff;
272 state->h0 += b * 5;
273
274 g0 = state->h0 + 5;
275 b = g0 >> 26;
276 g0 &= 0x3ffffff;
277 g1 = state->h1 + b;
278 b = g1 >> 26;
279 g1 &= 0x3ffffff;
280 g2 = state->h2 + b;
281 b = g2 >> 26;
282 g2 &= 0x3ffffff;
283 g3 = state->h3 + b;
284 b = g3 >> 26;
285 g3 &= 0x3ffffff;
286 g4 = state->h4 + b - (1 << 26);
287
288 b = (g4 >> 31) - 1;
289 nb = ~b;
290 state->h0 = (state->h0 & nb) | (g0 & b);
291 state->h1 = (state->h1 & nb) | (g1 & b);
292 state->h2 = (state->h2 & nb) | (g2 & b);
293 state->h3 = (state->h3 & nb) | (g3 & b);
294 state->h4 = (state->h4 & nb) | (g4 & b);
295
296 uint64_t f0 = ((state->h0) | (state->h1 << 26)) +
297 (uint64_t)CRYPTO_load_u32_le(&state->key[0]);
298 uint64_t f1 = ((state->h1 >> 6) | (state->h2 << 20)) +
299 (uint64_t)CRYPTO_load_u32_le(&state->key[4]);
300 uint64_t f2 = ((state->h2 >> 12) | (state->h3 << 14)) +
301 (uint64_t)CRYPTO_load_u32_le(&state->key[8]);
302 uint64_t f3 = ((state->h3 >> 18) | (state->h4 << 8)) +
303 (uint64_t)CRYPTO_load_u32_le(&state->key[12]);
304
305 CRYPTO_store_u32_le(&mac[0], (uint32_t)f0);
306 f1 += (f0 >> 32);
307 CRYPTO_store_u32_le(&mac[4], (uint32_t)f1);
308 f2 += (f1 >> 32);
309 CRYPTO_store_u32_le(&mac[8], (uint32_t)f2);
310 f3 += (f2 >> 32);
311 CRYPTO_store_u32_le(&mac[12], (uint32_t)f3);
312 }
313
314 #endif // !BORINGSSL_HAS_UINT128 || !OPENSSL_X86_64
315