Debian Codegraph @

external/boringssl/crypto/ec/p224-64.c

raw
1 /* Copyright (c) 2015, 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 /* A 64-bit implementation of the NIST P-224 elliptic curve point multiplication
16 *
17 * Inspired by Daniel J. Bernstein's public domain nistp224 implementation
18 * and Adam Langley's public domain 64-bit C implementation of curve25519. */
19
20 #include <openssl/base.h>
21
22 #if defined(OPENSSL_64_BIT) && !defined(OPENSSL_WINDOWS) && \
23 !defined(OPENSSL_SMALL)
24
25 #include <openssl/bn.h>
26 #include <openssl/ec.h>
27 #include <openssl/err.h>
28 #include <openssl/mem.h>
29
30 #include <string.h>
31
32 #include "internal.h"
33 #include "../internal.h"
34
35
36 typedef uint8_t u8;
37 typedef uint64_t u64;
38 typedef int64_t s64;
39
40 /* Field elements are represented as a_0 + 2^56*a_1 + 2^112*a_2 + 2^168*a_3
41 * using 64-bit coefficients called 'limbs', and sometimes (for multiplication
42 * results) as b_0 + 2^56*b_1 + 2^112*b_2 + 2^168*b_3 + 2^224*b_4 + 2^280*b_5 +
43 * 2^336*b_6 using 128-bit coefficients called 'widelimbs'. A 4-limb
44 * representation is an 'felem'; a 7-widelimb representation is a 'widefelem'.
45 * Even within felems, bits of adjacent limbs overlap, and we don't always
46 * reduce the representations: we ensure that inputs to each felem
47 * multiplication satisfy a_i < 2^60, so outputs satisfy b_i < 4*2^60*2^60, and
48 * fit into a 128-bit word without overflow. The coefficients are then again
49 * partially reduced to obtain an felem satisfying a_i < 2^57. We only reduce
50 * to the unique minimal representation at the end of the computation. */
51
52 typedef uint64_t limb;
53 typedef uint128_t widelimb;
54
55 typedef limb felem[4];
56 typedef widelimb widefelem[7];
57
58 /* Field element represented as a byte arrary. 28*8 = 224 bits is also the
59 * group order size for the elliptic curve, and we also use this type for
60 * scalars for point multiplication. */
61 typedef u8 felem_bytearray[28];
62
63 /* Precomputed multiples of the standard generator
64 * Points are given in coordinates (X, Y, Z) where Z normally is 1
65 * (0 for the point at infinity).
66 * For each field element, slice a_0 is word 0, etc.
67 *
68 * The table has 2 * 16 elements, starting with the following:
69 * index | bits | point
70 * ------+---------+------------------------------
71 * 0 | 0 0 0 0 | 0G
72 * 1 | 0 0 0 1 | 1G
73 * 2 | 0 0 1 0 | 2^56G
74 * 3 | 0 0 1 1 | (2^56 + 1)G
75 * 4 | 0 1 0 0 | 2^112G
76 * 5 | 0 1 0 1 | (2^112 + 1)G
77 * 6 | 0 1 1 0 | (2^112 + 2^56)G
78 * 7 | 0 1 1 1 | (2^112 + 2^56 + 1)G
79 * 8 | 1 0 0 0 | 2^168G
80 * 9 | 1 0 0 1 | (2^168 + 1)G
81 * 10 | 1 0 1 0 | (2^168 + 2^56)G
82 * 11 | 1 0 1 1 | (2^168 + 2^56 + 1)G
83 * 12 | 1 1 0 0 | (2^168 + 2^112)G
84 * 13 | 1 1 0 1 | (2^168 + 2^112 + 1)G
85 * 14 | 1 1 1 0 | (2^168 + 2^112 + 2^56)G
86 * 15 | 1 1 1 1 | (2^168 + 2^112 + 2^56 + 1)G
87 * followed by a copy of this with each element multiplied by 2^28.
88 *
89 * The reason for this is so that we can clock bits into four different
90 * locations when doing simple scalar multiplies against the base point,
91 * and then another four locations using the second 16 elements. */
92 static const felem g_pre_comp[2][16][3] = {
93 {{{0, 0, 0, 0}, {0, 0, 0, 0}, {0, 0, 0, 0}},
94 {{0x3280d6115c1d21, 0xc1d356c2112234, 0x7f321390b94a03, 0xb70e0cbd6bb4bf},
95 {0xd5819985007e34, 0x75a05a07476444, 0xfb4c22dfe6cd43, 0xbd376388b5f723},
96 {1, 0, 0, 0}},
97 {{0xfd9675666ebbe9, 0xbca7664d40ce5e, 0x2242df8d8a2a43, 0x1f49bbb0f99bc5},
98 {0x29e0b892dc9c43, 0xece8608436e662, 0xdc858f185310d0, 0x9812dd4eb8d321},
99 {1, 0, 0, 0}},
100 {{0x6d3e678d5d8eb8, 0x559eed1cb362f1, 0x16e9a3bbce8a3f, 0xeedcccd8c2a748},
101 {0xf19f90ed50266d, 0xabf2b4bf65f9df, 0x313865468fafec, 0x5cb379ba910a17},
102 {1, 0, 0, 0}},
103 {{0x0641966cab26e3, 0x91fb2991fab0a0, 0xefec27a4e13a0b, 0x0499aa8a5f8ebe},
104 {0x7510407766af5d, 0x84d929610d5450, 0x81d77aae82f706, 0x6916f6d4338c5b},
105 {1, 0, 0, 0}},
106 {{0xea95ac3b1f15c6, 0x086000905e82d4, 0xdd323ae4d1c8b1, 0x932b56be7685a3},
107 {0x9ef93dea25dbbf, 0x41665960f390f0, 0xfdec76dbe2a8a7, 0x523e80f019062a},
108 {1, 0, 0, 0}},
109 {{0x822fdd26732c73, 0xa01c83531b5d0f, 0x363f37347c1ba4, 0xc391b45c84725c},
110 {0xbbd5e1b2d6ad24, 0xddfbcde19dfaec, 0xc393da7e222a7f, 0x1efb7890ede244},
111 {1, 0, 0, 0}},
112 {{0x4c9e90ca217da1, 0xd11beca79159bb, 0xff8d33c2c98b7c, 0x2610b39409f849},
113 {0x44d1352ac64da0, 0xcdbb7b2c46b4fb, 0x966c079b753c89, 0xfe67e4e820b112},
114 {1, 0, 0, 0}},
115 {{0xe28cae2df5312d, 0xc71b61d16f5c6e, 0x79b7619a3e7c4c, 0x05c73240899b47},
116 {0x9f7f6382c73e3a, 0x18615165c56bda, 0x641fab2116fd56, 0x72855882b08394},
117 {1, 0, 0, 0}},
118 {{0x0469182f161c09, 0x74a98ca8d00fb5, 0xb89da93489a3e0, 0x41c98768fb0c1d},
119 {0xe5ea05fb32da81, 0x3dce9ffbca6855, 0x1cfe2d3fbf59e6, 0x0e5e03408738a7},
120 {1, 0, 0, 0}},
121 {{0xdab22b2333e87f, 0x4430137a5dd2f6, 0xe03ab9f738beb8, 0xcb0c5d0dc34f24},
122 {0x764a7df0c8fda5, 0x185ba5c3fa2044, 0x9281d688bcbe50, 0xc40331df893881},
123 {1, 0, 0, 0}},
124 {{0xb89530796f0f60, 0xade92bd26909a3, 0x1a0c83fb4884da, 0x1765bf22a5a984},
125 {0x772a9ee75db09e, 0x23bc6c67cec16f, 0x4c1edba8b14e2f, 0xe2a215d9611369},
126 {1, 0, 0, 0}},
127 {{0x571e509fb5efb3, 0xade88696410552, 0xc8ae85fada74fe, 0x6c7e4be83bbde3},
128 {0xff9f51160f4652, 0xb47ce2495a6539, 0xa2946c53b582f4, 0x286d2db3ee9a60},
129 {1, 0, 0, 0}},
130 {{0x40bbd5081a44af, 0x0995183b13926c, 0xbcefba6f47f6d0, 0x215619e9cc0057},
131 {0x8bc94d3b0df45e, 0xf11c54a3694f6f, 0x8631b93cdfe8b5, 0xe7e3f4b0982db9},
132 {1, 0, 0, 0}},
133 {{0xb17048ab3e1c7b, 0xac38f36ff8a1d8, 0x1c29819435d2c6, 0xc813132f4c07e9},
134 {0x2891425503b11f, 0x08781030579fea, 0xf5426ba5cc9674, 0x1e28ebf18562bc},
135 {1, 0, 0, 0}},
136 {{0x9f31997cc864eb, 0x06cd91d28b5e4c, 0xff17036691a973, 0xf1aef351497c58},
137 {0xdd1f2d600564ff, 0xdead073b1402db, 0x74a684435bd693, 0xeea7471f962558},
138 {1, 0, 0, 0}}},
139 {{{0, 0, 0, 0}, {0, 0, 0, 0}, {0, 0, 0, 0}},
140 {{0x9665266dddf554, 0x9613d78b60ef2d, 0xce27a34cdba417, 0xd35ab74d6afc31},
141 {0x85ccdd22deb15e, 0x2137e5783a6aab, 0xa141cffd8c93c6, 0x355a1830e90f2d},
142 {1, 0, 0, 0}},
143 {{0x1a494eadaade65, 0xd6da4da77fe53c, 0xe7992996abec86, 0x65c3553c6090e3},
144 {0xfa610b1fb09346, 0xf1c6540b8a4aaf, 0xc51a13ccd3cbab, 0x02995b1b18c28a},
145 {1, 0, 0, 0}},
146 {{0x7874568e7295ef, 0x86b419fbe38d04, 0xdc0690a7550d9a, 0xd3966a44beac33},
147 {0x2b7280ec29132f, 0xbeaa3b6a032df3, 0xdc7dd88ae41200, 0xd25e2513e3a100},
148 {1, 0, 0, 0}},
149 {{0x924857eb2efafd, 0xac2bce41223190, 0x8edaa1445553fc, 0x825800fd3562d5},
150 {0x8d79148ea96621, 0x23a01c3dd9ed8d, 0xaf8b219f9416b5, 0xd8db0cc277daea},
151 {1, 0, 0, 0}},
152 {{0x76a9c3b1a700f0, 0xe9acd29bc7e691, 0x69212d1a6b0327, 0x6322e97fe154be},
153 {0x469fc5465d62aa, 0x8d41ed18883b05, 0x1f8eae66c52b88, 0xe4fcbe9325be51},
154 {1, 0, 0, 0}},
155 {{0x825fdf583cac16, 0x020b857c7b023a, 0x683c17744b0165, 0x14ffd0a2daf2f1},
156 {0x323b36184218f9, 0x4944ec4e3b47d4, 0xc15b3080841acf, 0x0bced4b01a28bb},
157 {1, 0, 0, 0}},
158 {{0x92ac22230df5c4, 0x52f33b4063eda8, 0xcb3f19870c0c93, 0x40064f2ba65233},
159 {0xfe16f0924f8992, 0x012da25af5b517, 0x1a57bb24f723a6, 0x06f8bc76760def},
160 {1, 0, 0, 0}},
161 {{0x4a7084f7817cb9, 0xbcab0738ee9a78, 0x3ec11e11d9c326, 0xdc0fe90e0f1aae},
162 {0xcf639ea5f98390, 0x5c350aa22ffb74, 0x9afae98a4047b7, 0x956ec2d617fc45},
163 {1, 0, 0, 0}},
164 {{0x4306d648c1be6a, 0x9247cd8bc9a462, 0xf5595e377d2f2e, 0xbd1c3caff1a52e},
165 {0x045e14472409d0, 0x29f3e17078f773, 0x745a602b2d4f7d, 0x191837685cdfbb},
166 {1, 0, 0, 0}},
167 {{0x5b6ee254a8cb79, 0x4953433f5e7026, 0xe21faeb1d1def4, 0xc4c225785c09de},
168 {0x307ce7bba1e518, 0x31b125b1036db8, 0x47e91868839e8f, 0xc765866e33b9f3},
169 {1, 0, 0, 0}},
170 {{0x3bfece24f96906, 0x4794da641e5093, 0xde5df64f95db26, 0x297ecd89714b05},
171 {0x701bd3ebb2c3aa, 0x7073b4f53cb1d5, 0x13c5665658af16, 0x9895089d66fe58},
172 {1, 0, 0, 0}},
173 {{0x0fef05f78c4790, 0x2d773633b05d2e, 0x94229c3a951c94, 0xbbbd70df4911bb},
174 {0xb2c6963d2c1168, 0x105f47a72b0d73, 0x9fdf6111614080, 0x7b7e94b39e67b0},
175 {1, 0, 0, 0}},
176 {{0xad1a7d6efbe2b3, 0xf012482c0da69d, 0x6b3bdf12438345, 0x40d7558d7aa4d9},
177 {0x8a09fffb5c6d3d, 0x9a356e5d9ffd38, 0x5973f15f4f9b1c, 0xdcd5f59f63c3ea},
178 {1, 0, 0, 0}},
179 {{0xacf39f4c5ca7ab, 0x4c8071cc5fd737, 0xc64e3602cd1184, 0x0acd4644c9abba},
180 {0x6c011a36d8bf6e, 0xfecd87ba24e32a, 0x19f6f56574fad8, 0x050b204ced9405},
181 {1, 0, 0, 0}},
182 {{0xed4f1cae7d9a96, 0x5ceef7ad94c40a, 0x778e4a3bf3ef9b, 0x7405783dc3b55e},
183 {0x32477c61b6e8c6, 0xb46a97570f018b, 0x91176d0a7e95d1, 0x3df90fbc4c7d0e},
184 {1, 0, 0, 0}}}};
185
186 /* Helper functions to convert field elements to/from internal representation */
187 static void bin28_to_felem(felem out, const u8 in[28]) {
188 #if defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
189 out[0] = (*((const uint64_t *)(in + 20))) << 8 >> 8;
190 out[1] = (*((const uint64_t *)(in + 14))) >> 8;
191 out[2] = (*((const uint64_t *)(in + 7))) >> 8;
192 out[3] = *((const uint64_t *)(in)) >>8;
193 #else
194 out[0] = *((const uint64_t *)(in)) & 0x00ffffffffffffff;
195 out[1] = (*((const uint64_t *)(in + 7))) & 0x00ffffffffffffff;
196 out[2] = (*((const uint64_t *)(in + 14))) & 0x00ffffffffffffff;
197 out[3] = (*((const uint64_t *)(in + 20))) >> 8;
198 #endif
199 }
200
201 static void felem_to_bin28(u8 out[28], const felem in) {
202 size_t i;
203 for (i = 0; i < 7; ++i) {
204 #if defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
205 out[i] = *((u8 *)&in[3] + i + 1);
206 out[i + 7] = *((u8 *)&in[2] + i + 1);
207 out[i + 14] = *((u8 *)&in[1] + i + 1);
208 out[i + 21] = *((u8 *)&in[0] + i + 1);
209 #else
210 out[i] = in[0] >> (8 * i);
211 out[i + 7] = in[1] >> (8 * i);
212 out[i + 14] = in[2] >> (8 * i);
213 out[i + 21] = in[3] >> (8 * i);
214 #endif
215 }
216 }
217
218 /* To preserve endianness when using BN_bn2bin and BN_bin2bn */
219 static void flip_endian(u8 *out, const u8 *in, size_t len) {
220 size_t i;
221 for (i = 0; i < len; ++i) {
222 out[i] = in[len - 1 - i];
223 }
224 }
225
226 /* From OpenSSL BIGNUM to internal representation */
227 static int BN_to_felem(felem out, const BIGNUM *bn) {
228 /* BN_bn2bin eats leading zeroes */
229 felem_bytearray b_out;
230 memset(b_out, 0, sizeof(b_out));
231 size_t num_bytes = BN_num_bytes(bn);
232 if (num_bytes > sizeof(b_out) ||
233 BN_is_negative(bn)) {
234 OPENSSL_PUT_ERROR(EC, EC_R_BIGNUM_OUT_OF_RANGE);
235 return 0;
236 }
237
238 felem_bytearray b_in;
239 num_bytes = BN_bn2bin(bn, b_in);
240 #if defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
241 memcpy(b_out+sizeof(b_out)-num_bytes, b_in, num_bytes);
242 memset(b_out, 0, sizeof(b_out)-num_bytes);
243 #else
244 flip_endian(b_out, b_in, num_bytes);
245 #endif
246 bin28_to_felem(out, b_out);
247 return 1;
248 }
249
250 /* From internal representation to OpenSSL BIGNUM */
251 static BIGNUM *felem_to_BN(BIGNUM *out, const felem in) {
252 felem_bytearray b_out;
253 #if defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
254 felem_to_bin28(b_out, in);
255 #else
256 felem_bytearray b_in;
257 felem_to_bin28(b_in, in);
258 flip_endian(b_out, b_in, sizeof(b_out));
259 #endif
260 return BN_bin2bn(b_out, sizeof(b_out), out);
261 }
262
263 /* Field operations, using the internal representation of field elements.
264 * NB! These operations are specific to our point multiplication and cannot be
265 * expected to be correct in general - e.g., multiplication with a large scalar
266 * will cause an overflow. */
267
268 static void felem_assign(felem out, const felem in) {
269 out[0] = in[0];
270 out[1] = in[1];
271 out[2] = in[2];
272 out[3] = in[3];
273 }
274
275 /* Sum two field elements: out += in */
276 static void felem_sum(felem out, const felem in) {
277 out[0] += in[0];
278 out[1] += in[1];
279 out[2] += in[2];
280 out[3] += in[3];
281 }
282
283 /* Get negative value: out = -in */
284 /* Assumes in[i] < 2^57 */
285 static void felem_neg(felem out, const felem in) {
286 static const limb two58p2 = (((limb)1) << 58) + (((limb)1) << 2);
287 static const limb two58m2 = (((limb)1) << 58) - (((limb)1) << 2);
288 static const limb two58m42m2 =
289 (((limb)1) << 58) - (((limb)1) << 42) - (((limb)1) << 2);
290
291 /* Set to 0 mod 2^224-2^96+1 to ensure out > in */
292 out[0] = two58p2 - in[0];
293 out[1] = two58m42m2 - in[1];
294 out[2] = two58m2 - in[2];
295 out[3] = two58m2 - in[3];
296 }
297
298 /* Subtract field elements: out -= in */
299 /* Assumes in[i] < 2^57 */
300 static void felem_diff(felem out, const felem in) {
301 static const limb two58p2 = (((limb)1) << 58) + (((limb)1) << 2);
302 static const limb two58m2 = (((limb)1) << 58) - (((limb)1) << 2);
303 static const limb two58m42m2 =
304 (((limb)1) << 58) - (((limb)1) << 42) - (((limb)1) << 2);
305
306 /* Add 0 mod 2^224-2^96+1 to ensure out > in */
307 out[0] += two58p2;
308 out[1] += two58m42m2;
309 out[2] += two58m2;
310 out[3] += two58m2;
311
312 out[0] -= in[0];
313 out[1] -= in[1];
314 out[2] -= in[2];
315 out[3] -= in[3];
316 }
317
318 /* Subtract in unreduced 128-bit mode: out -= in */
319 /* Assumes in[i] < 2^119 */
320 static void widefelem_diff(widefelem out, const widefelem in) {
321 static const widelimb two120 = ((widelimb)1) << 120;
322 static const widelimb two120m64 =
323 (((widelimb)1) << 120) - (((widelimb)1) << 64);
324 static const widelimb two120m104m64 =
325 (((widelimb)1) << 120) - (((widelimb)1) << 104) - (((widelimb)1) << 64);
326
327 /* Add 0 mod 2^224-2^96+1 to ensure out > in */
328 out[0] += two120;
329 out[1] += two120m64;
330 out[2] += two120m64;
331 out[3] += two120;
332 out[4] += two120m104m64;
333 out[5] += two120m64;
334 out[6] += two120m64;
335
336 out[0] -= in[0];
337 out[1] -= in[1];
338 out[2] -= in[2];
339 out[3] -= in[3];
340 out[4] -= in[4];
341 out[5] -= in[5];
342 out[6] -= in[6];
343 }
344
345 /* Subtract in mixed mode: out128 -= in64 */
346 /* in[i] < 2^63 */
347 static void felem_diff_128_64(widefelem out, const felem in) {
348 static const widelimb two64p8 = (((widelimb)1) << 64) + (((widelimb)1) << 8);
349 static const widelimb two64m8 = (((widelimb)1) << 64) - (((widelimb)1) << 8);
350 static const widelimb two64m48m8 =
351 (((widelimb)1) << 64) - (((widelimb)1) << 48) - (((widelimb)1) << 8);
352
353 /* Add 0 mod 2^224-2^96+1 to ensure out > in */
354 out[0] += two64p8;
355 out[1] += two64m48m8;
356 out[2] += two64m8;
357 out[3] += two64m8;
358
359 out[0] -= in[0];
360 out[1] -= in[1];
361 out[2] -= in[2];
362 out[3] -= in[3];
363 }
364
365 /* Multiply a field element by a scalar: out = out * scalar
366 * The scalars we actually use are small, so results fit without overflow */
367 static void felem_scalar(felem out, const limb scalar) {
368 out[0] *= scalar;
369 out[1] *= scalar;
370 out[2] *= scalar;
371 out[3] *= scalar;
372 }
373
374 /* Multiply an unreduced field element by a scalar: out = out * scalar
375 * The scalars we actually use are small, so results fit without overflow */
376 static void widefelem_scalar(widefelem out, const widelimb scalar) {
377 out[0] *= scalar;
378 out[1] *= scalar;
379 out[2] *= scalar;
380 out[3] *= scalar;
381 out[4] *= scalar;
382 out[5] *= scalar;
383 out[6] *= scalar;
384 }
385
386 /* Square a field element: out = in^2 */
387 static void felem_square(widefelem out, const felem in) {
388 limb tmp0, tmp1, tmp2;
389 tmp0 = 2 * in[0];
390 tmp1 = 2 * in[1];
391 tmp2 = 2 * in[2];
392 out[0] = ((widelimb)in[0]) * in[0];
393 out[1] = ((widelimb)in[0]) * tmp1;
394 out[2] = ((widelimb)in[0]) * tmp2 + ((widelimb)in[1]) * in[1];
395 out[3] = ((widelimb)in[3]) * tmp0 + ((widelimb)in[1]) * tmp2;
396 out[4] = ((widelimb)in[3]) * tmp1 + ((widelimb)in[2]) * in[2];
397 out[5] = ((widelimb)in[3]) * tmp2;
398 out[6] = ((widelimb)in[3]) * in[3];
399 }
400
401 /* Multiply two field elements: out = in1 * in2 */
402 static void felem_mul(widefelem out, const felem in1, const felem in2) {
403 out[0] = ((widelimb)in1[0]) * in2[0];
404 out[1] = ((widelimb)in1[0]) * in2[1] + ((widelimb)in1[1]) * in2[0];
405 out[2] = ((widelimb)in1[0]) * in2[2] + ((widelimb)in1[1]) * in2[1] +
406 ((widelimb)in1[2]) * in2[0];
407 out[3] = ((widelimb)in1[0]) * in2[3] + ((widelimb)in1[1]) * in2[2] +
408 ((widelimb)in1[2]) * in2[1] + ((widelimb)in1[3]) * in2[0];
409 out[4] = ((widelimb)in1[1]) * in2[3] + ((widelimb)in1[2]) * in2[2] +
410 ((widelimb)in1[3]) * in2[1];
411 out[5] = ((widelimb)in1[2]) * in2[3] + ((widelimb)in1[3]) * in2[2];
412 out[6] = ((widelimb)in1[3]) * in2[3];
413 }
414
415 /* Reduce seven 128-bit coefficients to four 64-bit coefficients.
416 * Requires in[i] < 2^126,
417 * ensures out[0] < 2^56, out[1] < 2^56, out[2] < 2^56, out[3] <= 2^56 + 2^16 */
418 static void felem_reduce(felem out, const widefelem in) {
419 static const widelimb two127p15 =
420 (((widelimb)1) << 127) + (((widelimb)1) << 15);
421 static const widelimb two127m71 =
422 (((widelimb)1) << 127) - (((widelimb)1) << 71);
423 static const widelimb two127m71m55 =
424 (((widelimb)1) << 127) - (((widelimb)1) << 71) - (((widelimb)1) << 55);
425 widelimb output[5];
426
427 /* Add 0 mod 2^224-2^96+1 to ensure all differences are positive */
428 output[0] = in[0] + two127p15;
429 output[1] = in[1] + two127m71m55;
430 output[2] = in[2] + two127m71;
431 output[3] = in[3];
432 output[4] = in[4];
433
434 /* Eliminate in[4], in[5], in[6] */
435 output[4] += in[6] >> 16;
436 output[3] += (in[6] & 0xffff) << 40;
437 output[2] -= in[6];
438
439 output[3] += in[5] >> 16;
440 output[2] += (in[5] & 0xffff) << 40;
441 output[1] -= in[5];
442
443 output[2] += output[4] >> 16;
444 output[1] += (output[4] & 0xffff) << 40;
445 output[0] -= output[4];
446
447 /* Carry 2 -> 3 -> 4 */
448 output[3] += output[2] >> 56;
449 output[2] &= 0x00ffffffffffffff;
450
451 output[4] = output[3] >> 56;
452 output[3] &= 0x00ffffffffffffff;
453
454 /* Now output[2] < 2^56, output[3] < 2^56, output[4] < 2^72 */
455
456 /* Eliminate output[4] */
457 output[2] += output[4] >> 16;
458 /* output[2] < 2^56 + 2^56 = 2^57 */
459 output[1] += (output[4] & 0xffff) << 40;
460 output[0] -= output[4];
461
462 /* Carry 0 -> 1 -> 2 -> 3 */
463 output[1] += output[0] >> 56;
464 out[0] = output[0] & 0x00ffffffffffffff;
465
466 output[2] += output[1] >> 56;
467 /* output[2] < 2^57 + 2^72 */
468 out[1] = output[1] & 0x00ffffffffffffff;
469 output[3] += output[2] >> 56;
470 /* output[3] <= 2^56 + 2^16 */
471 out[2] = output[2] & 0x00ffffffffffffff;
472
473 /* out[0] < 2^56, out[1] < 2^56, out[2] < 2^56,
474 * out[3] <= 2^56 + 2^16 (due to final carry),
475 * so out < 2*p */
476 out[3] = output[3];
477 }
478
479 /* Reduce to unique minimal representation.
480 * Requires 0 <= in < 2*p (always call felem_reduce first) */
481 static void felem_contract(felem out, const felem in) {
482 static const int64_t two56 = ((limb)1) << 56;
483 /* 0 <= in < 2*p, p = 2^224 - 2^96 + 1 */
484 /* if in > p , reduce in = in - 2^224 + 2^96 - 1 */
485 int64_t tmp[4], a;
486 tmp[0] = in[0];
487 tmp[1] = in[1];
488 tmp[2] = in[2];
489 tmp[3] = in[3];
490 /* Case 1: a = 1 iff in >= 2^224 */
491 a = (in[3] >> 56);
492 tmp[0] -= a;
493 tmp[1] += a << 40;
494 tmp[3] &= 0x00ffffffffffffff;
495 /* Case 2: a = 0 iff p <= in < 2^224, i.e., the high 128 bits are all 1 and
496 * the lower part is non-zero */
497 a = ((in[3] & in[2] & (in[1] | 0x000000ffffffffff)) + 1) |
498 (((int64_t)(in[0] + (in[1] & 0x000000ffffffffff)) - 1) >> 63);
499 a &= 0x00ffffffffffffff;
500 /* turn a into an all-one mask (if a = 0) or an all-zero mask */
501 a = (a - 1) >> 63;
502 /* subtract 2^224 - 2^96 + 1 if a is all-one */
503 tmp[3] &= a ^ 0xffffffffffffffff;
504 tmp[2] &= a ^ 0xffffffffffffffff;
505 tmp[1] &= (a ^ 0xffffffffffffffff) | 0x000000ffffffffff;
506 tmp[0] -= 1 & a;
507
508 /* eliminate negative coefficients: if tmp[0] is negative, tmp[1] must
509 * be non-zero, so we only need one step */
510 a = tmp[0] >> 63;
511 tmp[0] += two56 & a;
512 tmp[1] -= 1 & a;
513
514 /* carry 1 -> 2 -> 3 */
515 tmp[2] += tmp[1] >> 56;
516 tmp[1] &= 0x00ffffffffffffff;
517
518 tmp[3] += tmp[2] >> 56;
519 tmp[2] &= 0x00ffffffffffffff;
520
521 /* Now 0 <= out < p */
522 out[0] = tmp[0];
523 out[1] = tmp[1];
524 out[2] = tmp[2];
525 out[3] = tmp[3];
526 }
527
528 /* Zero-check: returns 1 if input is 0, and 0 otherwise. We know that field
529 * elements are reduced to in < 2^225, so we only need to check three cases: 0,
530 * 2^224 - 2^96 + 1, and 2^225 - 2^97 + 2 */
531 static limb felem_is_zero(const felem in) {
532 limb zero = in[0] | in[1] | in[2] | in[3];
533 zero = (((int64_t)(zero)-1) >> 63) & 1;
534
535 limb two224m96p1 = (in[0] ^ 1) | (in[1] ^ 0x00ffff0000000000) |
536 (in[2] ^ 0x00ffffffffffffff) |
537 (in[3] ^ 0x00ffffffffffffff);
538 two224m96p1 = (((int64_t)(two224m96p1)-1) >> 63) & 1;
539 limb two225m97p2 = (in[0] ^ 2) | (in[1] ^ 0x00fffe0000000000) |
540 (in[2] ^ 0x00ffffffffffffff) |
541 (in[3] ^ 0x01ffffffffffffff);
542 two225m97p2 = (((int64_t)(two225m97p2)-1) >> 63) & 1;
543 return (zero | two224m96p1 | two225m97p2);
544 }
545
546 /* Invert a field element */
547 /* Computation chain copied from djb's code */
548 static void felem_inv(felem out, const felem in) {
549 felem ftmp, ftmp2, ftmp3, ftmp4;
550 widefelem tmp;
551 size_t i;
552
553 felem_square(tmp, in);
554 felem_reduce(ftmp, tmp); /* 2 */
555 felem_mul(tmp, in, ftmp);
556 felem_reduce(ftmp, tmp); /* 2^2 - 1 */
557 felem_square(tmp, ftmp);
558 felem_reduce(ftmp, tmp); /* 2^3 - 2 */
559 felem_mul(tmp, in, ftmp);
560 felem_reduce(ftmp, tmp); /* 2^3 - 1 */
561 felem_square(tmp, ftmp);
562 felem_reduce(ftmp2, tmp); /* 2^4 - 2 */
563 felem_square(tmp, ftmp2);
564 felem_reduce(ftmp2, tmp); /* 2^5 - 4 */
565 felem_square(tmp, ftmp2);
566 felem_reduce(ftmp2, tmp); /* 2^6 - 8 */
567 felem_mul(tmp, ftmp2, ftmp);
568 felem_reduce(ftmp, tmp); /* 2^6 - 1 */
569 felem_square(tmp, ftmp);
570 felem_reduce(ftmp2, tmp); /* 2^7 - 2 */
571 for (i = 0; i < 5; ++i) { /* 2^12 - 2^6 */
572 felem_square(tmp, ftmp2);
573 felem_reduce(ftmp2, tmp);
574 }
575 felem_mul(tmp, ftmp2, ftmp);
576 felem_reduce(ftmp2, tmp); /* 2^12 - 1 */
577 felem_square(tmp, ftmp2);
578 felem_reduce(ftmp3, tmp); /* 2^13 - 2 */
579 for (i = 0; i < 11; ++i) {/* 2^24 - 2^12 */
580 felem_square(tmp, ftmp3);
581 felem_reduce(ftmp3, tmp);
582 }
583 felem_mul(tmp, ftmp3, ftmp2);
584 felem_reduce(ftmp2, tmp); /* 2^24 - 1 */
585 felem_square(tmp, ftmp2);
586 felem_reduce(ftmp3, tmp); /* 2^25 - 2 */
587 for (i = 0; i < 23; ++i) {/* 2^48 - 2^24 */
588 felem_square(tmp, ftmp3);
589 felem_reduce(ftmp3, tmp);
590 }
591 felem_mul(tmp, ftmp3, ftmp2);
592 felem_reduce(ftmp3, tmp); /* 2^48 - 1 */
593 felem_square(tmp, ftmp3);
594 felem_reduce(ftmp4, tmp); /* 2^49 - 2 */
595 for (i = 0; i < 47; ++i) {/* 2^96 - 2^48 */
596 felem_square(tmp, ftmp4);
597 felem_reduce(ftmp4, tmp);
598 }
599 felem_mul(tmp, ftmp3, ftmp4);
600 felem_reduce(ftmp3, tmp); /* 2^96 - 1 */
601 felem_square(tmp, ftmp3);
602 felem_reduce(ftmp4, tmp); /* 2^97 - 2 */
603 for (i = 0; i < 23; ++i) {/* 2^120 - 2^24 */
604 felem_square(tmp, ftmp4);
605 felem_reduce(ftmp4, tmp);
606 }
607 felem_mul(tmp, ftmp2, ftmp4);
608 felem_reduce(ftmp2, tmp); /* 2^120 - 1 */
609 for (i = 0; i < 6; ++i) { /* 2^126 - 2^6 */
610 felem_square(tmp, ftmp2);
611 felem_reduce(ftmp2, tmp);
612 }
613 felem_mul(tmp, ftmp2, ftmp);
614 felem_reduce(ftmp, tmp); /* 2^126 - 1 */
615 felem_square(tmp, ftmp);
616 felem_reduce(ftmp, tmp); /* 2^127 - 2 */
617 felem_mul(tmp, ftmp, in);
618 felem_reduce(ftmp, tmp); /* 2^127 - 1 */
619 for (i = 0; i < 97; ++i) {/* 2^224 - 2^97 */
620 felem_square(tmp, ftmp);
621 felem_reduce(ftmp, tmp);
622 }
623 felem_mul(tmp, ftmp, ftmp3);
624 felem_reduce(out, tmp); /* 2^224 - 2^96 - 1 */
625 }
626
627 /* Copy in constant time:
628 * if icopy == 1, copy in to out,
629 * if icopy == 0, copy out to itself. */
630 static void copy_conditional(felem out, const felem in, limb icopy) {
631 size_t i;
632 /* icopy is a (64-bit) 0 or 1, so copy is either all-zero or all-one */
633 const limb copy = -icopy;
634 for (i = 0; i < 4; ++i) {
635 const limb tmp = copy & (in[i] ^ out[i]);
636 out[i] ^= tmp;
637 }
638 }
639
640 /* ELLIPTIC CURVE POINT OPERATIONS
641 *
642 * Points are represented in Jacobian projective coordinates:
643 * (X, Y, Z) corresponds to the affine point (X/Z^2, Y/Z^3),
644 * or to the point at infinity if Z == 0. */
645
646 /* Double an elliptic curve point:
647 * (X', Y', Z') = 2 * (X, Y, Z), where
648 * X' = (3 * (X - Z^2) * (X + Z^2))^2 - 8 * X * Y^2
649 * Y' = 3 * (X - Z^2) * (X + Z^2) * (4 * X * Y^2 - X') - 8 * Y^2
650 * Z' = (Y + Z)^2 - Y^2 - Z^2 = 2 * Y * Z
651 * Outputs can equal corresponding inputs, i.e., x_out == x_in is allowed,
652 * while x_out == y_in is not (maybe this works, but it's not tested). */
653 static void point_double(felem x_out, felem y_out, felem z_out,
654 const felem x_in, const felem y_in, const felem z_in) {
655 widefelem tmp, tmp2;
656 felem delta, gamma, beta, alpha, ftmp, ftmp2;
657
658 felem_assign(ftmp, x_in);
659 felem_assign(ftmp2, x_in);
660
661 /* delta = z^2 */
662 felem_square(tmp, z_in);
663 felem_reduce(delta, tmp);
664
665 /* gamma = y^2 */
666 felem_square(tmp, y_in);
667 felem_reduce(gamma, tmp);
668
669 /* beta = x*gamma */
670 felem_mul(tmp, x_in, gamma);
671 felem_reduce(beta, tmp);
672
673 /* alpha = 3*(x-delta)*(x+delta) */
674 felem_diff(ftmp, delta);
675 /* ftmp[i] < 2^57 + 2^58 + 2 < 2^59 */
676 felem_sum(ftmp2, delta);
677 /* ftmp2[i] < 2^57 + 2^57 = 2^58 */
678 felem_scalar(ftmp2, 3);
679 /* ftmp2[i] < 3 * 2^58 < 2^60 */
680 felem_mul(tmp, ftmp, ftmp2);
681 /* tmp[i] < 2^60 * 2^59 * 4 = 2^121 */
682 felem_reduce(alpha, tmp);
683
684 /* x' = alpha^2 - 8*beta */
685 felem_square(tmp, alpha);
686 /* tmp[i] < 4 * 2^57 * 2^57 = 2^116 */
687 felem_assign(ftmp, beta);
688 felem_scalar(ftmp, 8);
689 /* ftmp[i] < 8 * 2^57 = 2^60 */
690 felem_diff_128_64(tmp, ftmp);
691 /* tmp[i] < 2^116 + 2^64 + 8 < 2^117 */
692 felem_reduce(x_out, tmp);
693
694 /* z' = (y + z)^2 - gamma - delta */
695 felem_sum(delta, gamma);
696 /* delta[i] < 2^57 + 2^57 = 2^58 */
697 felem_assign(ftmp, y_in);
698 felem_sum(ftmp, z_in);
699 /* ftmp[i] < 2^57 + 2^57 = 2^58 */
700 felem_square(tmp, ftmp);
701 /* tmp[i] < 4 * 2^58 * 2^58 = 2^118 */
702 felem_diff_128_64(tmp, delta);
703 /* tmp[i] < 2^118 + 2^64 + 8 < 2^119 */
704 felem_reduce(z_out, tmp);
705
706 /* y' = alpha*(4*beta - x') - 8*gamma^2 */
707 felem_scalar(beta, 4);
708 /* beta[i] < 4 * 2^57 = 2^59 */
709 felem_diff(beta, x_out);
710 /* beta[i] < 2^59 + 2^58 + 2 < 2^60 */
711 felem_mul(tmp, alpha, beta);
712 /* tmp[i] < 4 * 2^57 * 2^60 = 2^119 */
713 felem_square(tmp2, gamma);
714 /* tmp2[i] < 4 * 2^57 * 2^57 = 2^116 */
715 widefelem_scalar(tmp2, 8);
716 /* tmp2[i] < 8 * 2^116 = 2^119 */
717 widefelem_diff(tmp, tmp2);
718 /* tmp[i] < 2^119 + 2^120 < 2^121 */
719 felem_reduce(y_out, tmp);
720 }
721
722 /* Add two elliptic curve points:
723 * (X_1, Y_1, Z_1) + (X_2, Y_2, Z_2) = (X_3, Y_3, Z_3), where
724 * X_3 = (Z_1^3 * Y_2 - Z_2^3 * Y_1)^2 - (Z_1^2 * X_2 - Z_2^2 * X_1)^3 -
725 * 2 * Z_2^2 * X_1 * (Z_1^2 * X_2 - Z_2^2 * X_1)^2
726 * Y_3 = (Z_1^3 * Y_2 - Z_2^3 * Y_1) * (Z_2^2 * X_1 * (Z_1^2 * X_2 - Z_2^2 *
727 * X_1)^2 - X_3) -
728 * Z_2^3 * Y_1 * (Z_1^2 * X_2 - Z_2^2 * X_1)^3
729 * Z_3 = (Z_1^2 * X_2 - Z_2^2 * X_1) * (Z_1 * Z_2)
730 *
731 * This runs faster if 'mixed' is set, which requires Z_2 = 1 or Z_2 = 0. */
732
733 /* This function is not entirely constant-time: it includes a branch for
734 * checking whether the two input points are equal, (while not equal to the
735 * point at infinity). This case never happens during single point
736 * multiplication, so there is no timing leak for ECDH or ECDSA signing. */
737 static void point_add(felem x3, felem y3, felem z3, const felem x1,
738 const felem y1, const felem z1, const int mixed,
739 const felem x2, const felem y2, const felem z2) {
740 felem ftmp, ftmp2, ftmp3, ftmp4, ftmp5, x_out, y_out, z_out;
741 widefelem tmp, tmp2;
742 limb z1_is_zero, z2_is_zero, x_equal, y_equal;
743
744 if (!mixed) {
745 /* ftmp2 = z2^2 */
746 felem_square(tmp, z2);
747 felem_reduce(ftmp2, tmp);
748
749 /* ftmp4 = z2^3 */
750 felem_mul(tmp, ftmp2, z2);
751 felem_reduce(ftmp4, tmp);
752
753 /* ftmp4 = z2^3*y1 */
754 felem_mul(tmp2, ftmp4, y1);
755 felem_reduce(ftmp4, tmp2);
756
757 /* ftmp2 = z2^2*x1 */
758 felem_mul(tmp2, ftmp2, x1);
759 felem_reduce(ftmp2, tmp2);
760 } else {
761 /* We'll assume z2 = 1 (special case z2 = 0 is handled later) */
762
763 /* ftmp4 = z2^3*y1 */
764 felem_assign(ftmp4, y1);
765
766 /* ftmp2 = z2^2*x1 */
767 felem_assign(ftmp2, x1);
768 }
769
770 /* ftmp = z1^2 */
771 felem_square(tmp, z1);
772 felem_reduce(ftmp, tmp);
773
774 /* ftmp3 = z1^3 */
775 felem_mul(tmp, ftmp, z1);
776 felem_reduce(ftmp3, tmp);
777
778 /* tmp = z1^3*y2 */
779 felem_mul(tmp, ftmp3, y2);
780 /* tmp[i] < 4 * 2^57 * 2^57 = 2^116 */
781
782 /* ftmp3 = z1^3*y2 - z2^3*y1 */
783 felem_diff_128_64(tmp, ftmp4);
784 /* tmp[i] < 2^116 + 2^64 + 8 < 2^117 */
785 felem_reduce(ftmp3, tmp);
786
787 /* tmp = z1^2*x2 */
788 felem_mul(tmp, ftmp, x2);
789 /* tmp[i] < 4 * 2^57 * 2^57 = 2^116 */
790
791 /* ftmp = z1^2*x2 - z2^2*x1 */
792 felem_diff_128_64(tmp, ftmp2);
793 /* tmp[i] < 2^116 + 2^64 + 8 < 2^117 */
794 felem_reduce(ftmp, tmp);
795
796 /* the formulae are incorrect if the points are equal
797 * so we check for this and do doubling if this happens */
798 x_equal = felem_is_zero(ftmp);
799 y_equal = felem_is_zero(ftmp3);
800 z1_is_zero = felem_is_zero(z1);
801 z2_is_zero = felem_is_zero(z2);
802 /* In affine coordinates, (X_1, Y_1) == (X_2, Y_2) */
803 if (x_equal && y_equal && !z1_is_zero && !z2_is_zero) {
804 point_double(x3, y3, z3, x1, y1, z1);
805 return;
806 }
807
808 /* ftmp5 = z1*z2 */
809 if (!mixed) {
810 felem_mul(tmp, z1, z2);
811 felem_reduce(ftmp5, tmp);
812 } else {
813 /* special case z2 = 0 is handled later */
814 felem_assign(ftmp5, z1);
815 }
816
817 /* z_out = (z1^2*x2 - z2^2*x1)*(z1*z2) */
818 felem_mul(tmp, ftmp, ftmp5);
819 felem_reduce(z_out, tmp);
820
821 /* ftmp = (z1^2*x2 - z2^2*x1)^2 */
822 felem_assign(ftmp5, ftmp);
823 felem_square(tmp, ftmp);
824 felem_reduce(ftmp, tmp);
825
826 /* ftmp5 = (z1^2*x2 - z2^2*x1)^3 */
827 felem_mul(tmp, ftmp, ftmp5);
828 felem_reduce(ftmp5, tmp);
829
830 /* ftmp2 = z2^2*x1*(z1^2*x2 - z2^2*x1)^2 */
831 felem_mul(tmp, ftmp2, ftmp);
832 felem_reduce(ftmp2, tmp);
833
834 /* tmp = z2^3*y1*(z1^2*x2 - z2^2*x1)^3 */
835 felem_mul(tmp, ftmp4, ftmp5);
836 /* tmp[i] < 4 * 2^57 * 2^57 = 2^116 */
837
838 /* tmp2 = (z1^3*y2 - z2^3*y1)^2 */
839 felem_square(tmp2, ftmp3);
840 /* tmp2[i] < 4 * 2^57 * 2^57 < 2^116 */
841
842 /* tmp2 = (z1^3*y2 - z2^3*y1)^2 - (z1^2*x2 - z2^2*x1)^3 */
843 felem_diff_128_64(tmp2, ftmp5);
844 /* tmp2[i] < 2^116 + 2^64 + 8 < 2^117 */
845
846 /* ftmp5 = 2*z2^2*x1*(z1^2*x2 - z2^2*x1)^2 */
847 felem_assign(ftmp5, ftmp2);
848 felem_scalar(ftmp5, 2);
849 /* ftmp5[i] < 2 * 2^57 = 2^58 */
850
851 /* x_out = (z1^3*y2 - z2^3*y1)^2 - (z1^2*x2 - z2^2*x1)^3 -
852 2*z2^2*x1*(z1^2*x2 - z2^2*x1)^2 */
853 felem_diff_128_64(tmp2, ftmp5);
854 /* tmp2[i] < 2^117 + 2^64 + 8 < 2^118 */
855 felem_reduce(x_out, tmp2);
856
857 /* ftmp2 = z2^2*x1*(z1^2*x2 - z2^2*x1)^2 - x_out */
858 felem_diff(ftmp2, x_out);
859 /* ftmp2[i] < 2^57 + 2^58 + 2 < 2^59 */
860
861 /* tmp2 = (z1^3*y2 - z2^3*y1)*(z2^2*x1*(z1^2*x2 - z2^2*x1)^2 - x_out) */
862 felem_mul(tmp2, ftmp3, ftmp2);
863 /* tmp2[i] < 4 * 2^57 * 2^59 = 2^118 */
864
865 /* y_out = (z1^3*y2 - z2^3*y1)*(z2^2*x1*(z1^2*x2 - z2^2*x1)^2 - x_out) -
866 z2^3*y1*(z1^2*x2 - z2^2*x1)^3 */
867 widefelem_diff(tmp2, tmp);
868 /* tmp2[i] < 2^118 + 2^120 < 2^121 */
869 felem_reduce(y_out, tmp2);
870
871 /* the result (x_out, y_out, z_out) is incorrect if one of the inputs is
872 * the point at infinity, so we need to check for this separately */
873
874 /* if point 1 is at infinity, copy point 2 to output, and vice versa */
875 copy_conditional(x_out, x2, z1_is_zero);
876 copy_conditional(x_out, x1, z2_is_zero);
877 copy_conditional(y_out, y2, z1_is_zero);
878 copy_conditional(y_out, y1, z2_is_zero);
879 copy_conditional(z_out, z2, z1_is_zero);
880 copy_conditional(z_out, z1, z2_is_zero);
881 felem_assign(x3, x_out);
882 felem_assign(y3, y_out);
883 felem_assign(z3, z_out);
884 }
885
886 /* select_point selects the |idx|th point from a precomputation table and
887 * copies it to out. */
888 static void select_point(const u64 idx, size_t size,
889 const felem pre_comp[/*size*/][3], felem out[3]) {
890 limb *outlimbs = &out[0][0];
891 memset(outlimbs, 0, 3 * sizeof(felem));
892
893 size_t i;
894 for (i = 0; i < size; i++) {
895 const limb *inlimbs = &pre_comp[i][0][0];
896 u64 mask = i ^ idx;
897 mask |= mask >> 4;
898 mask |= mask >> 2;
899 mask |= mask >> 1;
900 mask &= 1;
901 mask--;
902 size_t j;
903 for (j = 0; j < 4 * 3; j++) {
904 outlimbs[j] |= inlimbs[j] & mask;
905 }
906 }
907 }
908
909 /* get_bit returns the |i|th bit in |in| */
910 static char get_bit(const felem_bytearray in, size_t i) {
911 if (i >= 224) {
912 return 0;
913 }
914 return (in[i >> 3] >> (i & 7)) & 1;
915 }
916
917 /* Interleaved point multiplication using precomputed point multiples:
918 * The small point multiples 0*P, 1*P, ..., 16*P are in pre_comp[],
919 * the scalars in scalars[]. If g_scalar is non-NULL, we also add this multiple
920 * of the generator, using certain (large) precomputed multiples in g_pre_comp.
921 * Output point (X, Y, Z) is stored in x_out, y_out, z_out */
922 static void batch_mul(felem x_out, felem y_out, felem z_out,
923 const felem_bytearray scalars[],
924 const size_t num_points, const u8 *g_scalar,
925 const felem pre_comp[][17][3]) {
926 felem nq[3], tmp[4];
927 u64 bits;
928 u8 sign, digit;
929
930 /* set nq to the point at infinity */
931 memset(nq, 0, 3 * sizeof(felem));
932
933 /* Loop over all scalars msb-to-lsb, interleaving additions
934 * of multiples of the generator (two in each of the last 28 rounds)
935 * and additions of other points multiples (every 5th round). */
936 int skip = 1; /* save two point operations in the first round */
937 size_t i = num_points != 0 ? 220 : 27;
938 for (;;) {
939 /* double */
940 if (!skip) {
941 point_double(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2]);
942 }
943
944 /* add multiples of the generator */
945 if (g_scalar != NULL && i <= 27) {
946 /* first, look 28 bits upwards */
947 bits = get_bit(g_scalar, i + 196) << 3;
948 bits |= get_bit(g_scalar, i + 140) << 2;
949 bits |= get_bit(g_scalar, i + 84) << 1;
950 bits |= get_bit(g_scalar, i + 28);
951 /* select the point to add, in constant time */
952 select_point(bits, 16, g_pre_comp[1], tmp);
953
954 if (!skip) {
955 point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], 1 /* mixed */,
956 tmp[0], tmp[1], tmp[2]);
957 } else {
958 memcpy(nq, tmp, 3 * sizeof(felem));
959 skip = 0;
960 }
961
962 /* second, look at the current position */
963 bits = get_bit(g_scalar, i + 168) << 3;
964 bits |= get_bit(g_scalar, i + 112) << 2;
965 bits |= get_bit(g_scalar, i + 56) << 1;
966 bits |= get_bit(g_scalar, i);
967 /* select the point to add, in constant time */
968 select_point(bits, 16, g_pre_comp[0], tmp);
969 point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], 1 /* mixed */, tmp[0],
970 tmp[1], tmp[2]);
971 }
972
973 /* do other additions every 5 doublings */
974 if (num_points != 0 && i % 5 == 0) {
975 /* loop over all scalars */
976 size_t num;
977 for (num = 0; num < num_points; ++num) {
978 bits = get_bit(scalars[num], i + 4) << 5;
979 bits |= get_bit(scalars[num], i + 3) << 4;
980 bits |= get_bit(scalars[num], i + 2) << 3;
981 bits |= get_bit(scalars[num], i + 1) << 2;
982 bits |= get_bit(scalars[num], i) << 1;
983 bits |= get_bit(scalars[num], i - 1);
984 ec_GFp_nistp_recode_scalar_bits(&sign, &digit, bits);
985
986 /* select the point to add or subtract */
987 select_point(digit, 17, pre_comp[num], tmp);
988 felem_neg(tmp[3], tmp[1]); /* (X, -Y, Z) is the negative point */
989 copy_conditional(tmp[1], tmp[3], sign);
990
991 if (!skip) {
992 point_add(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2], 0 /* mixed */,
993 tmp[0], tmp[1], tmp[2]);
994 } else {
995 memcpy(nq, tmp, 3 * sizeof(felem));
996 skip = 0;
997 }
998 }
999 }
1000
1001 if (i == 0) {
1002 break;
1003 }
1004 --i;
1005 }
1006 felem_assign(x_out, nq[0]);
1007 felem_assign(y_out, nq[1]);
1008 felem_assign(z_out, nq[2]);
1009 }
1010
1011 /* Takes the Jacobian coordinates (X, Y, Z) of a point and returns
1012 * (X', Y') = (X/Z^2, Y/Z^3) */
1013 static int ec_GFp_nistp224_point_get_affine_coordinates(const EC_GROUP *group,
1014 const EC_POINT *point,
1015 BIGNUM *x, BIGNUM *y,
1016 BN_CTX *ctx) {
1017 felem z1, z2, x_in, y_in, x_out, y_out;
1018 widefelem tmp;
1019
1020 if (EC_POINT_is_at_infinity(group, point)) {
1021 OPENSSL_PUT_ERROR(EC, EC_R_POINT_AT_INFINITY);
1022 return 0;
1023 }
1024
1025 if (!BN_to_felem(x_in, &point->X) ||
1026 !BN_to_felem(y_in, &point->Y) ||
1027 !BN_to_felem(z1, &point->Z)) {
1028 return 0;
1029 }
1030
1031 felem_inv(z2, z1);
1032 felem_square(tmp, z2);
1033 felem_reduce(z1, tmp);
1034 felem_mul(tmp, x_in, z1);
1035 felem_reduce(x_in, tmp);
1036 felem_contract(x_out, x_in);
1037 if (x != NULL && !felem_to_BN(x, x_out)) {
1038 OPENSSL_PUT_ERROR(EC, ERR_R_BN_LIB);
1039 return 0;
1040 }
1041
1042 felem_mul(tmp, z1, z2);
1043 felem_reduce(z1, tmp);
1044 felem_mul(tmp, y_in, z1);
1045 felem_reduce(y_in, tmp);
1046 felem_contract(y_out, y_in);
1047 if (y != NULL && !felem_to_BN(y, y_out)) {
1048 OPENSSL_PUT_ERROR(EC, ERR_R_BN_LIB);
1049 return 0;
1050 }
1051
1052 return 1;
1053 }
1054
1055 static int ec_GFp_nistp224_points_mul(const EC_GROUP *group,
1056 EC_POINT *r,
1057 const BIGNUM *g_scalar,
1058 const EC_POINT *p_,
1059 const BIGNUM *p_scalar_,
1060 BN_CTX *ctx) {
1061 /* TODO: This function used to take |points| and |scalars| as arrays of
1062 * |num| elements. The code below should be simplified to work in terms of
1063 * |p_| and |p_scalar_|. */
1064 size_t num = p_ != NULL ? 1 : 0;
1065 const EC_POINT **points = p_ != NULL ? &p_ : NULL;
1066 BIGNUM const *const *scalars = p_ != NULL ? &p_scalar_ : NULL;
1067
1068 int ret = 0;
1069 BN_CTX *new_ctx = NULL;
1070 BIGNUM *x, *y, *z, *tmp_scalar;
1071 felem_bytearray g_secret;
1072 felem_bytearray *secrets = NULL;
1073 felem(*pre_comp)[17][3] = NULL;
1074 felem_bytearray tmp;
1075 size_t num_points = num;
1076 felem x_in, y_in, z_in, x_out, y_out, z_out;
1077 const EC_POINT *p = NULL;
1078 const BIGNUM *p_scalar = NULL;
1079
1080 if (ctx == NULL) {
1081 ctx = BN_CTX_new();
1082 new_ctx = ctx;
1083 if (ctx == NULL) {
1084 return 0;
1085 }
1086 }
1087
1088 BN_CTX_start(ctx);
1089 if ((x = BN_CTX_get(ctx)) == NULL ||
1090 (y = BN_CTX_get(ctx)) == NULL ||
1091 (z = BN_CTX_get(ctx)) == NULL ||
1092 (tmp_scalar = BN_CTX_get(ctx)) == NULL) {
1093 goto err;
1094 }
1095
1096 if (num_points > 0) {
1097 secrets = OPENSSL_malloc(num_points * sizeof(felem_bytearray));
1098 pre_comp = OPENSSL_malloc(num_points * sizeof(felem[17][3]));
1099 if (secrets == NULL ||
1100 pre_comp == NULL) {
1101 OPENSSL_PUT_ERROR(EC, ERR_R_MALLOC_FAILURE);
1102 goto err;
1103 }
1104
1105 /* we treat NULL scalars as 0, and NULL points as points at infinity,
1106 * i.e., they contribute nothing to the linear combination */
1107 memset(secrets, 0, num_points * sizeof(felem_bytearray));
1108 memset(pre_comp, 0, num_points * 17 * 3 * sizeof(felem));
1109 size_t i;
1110 for (i = 0; i < num_points; ++i) {
1111 if (i == num) {
1112 /* the generator */
1113 p = EC_GROUP_get0_generator(group);
1114 p_scalar = g_scalar;
1115 } else {
1116 /* the i^th point */
1117 p = points[i];
1118 p_scalar = scalars[i];
1119 }
1120
1121 if (p_scalar != NULL && p != NULL) {
1122 size_t num_bytes;
1123 /* reduce g_scalar to 0 <= g_scalar < 2^224 */
1124 if (BN_num_bits(p_scalar) > 224 || BN_is_negative(p_scalar)) {
1125 /* this is an unusual input, and we don't guarantee
1126 * constant-timeness */
1127 if (!BN_nnmod(tmp_scalar, p_scalar, &group->order, ctx)) {
1128 OPENSSL_PUT_ERROR(EC, ERR_R_BN_LIB);
1129 goto err;
1130 }
1131 num_bytes = BN_bn2bin(tmp_scalar, tmp);
1132 } else {
1133 num_bytes = BN_bn2bin(p_scalar, tmp);
1134 }
1135
1136 flip_endian(secrets[i], tmp, num_bytes);
1137 /* precompute multiples */
1138 if (!BN_to_felem(x_out, &p->X) ||
1139 !BN_to_felem(y_out, &p->Y) ||
1140 !BN_to_felem(z_out, &p->Z)) {
1141 goto err;
1142 }
1143
1144 felem_assign(pre_comp[i][1][0], x_out);
1145 felem_assign(pre_comp[i][1][1], y_out);
1146 felem_assign(pre_comp[i][1][2], z_out);
1147
1148 size_t j;
1149 for (j = 2; j <= 16; ++j) {
1150 if (j & 1) {
1151 point_add(pre_comp[i][j][0], pre_comp[i][j][1], pre_comp[i][j][2],
1152 pre_comp[i][1][0], pre_comp[i][1][1], pre_comp[i][1][2],
1153 0, pre_comp[i][j - 1][0], pre_comp[i][j - 1][1],
1154 pre_comp[i][j - 1][2]);
1155 } else {
1156 point_double(pre_comp[i][j][0], pre_comp[i][j][1],
1157 pre_comp[i][j][2], pre_comp[i][j / 2][0],
1158 pre_comp[i][j / 2][1], pre_comp[i][j / 2][2]);
1159 }
1160 }
1161 }
1162 }
1163 }
1164
1165 if (g_scalar != NULL) {
1166 memset(g_secret, 0, sizeof(g_secret));
1167 size_t num_bytes;
1168 /* reduce g_scalar to 0 <= g_scalar < 2^224 */
1169 if (BN_num_bits(g_scalar) > 224 || BN_is_negative(g_scalar)) {
1170 /* this is an unusual input, and we don't guarantee constant-timeness */
1171 if (!BN_nnmod(tmp_scalar, g_scalar, &group->order, ctx)) {
1172 OPENSSL_PUT_ERROR(EC, ERR_R_BN_LIB);
1173 goto err;
1174 }
1175 num_bytes = BN_bn2bin(tmp_scalar, tmp);
1176 } else {
1177 num_bytes = BN_bn2bin(g_scalar, tmp);
1178 }
1179
1180 flip_endian(g_secret, tmp, num_bytes);
1181 }
1182 batch_mul(x_out, y_out, z_out, (const felem_bytearray(*))secrets,
1183 num_points, g_scalar != NULL ? g_secret : NULL,
1184 (const felem(*)[17][3])pre_comp);
1185
1186 /* reduce the output to its unique minimal representation */
1187 felem_contract(x_in, x_out);
1188 felem_contract(y_in, y_out);
1189 felem_contract(z_in, z_out);
1190 if (!felem_to_BN(x, x_in) ||
1191 !felem_to_BN(y, y_in) ||
1192 !felem_to_BN(z, z_in)) {
1193 OPENSSL_PUT_ERROR(EC, ERR_R_BN_LIB);
1194 goto err;
1195 }
1196 ret = ec_point_set_Jprojective_coordinates_GFp(group, r, x, y, z, ctx);
1197
1198 err:
1199 BN_CTX_end(ctx);
1200 BN_CTX_free(new_ctx);
1201 OPENSSL_free(secrets);
1202 OPENSSL_free(pre_comp);
1203 return ret;
1204 }
1205
1206 const EC_METHOD *EC_GFp_nistp224_method(void) {
1207 static const EC_METHOD ret = {ec_GFp_simple_group_init,
1208 ec_GFp_simple_group_finish,
1209 ec_GFp_simple_group_copy,
1210 ec_GFp_simple_group_set_curve,
1211 ec_GFp_nistp224_point_get_affine_coordinates,
1212 ec_GFp_nistp224_points_mul,
1213 0 /* check_pub_key_order */,
1214 ec_GFp_simple_field_mul,
1215 ec_GFp_simple_field_sqr,
1216 0 /* field_encode */,
1217 0 /* field_decode */};
1218
1219 return &ret;
1220 }
1221
1222 #endif /* 64_BIT && !WINDOWS && !SMALL */