]> git.ipfire.org Git - thirdparty/openssl.git/blame - crypto/ec/ecp_nistz256.c
projects / thirdparty / openssl.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Fixed error introduced in commit f2be92b94dad3c6cbdf79d99a324804094cf1617
[thirdparty/openssl.git] / crypto / ec / ecp_nistz256.c
CommitLineData
4d3fa06f
AP		 1/******************************************************************************
2 * *
3 * Copyright 2014 Intel Corporation *
4 * *
5 * Licensed under the Apache License, Version 2.0 (the "License"); *
6 * you may not use this file except in compliance with the License. *
7 * You may obtain a copy of the License at *
8 * *
9 * http://www.apache.org/licenses/LICENSE-2.0 *
10 * *
11 * Unless required by applicable law or agreed to in writing, software *
12 * distributed under the License is distributed on an "AS IS" BASIS, *
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. *
14 * See the License for the specific language governing permissions and *
15 * limitations under the License. *
16 * *
17 ******************************************************************************
18 * *
19 * Developers and authors: *
20 * Shay Gueron (1, 2), and Vlad Krasnov (1) *
21 * (1) Intel Corporation, Israel Development Center *
22 * (2) University of Haifa *
23 * Reference: *
24 * S.Gueron and V.Krasnov, "Fast Prime Field Elliptic Curve Cryptography with *
25 * 256 Bit Primes" *
26 * *
27 ******************************************************************************/
28
29#include <string.h>
30
31#include <openssl/bn.h>
32#include <openssl/err.h>
33#include <openssl/ec.h>
34#include "cryptlib.h"
35
36#include "ec_lcl.h"
37
38#if BN_BITS2 != 64
39# define TOBN(hi,lo) lo,hi
40#else
41# define TOBN(hi,lo) ((BN_ULONG)hi<<32|lo)
42#endif
43
44#if defined(__GNUC__)
45# define ALIGN32 __attribute((aligned(32)))
46#elif defined(_MSC_VER)
47# define ALIGN32 __declspec(align(32))
48#else
49# define ALIGN32
50#endif
51
52#define ALIGNPTR(p,N) ((unsigned char *)p+N-(size_t)p%N)
53#define P256_LIMBS (256/BN_BITS2)
54
55typedef unsigned short u16;
56
57typedef struct {
58 BN_ULONG X[P256_LIMBS];
59 BN_ULONG Y[P256_LIMBS];
60 BN_ULONG Z[P256_LIMBS];
61} P256_POINT;
62
63typedef struct {
64 BN_ULONG X[P256_LIMBS];
65 BN_ULONG Y[P256_LIMBS];
66} P256_POINT_AFFINE;
67
68typedef P256_POINT_AFFINE PRECOMP256_ROW[64];
69
70/* structure for precomputed multiples of the generator */
71typedef struct ec_pre_comp_st {
72 const EC_GROUP *group; /* Parent EC_GROUP object */
73 size_t w; /* Window size */
74 /* Constant time access to the X and Y coordinates of the pre-computed,
75 * generator multiplies, in the Montgomery domain. Pre-calculated
76 * multiplies are stored in affine form. */
77 PRECOMP256_ROW *precomp;
78 void *precomp_storage;
79 int references;
80} EC_PRE_COMP;
81
82/* Functions implemented in assembly */
83/* Modular mul by 2: res = 2*a mod P */
84void ecp_nistz256_mul_by_2(BN_ULONG res[P256_LIMBS],
85 const BN_ULONG a[P256_LIMBS]);
86/* Modular div by 2: res = a/2 mod P */
87void ecp_nistz256_div_by_2(BN_ULONG res[P256_LIMBS],
88 const BN_ULONG a[P256_LIMBS]);
89/* Modular mul by 3: res = 3*a mod P */
90void ecp_nistz256_mul_by_3(BN_ULONG res[P256_LIMBS],
91 const BN_ULONG a[P256_LIMBS]);
92/* Modular add: res = a+b mod P */
93void ecp_nistz256_add(BN_ULONG res[P256_LIMBS],
94 const BN_ULONG a[P256_LIMBS],
95 const BN_ULONG b[P256_LIMBS]);
96/* Modular sub: res = a-b mod P */
97void ecp_nistz256_sub(BN_ULONG res[P256_LIMBS],
98 const BN_ULONG a[P256_LIMBS],
99 const BN_ULONG b[P256_LIMBS]);
100/* Modular neg: res = -a mod P */
101void ecp_nistz256_neg(BN_ULONG res[P256_LIMBS], const BN_ULONG a[P256_LIMBS]);
102/* Montgomery mul: res = a*b*2^-256 mod P */
103void ecp_nistz256_mul_mont(BN_ULONG res[P256_LIMBS],
104 const BN_ULONG a[P256_LIMBS],
105 const BN_ULONG b[P256_LIMBS]);
106/* Montgomery sqr: res = a*a*2^-256 mod P */
107void ecp_nistz256_sqr_mont(BN_ULONG res[P256_LIMBS],
108 const BN_ULONG a[P256_LIMBS]);
109/* Convert a number from Montgomery domain, by multiplying with 1 */
110void ecp_nistz256_from_mont(BN_ULONG res[P256_LIMBS],
111 const BN_ULONG in[P256_LIMBS]);
112/* Convert a number to Montgomery domain, by multiplying with 2^512 mod P*/
113void ecp_nistz256_to_mont(BN_ULONG res[P256_LIMBS],
114 const BN_ULONG in[P256_LIMBS]);
115/* Functions that perform constant time access to the precomputed tables */
116void ecp_nistz256_select_w5(P256_POINT * val,
117 const P256_POINT * in_t, int index);
118void ecp_nistz256_select_w7(P256_POINT_AFFINE * val,
119 const P256_POINT_AFFINE * in_t, int index);
120
121/* One converted into the Montgomery domain */
122static const BN_ULONG ONE[P256_LIMBS] = {
123 TOBN(0x00000000, 0x00000001), TOBN(0xffffffff, 0x00000000),
124 TOBN(0xffffffff, 0xffffffff), TOBN(0x00000000, 0xfffffffe)
125};
126
127static void *ec_pre_comp_dup(void *);
128static void ec_pre_comp_free(void *);
129static void ec_pre_comp_clear_free(void *);
130static EC_PRE_COMP *ec_pre_comp_new(const EC_GROUP * group);
131
132/* Precomputed tables for the default generator */
133#include "ecp_nistz256_table.c"
134
135/* Recode window to a signed digit, see ecp_nistputil.c for details */
136static unsigned int _booth_recode_w5(unsigned int in)
137{
138 unsigned int s, d;
139
140 s = ~((in >> 5) - 1);
141 d = (1 << 6) - in - 1;
142 d = (d & s) | (in & ~s);
143 d = (d >> 1) + (d & 1);
144
145 return (d << 1) + (s & 1);
146}
147
148static unsigned int _booth_recode_w7(unsigned int in)
149{
150 unsigned int s, d;
151
152 s = ~((in >> 7) - 1);
153 d = (1 << 8) - in - 1;
154 d = (d & s) | (in & ~s);
155 d = (d >> 1) + (d & 1);
156
157 return (d << 1) + (s & 1);
158}
159
160static void copy_conditional(BN_ULONG dst[P256_LIMBS],
161 const BN_ULONG src[P256_LIMBS], BN_ULONG move)
162{
163 BN_ULONG mask1 = -move;
164 BN_ULONG mask2 = ~mask1;
165
166 dst[0] = (src[0] & mask1) ^ (dst[0] & mask2);
167 dst[1] = (src[1] & mask1) ^ (dst[1] & mask2);
168 dst[2] = (src[2] & mask1) ^ (dst[2] & mask2);
169 dst[3] = (src[3] & mask1) ^ (dst[3] & mask2);
170 if (P256_LIMBS == 8) {
171 dst[4] = (src[4] & mask1) ^ (dst[4] & mask2);
172 dst[5] = (src[5] & mask1) ^ (dst[5] & mask2);
173 dst[6] = (src[6] & mask1) ^ (dst[6] & mask2);
174 dst[7] = (src[7] & mask1) ^ (dst[7] & mask2);
175 }
176}
177
178static BN_ULONG is_zero(BN_ULONG in)
179{
180 in |= (0 - in);
181 in = ~in;
182 in &= BN_MASK2;
183 in >>= BN_BITS2 - 1;
184 return in;
185}
186
187static BN_ULONG is_equal(const BN_ULONG a[P256_LIMBS],
188 const BN_ULONG b[P256_LIMBS])
189{
190 BN_ULONG res;
191
192 res = a[0] ^ b[0];
193 res |= a[1] ^ b[1];
194 res |= a[2] ^ b[2];
195 res |= a[3] ^ b[3];
196 if (P256_LIMBS == 8) {
197 res |= a[4] ^ b[4];
198 res |= a[5] ^ b[5];
199 res |= a[6] ^ b[6];
200 res |= a[7] ^ b[7];
201 }
202
203 return is_zero(res);
204}
205
206static BN_ULONG is_one(const BN_ULONG a[P256_LIMBS])
207{
208 BN_ULONG res;
209
210 res = a[0] ^ ONE[0];
211 res |= a[1] ^ ONE[1];
212 res |= a[2] ^ ONE[2];
213 res |= a[3] ^ ONE[3];
214 if (P256_LIMBS == 8) {
215 res |= a[4] ^ ONE[4];
216 res |= a[5] ^ ONE[5];
217 res |= a[6] ^ ONE[6];
218 }
219
220 return is_zero(res);
221}
222
223#ifndef ECP_NISTZ256_REFERENCE_IMPLEMENTATION
224void ecp_nistz256_point_double(P256_POINT * r, const P256_POINT * a);
225void ecp_nistz256_point_add(P256_POINT * r,
226 const P256_POINT * a, const P256_POINT * b);
227void ecp_nistz256_point_add_affine(P256_POINT * r,
228 const P256_POINT * a,
229 const P256_POINT_AFFINE * b);
230#else
231/* Point double: r = 2*a */
232static void ecp_nistz256_point_double(P256_POINT * r, const P256_POINT * a)
233{
234 BN_ULONG S[P256_LIMBS];
235 BN_ULONG M[P256_LIMBS];
236 BN_ULONG Zsqr[P256_LIMBS];
237 BN_ULONG tmp0[P256_LIMBS];
238
239 const BN_ULONG *in_x = a->X;
240 const BN_ULONG *in_y = a->Y;
241 const BN_ULONG *in_z = a->Z;
242
243 BN_ULONG *res_x = r->X;
244 BN_ULONG *res_y = r->Y;
245 BN_ULONG *res_z = r->Z;
246
247 ecp_nistz256_mul_by_2(S, in_y);
248
249 ecp_nistz256_sqr_mont(Zsqr, in_z);
250
251 ecp_nistz256_sqr_mont(S, S);
252
253 ecp_nistz256_mul_mont(res_z, in_z, in_y);
254 ecp_nistz256_mul_by_2(res_z, res_z);
255
256 ecp_nistz256_add(M, in_x, Zsqr);
257 ecp_nistz256_sub(Zsqr, in_x, Zsqr);
258
259 ecp_nistz256_sqr_mont(res_y, S);
260 ecp_nistz256_div_by_2(res_y, res_y);
261
262 ecp_nistz256_mul_mont(M, M, Zsqr);
263 ecp_nistz256_mul_by_3(M, M);
264
265 ecp_nistz256_mul_mont(S, S, in_x);
266 ecp_nistz256_mul_by_2(tmp0, S);
267
268 ecp_nistz256_sqr_mont(res_x, M);
269
270 ecp_nistz256_sub(res_x, res_x, tmp0);
271 ecp_nistz256_sub(S, S, res_x);
272
273 ecp_nistz256_mul_mont(S, S, M);
274 ecp_nistz256_sub(res_y, S, res_y);
275}
276
277/* Point addition: r = a+b */
278static void ecp_nistz256_point_add(P256_POINT * r,
279 const P256_POINT * a, const P256_POINT * b)
280{
281 BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS];
282 BN_ULONG U1[P256_LIMBS], S1[P256_LIMBS];
283 BN_ULONG Z1sqr[P256_LIMBS];
284 BN_ULONG Z2sqr[P256_LIMBS];
285 BN_ULONG H[P256_LIMBS], R[P256_LIMBS];
286 BN_ULONG Hsqr[P256_LIMBS];
287 BN_ULONG Rsqr[P256_LIMBS];
288 BN_ULONG Hcub[P256_LIMBS];
289
290 BN_ULONG res_x[P256_LIMBS];
291 BN_ULONG res_y[P256_LIMBS];
292 BN_ULONG res_z[P256_LIMBS];
293
294 BN_ULONG in1infty, in2infty;
295
296 const BN_ULONG *in1_x = a->X;
297 const BN_ULONG *in1_y = a->Y;
298 const BN_ULONG *in1_z = a->Z;
299
300 const BN_ULONG *in2_x = b->X;
301 const BN_ULONG *in2_y = b->Y;
302 const BN_ULONG *in2_z = b->Z;
303
304 /* We encode infinity as (0,0), which is not on the curve,
305 * so it is OK. */
306 in1infty = in1_x[0] | in1_x[1] | in1_x[2] | in1_x[3] |
307 in1_y[0] | in1_y[1] | in1_y[2] | in1_y[3];
308 if (P256_LIMBS == 8)
309 in1infty |= in1_x[4] | in1_x[5] | in1_x[6] | in1_x[7] |
310 in1_y[4] | in1_y[5] | in1_y[6] | in1_y[7];
311
312 in2infty = in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] |
313 in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3];
314 if (P256_LIMBS == 8)
315 in2infty |= in2_x[4] | in2_x[5] | in2_x[6] | in2_x[7] |
316 in2_y[4] | in2_y[5] | in2_y[6] | in2_y[7];
317
318 in1infty = is_zero(in1infty);
319 in2infty = is_zero(in2infty);
320
321 ecp_nistz256_sqr_mont(Z2sqr, in2_z); /* Z2^2 */
322 ecp_nistz256_sqr_mont(Z1sqr, in1_z); /* Z1^2 */
323
324 ecp_nistz256_mul_mont(S1, Z2sqr, in2_z); /* S1 = Z2^3 */
325 ecp_nistz256_mul_mont(S2, Z1sqr, in1_z); /* S2 = Z1^3 */
326
327 ecp_nistz256_mul_mont(S1, S1, in1_y); /* S1 = Y1*Z2^3 */
328 ecp_nistz256_mul_mont(S2, S2, in2_y); /* S2 = Y2*Z1^3 */
329 ecp_nistz256_sub(R, S2, S1); /* R = S2 - S1 */
330
331 ecp_nistz256_mul_mont(U1, in1_x, Z2sqr); /* U1 = X1*Z2^2 */
332 ecp_nistz256_mul_mont(U2, in2_x, Z1sqr); /* U2 = X2*Z1^2 */
333 ecp_nistz256_sub(H, U2, U1); /* H = U2 - U1 */
334
335 /* This should not happen during sign/ecdh,
336 * so no constant time violation */
337 if (is_equal(U1, U2) && !in1infty && !in2infty) {
338 if (is_equal(S1, S2)) {
339 ecp_nistz256_point_double(r, a);
340 return;
341 } else {
342 memset(r, 0, sizeof(*r));
343 return;
344 }
345 }
346
347 ecp_nistz256_sqr_mont(Rsqr, R); /* R^2 */
348 ecp_nistz256_mul_mont(res_z, H, in1_z); /* Z3 = H*Z1*Z2 */
349 ecp_nistz256_sqr_mont(Hsqr, H); /* H^2 */
350 ecp_nistz256_mul_mont(res_z, res_z, in2_z); /* Z3 = H*Z1*Z2 */
351 ecp_nistz256_mul_mont(Hcub, Hsqr, H); /* H^3 */
352
353 ecp_nistz256_mul_mont(U2, U1, Hsqr); /* U1*H^2 */
354 ecp_nistz256_mul_by_2(Hsqr, U2); /* 2*U1*H^2 */
355
356 ecp_nistz256_sub(res_x, Rsqr, Hsqr);
357 ecp_nistz256_sub(res_x, res_x, Hcub);
358
359 ecp_nistz256_sub(res_y, U2, res_x);
360
361 ecp_nistz256_mul_mont(S2, S1, Hcub);
362 ecp_nistz256_mul_mont(res_y, R, res_y);
363 ecp_nistz256_sub(res_y, res_y, S2);
364
365 copy_conditional(res_x, in2_x, in1infty);
366 copy_conditional(res_y, in2_y, in1infty);
367 copy_conditional(res_z, in2_z, in1infty);
368
369 copy_conditional(res_x, in1_x, in2infty);
370 copy_conditional(res_y, in1_y, in2infty);
371 copy_conditional(res_z, in1_z, in2infty);
372
373 memcpy(r->X, res_x, sizeof(res_x));
374 memcpy(r->Y, res_y, sizeof(res_y));
375 memcpy(r->Z, res_z, sizeof(res_z));
376}
377
378/* Point addition when b is known to be affine: r = a+b */
379static void ecp_nistz256_point_add_affine(P256_POINT * r,
380 const P256_POINT * a,
381 const P256_POINT_AFFINE * b)
382{
383 BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS];
384 BN_ULONG Z1sqr[P256_LIMBS];
385 BN_ULONG H[P256_LIMBS], R[P256_LIMBS];
386 BN_ULONG Hsqr[P256_LIMBS];
387 BN_ULONG Rsqr[P256_LIMBS];
388 BN_ULONG Hcub[P256_LIMBS];
389
390 BN_ULONG res_x[P256_LIMBS];
391 BN_ULONG res_y[P256_LIMBS];
392 BN_ULONG res_z[P256_LIMBS];
393
394 BN_ULONG in1infty, in2infty;
395
396 const BN_ULONG *in1_x = a->X;
397 const BN_ULONG *in1_y = a->Y;
398 const BN_ULONG *in1_z = a->Z;
399
400 const BN_ULONG *in2_x = b->X;
401 const BN_ULONG *in2_y = b->Y;
402
403 /* In affine representation we encode infty as (0,0),
404 * which is not on the curve, so it is OK */
405 in1infty = in1_x[0] | in1_x[1] | in1_x[2] | in1_x[3] |
406 in1_y[0] | in1_y[1] | in1_y[2] | in1_y[3];
407 if (P256_LIMBS == 8)
408 in1infty |= in1_x[4] | in1_x[5] | in1_x[6] | in1_x[7] |
409 in1_y[4] | in1_y[5] | in1_y[6] | in1_y[7];
410
411 in2infty = in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] |
412 in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3];
413 if (P256_LIMBS == 8)
414 in2infty |= in2_x[4] | in2_x[5] | in2_x[6] | in2_x[7] |
415 in2_y[4] | in2_y[5] | in2_y[6] | in2_y[7];
416
417 in1infty = is_zero(in1infty);
418 in2infty = is_zero(in2infty);
419
420 ecp_nistz256_sqr_mont(Z1sqr, in1_z); /* Z1^2 */
421
422 ecp_nistz256_mul_mont(U2, in2_x, Z1sqr); /* U2 = X2*Z1^2 */
423 ecp_nistz256_sub(H, U2, in1_x); /* H = U2 - U1 */
424
425 ecp_nistz256_mul_mont(S2, Z1sqr, in1_z); /* S2 = Z1^3 */
426
427 ecp_nistz256_mul_mont(res_z, H, in1_z); /* Z3 = H*Z1*Z2 */
428
429 ecp_nistz256_mul_mont(S2, S2, in2_y); /* S2 = Y2*Z1^3 */
430 ecp_nistz256_sub(R, S2, in1_y); /* R = S2 - S1 */
431
432 ecp_nistz256_sqr_mont(Hsqr, H); /* H^2 */
433 ecp_nistz256_sqr_mont(Rsqr, R); /* R^2 */
434 ecp_nistz256_mul_mont(Hcub, Hsqr, H); /* H^3 */
435
436 ecp_nistz256_mul_mont(U2, in1_x, Hsqr); /* U1*H^2 */
437 ecp_nistz256_mul_by_2(Hsqr, U2); /* 2*U1*H^2 */
438
439 ecp_nistz256_sub(res_x, Rsqr, Hsqr);
440 ecp_nistz256_sub(res_x, res_x, Hcub);
441 ecp_nistz256_sub(H, U2, res_x);
442
443 ecp_nistz256_mul_mont(S2, in1_y, Hcub);
444 ecp_nistz256_mul_mont(H, H, R);
445 ecp_nistz256_sub(res_y, H, S2);
446
447 copy_conditional(res_x, in2_x, in1infty);
448 copy_conditional(res_x, in1_x, in2infty);
449
450 copy_conditional(res_y, in2_y, in1infty);
451 copy_conditional(res_y, in1_y, in2infty);
452
453 copy_conditional(res_z, ONE, in1infty);
454 copy_conditional(res_z, in1_z, in2infty);
455
456 memcpy(r->X, res_x, sizeof(res_x));
457 memcpy(r->Y, res_y, sizeof(res_y));
458 memcpy(r->Z, res_z, sizeof(res_z));
459}
460#endif
461
462/* r = in^-1 mod p */
463static void ecp_nistz256_mod_inverse(BN_ULONG r[P256_LIMBS],
464 const BN_ULONG in[P256_LIMBS])
465{
466 /* The poly is ffffffff 00000001 00000000 00000000 00000000 ffffffff ffffffff ffffffff
467 We use FLT and used poly-2 as exponent */
468 BN_ULONG p2[P256_LIMBS];
469 BN_ULONG p4[P256_LIMBS];
470 BN_ULONG p8[P256_LIMBS];
471 BN_ULONG p16[P256_LIMBS];
472 BN_ULONG p32[P256_LIMBS];
473 BN_ULONG res[P256_LIMBS];
474 int i;
475
476 ecp_nistz256_sqr_mont(res, in);
477 ecp_nistz256_mul_mont(p2, res, in); /* 3*p */
478
479 ecp_nistz256_sqr_mont(res, p2);
480 ecp_nistz256_sqr_mont(res, res);
481 ecp_nistz256_mul_mont(p4, res, p2); /* f*p */
482
483 ecp_nistz256_sqr_mont(res, p4);
484 ecp_nistz256_sqr_mont(res, res);
485 ecp_nistz256_sqr_mont(res, res);
486 ecp_nistz256_sqr_mont(res, res);
487 ecp_nistz256_mul_mont(p8, res, p4); /* ff*p */
488
489 ecp_nistz256_sqr_mont(res, p8);
490 for (i = 0; i < 7; i++)
491 ecp_nistz256_sqr_mont(res, res);
492 ecp_nistz256_mul_mont(p16, res, p8); /* ffff*p */
493
494 ecp_nistz256_sqr_mont(res, p16);
495 for (i = 0; i < 15; i++)
496 ecp_nistz256_sqr_mont(res, res);
497 ecp_nistz256_mul_mont(p32, res, p16); /* ffffffff*p */
498
499 ecp_nistz256_sqr_mont(res, p32);
500 for (i = 0; i < 31; i++)
501 ecp_nistz256_sqr_mont(res, res);
502 ecp_nistz256_mul_mont(res, res, in);
503
504 for (i = 0; i < 32 * 4; i++)
505 ecp_nistz256_sqr_mont(res, res);
506 ecp_nistz256_mul_mont(res, res, p32);
507
508 for (i = 0; i < 32; i++)
509 ecp_nistz256_sqr_mont(res, res);
510 ecp_nistz256_mul_mont(res, res, p32);
511
512 for (i = 0; i < 16; i++)
513 ecp_nistz256_sqr_mont(res, res);
514 ecp_nistz256_mul_mont(res, res, p16);
515
516 for (i = 0; i < 8; i++)
517 ecp_nistz256_sqr_mont(res, res);
518 ecp_nistz256_mul_mont(res, res, p8);
519
520 ecp_nistz256_sqr_mont(res, res);
521 ecp_nistz256_sqr_mont(res, res);
522 ecp_nistz256_sqr_mont(res, res);
523 ecp_nistz256_sqr_mont(res, res);
524 ecp_nistz256_mul_mont(res, res, p4);
525
526 ecp_nistz256_sqr_mont(res, res);
527 ecp_nistz256_sqr_mont(res, res);
528 ecp_nistz256_mul_mont(res, res, p2);
529
530 ecp_nistz256_sqr_mont(res, res);
531 ecp_nistz256_sqr_mont(res, res);
532 ecp_nistz256_mul_mont(res, res, in);
533
534 memcpy(r, res, sizeof(res));
535}
536
537/* ecp_nistz256_bignum_to_field_elem copies the contents of |in| to |out| and
538 * returns one if it fits. Otherwise it returns zero. */
539static int ecp_nistz256_bignum_to_field_elem(BN_ULONG out[P256_LIMBS],
540 const BIGNUM * in)
541{
542 if (in->top > P256_LIMBS)
543 return 0;
544
545 memset(out, 0, sizeof(BN_ULONG) * P256_LIMBS);
546 memcpy(out, in->d, sizeof(BN_ULONG) * in->top);
547 return 1;
548}
549
550/* r = sum(scalar[i]*point[i]) */
551static void ecp_nistz256_windowed_mul(const EC_GROUP * group,
552 P256_POINT * r,
553 const BIGNUM ** scalar,
554 const EC_POINT ** point,
555 int num, BN_CTX * ctx)
556{
557 int i, j;
558 unsigned int index;
559 unsigned char (*p_str)[33] = NULL;
560 const unsigned int window_size = 5;
561 const unsigned int mask = (1 << (window_size + 1)) - 1;
562 unsigned int wvalue;
563 BN_ULONG tmp[P256_LIMBS];
564 ALIGN32 P256_POINT h;
565 const BIGNUM **scalars = NULL;
566 P256_POINT(*table)[16] = NULL;
567 void *table_storage = NULL;
568
569 if ((table_storage =
570 OPENSSL_malloc(num * 16 * sizeof(P256_POINT) + 64)) == NULL
571 || (p_str =
572 OPENSSL_malloc(num * 33 * sizeof(unsigned char))) == NULL
573 || (scalars = OPENSSL_malloc(num * sizeof(BIGNUM *))) == NULL) {
574 ECerr(EC_F_NISTZ256_POINTS_MUL_W, ERR_R_MALLOC_FAILURE);
575 goto err;
576 } else {
577 table = (void *)ALIGNPTR(table_storage, 64);
578 }
579
580 for (i = 0; i < num; i++) {
581 P256_POINT *row = table[i];
582
583 if ((BN_num_bits(scalar[i]) > 256) || BN_is_negative(scalar[i])) {
584 BIGNUM *mod;
585
586 if ((mod = BN_CTX_get(ctx)) == NULL)
587 goto err;
588 if (!BN_nnmod(mod, scalar[i], &group->order, ctx)) {
589 ECerr(EC_F_NISTZ256_POINTS_MUL_W, ERR_R_BN_LIB);
590 goto err;
591 }
592 scalars[i] = mod;
593 } else
594 scalars[i] = scalar[i];
595
596 for (j = 0; j < scalars[i]->top * BN_BYTES; j += BN_BYTES) {
597 BN_ULONG d = scalars[i]->d[j / BN_BYTES];
598
599 p_str[i][j + 0] = d & 0xff;
600 p_str[i][j + 1] = (d >> 8) & 0xff;
601 p_str[i][j + 2] = (d >> 16) & 0xff;
602 p_str[i][j + 3] = (d >>= 24) & 0xff;
603 if (BN_BYTES == 8) {
604 d >>= 8;
605 p_str[i][j + 4] = d & 0xff;
606 p_str[i][j + 5] = (d >> 8) & 0xff;
607 p_str[i][j + 6] = (d >> 16) & 0xff;
608 p_str[i][j + 7] = (d >> 24) & 0xff;
609 }
610 }
611 for (; j < 33; j++)
612 p_str[i][j] = 0;
613
614 /* table[0] is implicitly (0,0,0) (the point at infinity),
615 * therefore it is not stored. All other values are actually
616 * stored with an offset of -1 in table.
617 */
618
619 if (!ecp_nistz256_bignum_to_field_elem(row[1 - 1].X, &point[i]->X)
620 || !ecp_nistz256_bignum_to_field_elem(row[1 - 1].Y, &point[i]->Y)
621 || !ecp_nistz256_bignum_to_field_elem(row[1 - 1].Z, &point[i]->Z)) {
622 ECerr(EC_F_NISTZ256_POINTS_MUL_W, EC_R_COORDINATES_OUT_OF_RANGE);
623 goto err;
624 }
625
626 ecp_nistz256_point_double(&row[ 2 - 1], &row[ 1 - 1]);
627 ecp_nistz256_point_add (&row[ 3 - 1], &row[ 2 - 1], &row[1 - 1]);
628 ecp_nistz256_point_double(&row[ 4 - 1], &row[ 2 - 1]);
629 ecp_nistz256_point_double(&row[ 6 - 1], &row[ 3 - 1]);
630 ecp_nistz256_point_double(&row[ 8 - 1], &row[ 4 - 1]);
631 ecp_nistz256_point_double(&row[12 - 1], &row[ 6 - 1]);
632 ecp_nistz256_point_add (&row[ 5 - 1], &row[ 4 - 1], &row[1 - 1]);
633 ecp_nistz256_point_add (&row[ 7 - 1], &row[ 6 - 1], &row[1 - 1]);
634 ecp_nistz256_point_add (&row[ 9 - 1], &row[ 8 - 1], &row[1 - 1]);
635 ecp_nistz256_point_add (&row[13 - 1], &row[12 - 1], &row[1 - 1]);
636 ecp_nistz256_point_double(&row[14 - 1], &row[ 7 - 1]);
637 ecp_nistz256_point_double(&row[10 - 1], &row[ 5 - 1]);
638 ecp_nistz256_point_add (&row[15 - 1], &row[14 - 1], &row[1 - 1]);
639 ecp_nistz256_point_add (&row[11 - 1], &row[10 - 1], &row[1 - 1]);
640 ecp_nistz256_point_add (&row[16 - 1], &row[15 - 1], &row[1 - 1]);
641 }
642
643 index = 255;
644
645 wvalue = p_str[0][(index - 1) / 8];
646 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
647
648 ecp_nistz256_select_w5(r, table[0], _booth_recode_w5(wvalue) >> 1);
649
650 while (index >= 5) {
651 for (i = (index == 255 ? 1 : 0); i < num; i++) {
652 unsigned int off = (index - 1) / 8;
653
654 wvalue = p_str[i][off] | p_str[i][off + 1] << 8;
655 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
656
657 wvalue = _booth_recode_w5(wvalue);
658
659 ecp_nistz256_select_w5(&h, table[i], wvalue >> 1);
660
661 ecp_nistz256_neg(tmp, h.Y);
662 copy_conditional(h.Y, tmp, (wvalue & 1));
663
664 ecp_nistz256_point_add(r, r, &h);
665 }
666
667 index -= window_size;
668
669 ecp_nistz256_point_double(r, r);
670 ecp_nistz256_point_double(r, r);
671 ecp_nistz256_point_double(r, r);
672 ecp_nistz256_point_double(r, r);
673 ecp_nistz256_point_double(r, r);
674 }
675
676 /* Final window */
677 for (i = 0; i < num; i++) {
678 wvalue = p_str[i][0];
679 wvalue = (wvalue << 1) & mask;
680
681 wvalue = _booth_recode_w5(wvalue);
682
683 ecp_nistz256_select_w5(&h, table[i], wvalue >> 1);
684
685 ecp_nistz256_neg(tmp, h.Y);
686 copy_conditional(h.Y, tmp, wvalue & 1);
687
688 ecp_nistz256_point_add(r, r, &h);
689 }
690
691err:
692 if (table_storage)
693 OPENSSL_free(table_storage);
694 if (p_str)
695 OPENSSL_free(p_str);
696 if (scalars)
697 OPENSSL_free(scalars);
698}
699
700/* Coordinates of G, for which we have precomputed tables */
701const static BN_ULONG def_xG[P256_LIMBS] = {
702 TOBN(0x79e730d4, 0x18a9143c), TOBN(0x75ba95fc, 0x5fedb601),
703 TOBN(0x79fb732b, 0x77622510), TOBN(0x18905f76, 0xa53755c6)
704};
705
706const static BN_ULONG def_yG[P256_LIMBS] = {
707 TOBN(0xddf25357, 0xce95560a), TOBN(0x8b4ab8e4, 0xba19e45c),
708 TOBN(0xd2e88688, 0xdd21f325), TOBN(0x8571ff18, 0x25885d85)
709};
710
711/* ecp_nistz256_is_affine_G returns one if |generator| is the standard,
712 * P-256 generator. */
713static int ecp_nistz256_is_affine_G(const EC_POINT * generator)
714{
715 return (generator->X.top == P256_LIMBS) &&
716 (generator->Y.top == P256_LIMBS) &&
717 (generator->Z.top == (P256_LIMBS - P256_LIMBS / 8)) &&
718 is_equal(generator->X.d, def_xG) &&
719 is_equal(generator->Y.d, def_yG) && is_one(generator->Z.d);
720}
721
722static int ecp_nistz256_mult_precompute(EC_GROUP * group, BN_CTX * ctx)
723{
724 /* We precompute a table for a Booth encoded exponent (wNAF) based
725 * computation. Each table holds 64 values for safe access, with an
726 * implicit value of infinity at index zero. We use window of size 7,
727 * and therefore require ceil(256/7) = 37 tables. */
728 BIGNUM *order;
729 EC_POINT *P = NULL, *T = NULL;
730 const EC_POINT *generator;
731 EC_PRE_COMP *pre_comp;
732 int i, j, k, ret = 0;
733 size_t w;
734
735 PRECOMP256_ROW *preComputedTable = NULL;
736 unsigned char *precomp_storage = NULL;
737
738 /* if there is an old EC_PRE_COMP object, throw it away */
739 EC_EX_DATA_free_data(&group->extra_data, ec_pre_comp_dup,
740 ec_pre_comp_free, ec_pre_comp_clear_free);
741
742 generator = EC_GROUP_get0_generator(group);
743 if (generator == NULL) {
744 ECerr(EC_F_NISTZ256_PRECOMPUTE_MULT, EC_R_UNDEFINED_GENERATOR);
745 return 0;
746 }
747
748 if (ecp_nistz256_is_affine_G(generator)) {
749 /* No need to calculate tables for the standard generator
750 * because we have them statically. */
751 return 1;
752 }
753
754 if ((pre_comp = ec_pre_comp_new(group)) == NULL)
755 return 0;
756
757 if (ctx == NULL) {
758 ctx = BN_CTX_new();
759 if (ctx == NULL)
760 goto err;
761 }
762
763 BN_CTX_start(ctx);
764 order = BN_CTX_get(ctx);
765
766 if (order == NULL)
767 goto err;
768
769 if (!EC_GROUP_get_order(group, order, ctx))
770 goto err;
771
772 if (BN_is_zero(order)) {
773 ECerr(EC_F_NISTZ256_PRECOMPUTE_MULT, EC_R_UNKNOWN_ORDER);
774 goto err;
775 }
776
777 w = 7;
778
779 if ((precomp_storage =
780 OPENSSL_malloc(37 * 64 * sizeof(P256_POINT_AFFINE) + 64)) == NULL) {
781 ECerr(EC_F_NISTZ256_PRECOMPUTE_MULT, ERR_R_MALLOC_FAILURE);
782 goto err;
783 } else {
784 preComputedTable = (void *)ALIGNPTR(precomp_storage, 64);
785 }
786
787 P = EC_POINT_new(group);
788 T = EC_POINT_new(group);
789
790 /* The zero entry is implicitly infinity, and we skip it,
791 * storing other values with -1 offset. */
792 EC_POINT_copy(T, generator);
793
794 for (k = 0; k < 64; k++) {
795 EC_POINT_copy(P, T);
796 for (j = 0; j < 37; j++) {
797 /* It would be faster to use
798 * ec_GFp_simple_points_make_affine and make multiple
799 * points affine at the same time. */
800 ec_GFp_simple_make_affine(group, P, ctx);
801 ecp_nistz256_bignum_to_field_elem(preComputedTable[j]
802 [k].X, &P->X);
803 ecp_nistz256_bignum_to_field_elem(preComputedTable[j]
804 [k].Y, &P->Y);
805 for (i = 0; i < 7; i++)
806 ec_GFp_simple_dbl(group, P, P, ctx);
807 }
808 ec_GFp_simple_add(group, T, T, generator, ctx);
809 }
810
811 pre_comp->group = group;
812 pre_comp->w = w;
813 pre_comp->precomp = preComputedTable;
814 pre_comp->precomp_storage = precomp_storage;
815
816 precomp_storage = NULL;
817
818 if (!EC_EX_DATA_set_data(&group->extra_data, pre_comp,
819 ec_pre_comp_dup, ec_pre_comp_free,
820 ec_pre_comp_clear_free)) {
821 goto err;
822 }
823
824 pre_comp = NULL;
825
826 ret = 1;
827
828err:
829 if (ctx != NULL)
830 BN_CTX_end(ctx);
831 if (pre_comp)
832 ec_pre_comp_free(pre_comp);
833 if (precomp_storage)
834 OPENSSL_free(precomp_storage);
835 if (P)
836 EC_POINT_free(P);
837 if (T)
838 EC_POINT_free(T);
839 return ret;
840}
841
842/*
843 * Note that by default ECP_NISTZ256_AVX2 is undefined. While it's great
844 * code processing 4 points in parallel, corresponding serial operation
845 * is several times slower, because it uses 29x29=58-bit multiplication
846 * as opposite to 64x64=128-bit in integer-only scalar case. As result
847 * it doesn't provide *significant* performance improvement. Note that
848 * just defining ECP_NISTZ256_AVX2 is not sufficient to make it work,
849 * you'd need to compile even asm/ecp_nistz256-avx.pl module.
850 */
851#if defined(ECP_NISTZ256_AVX2)
852# if !(defined(__x86_64) || defined(__x86_64__)) || \
853 defined(_M_AMD64) || defined(_MX64)) || \
854 !(defined(__GNUC__) || defined(_MSC_VER)) /* this is for ALIGN32 */
855# undef ECP_NISTZ256_AVX2
856# else
857/* Constant time access, loading four values, from four consecutive tables */
858void ecp_nistz256_avx2_select_w7(P256_POINT_AFFINE * val,
859 const P256_POINT_AFFINE * in_t, int index);
860void ecp_nistz256_avx2_multi_select_w7(void *result, const void *in, int index0,
861 int index1, int index2, int index3);
862void ecp_nistz256_avx2_transpose_convert(void *RESULTx4, const void *in);
863void ecp_nistz256_avx2_convert_transpose_back(void *result, const void *Ax4);
864void ecp_nistz256_avx2_point_add_affine_x4(void *RESULTx4, const void *Ax4,
865 const void *Bx4);
866void ecp_nistz256_avx2_point_add_affines_x4(void *RESULTx4, const void *Ax4,
867 const void *Bx4);
868void ecp_nistz256_avx2_to_mont(void *RESULTx4, const void *Ax4);
869void ecp_nistz256_avx2_from_mont(void *RESULTx4, const void *Ax4);
870void ecp_nistz256_avx2_set1(void *RESULTx4);
871int ecp_nistz_avx2_eligible(void);
872
873static void booth_recode_w7(unsigned char *sign,
874 unsigned char *digit, unsigned char in)
875{
876 unsigned char s, d;
877
878 s = ~((in >> 7) - 1);
879 d = (1 << 8) - in - 1;
880 d = (d & s) | (in & ~s);
881 d = (d >> 1) + (d & 1);
882
883 *sign = s & 1;
884 *digit = d;
885}
886
887/* ecp_nistz256_avx2_mul_g performs multiplication by G, using only the
888 * precomputed table. It does 4 affine point additions in parallel,
889 * significantly speeding up point multiplication for a fixed value. */
890static void ecp_nistz256_avx2_mul_g(P256_POINT * r,
891 unsigned char p_str[33],
892 const
893 P256_POINT_AFFINE(*preComputedTable)[64])
894{
895 const unsigned int window_size = 7;
896 const unsigned int mask = (1 << (window_size + 1)) - 1;
897 unsigned int wvalue;
898 /* Using 4 windows at a time */
899 unsigned char sign0, digit0;
900 unsigned char sign1, digit1;
901 unsigned char sign2, digit2;
902 unsigned char sign3, digit3;
903 unsigned int index = 0;
904 BN_ULONG tmp[P256_LIMBS];
905 int i;
906
907 ALIGN32 BN_ULONG aX4[4 * 9 * 3] = { 0 };
908 ALIGN32 BN_ULONG bX4[4 * 9 * 2] = { 0 };
909 ALIGN32 P256_POINT_AFFINE point_arr[P256_LIMBS];
910 ALIGN32 P256_POINT res_point_arr[P256_LIMBS];
911
912 /* Initial four windows */
913 wvalue = *((u16 *) & p_str[0]);
914 wvalue = (wvalue << 1) & mask;
915 index += window_size;
916 booth_recode_w7(&sign0, &digit0, wvalue);
917 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
918 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
919 index += window_size;
920 booth_recode_w7(&sign1, &digit1, wvalue);
921 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
922 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
923 index += window_size;
924 booth_recode_w7(&sign2, &digit2, wvalue);
925 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
926 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
927 index += window_size;
928 booth_recode_w7(&sign3, &digit3, wvalue);
929
930 ecp_nistz256_avx2_multi_select_w7(point_arr, preComputedTable[0],
931 digit0, digit1, digit2, digit3);
932
933 ecp_nistz256_neg(tmp, point_arr[0].Y);
934 copy_conditional(point_arr[0].Y, tmp, sign0);
935 ecp_nistz256_neg(tmp, point_arr[1].Y);
936 copy_conditional(point_arr[1].Y, tmp, sign1);
937 ecp_nistz256_neg(tmp, point_arr[2].Y);
938 copy_conditional(point_arr[2].Y, tmp, sign2);
939 ecp_nistz256_neg(tmp, point_arr[3].Y);
940 copy_conditional(point_arr[3].Y, tmp, sign3);
941
942 ecp_nistz256_avx2_transpose_convert(aX4, point_arr);
943 ecp_nistz256_avx2_to_mont(aX4, aX4);
944 ecp_nistz256_avx2_to_mont(&aX4[4 * 9], &aX4[4 * 9]);
945 ecp_nistz256_avx2_set1(&aX4[4 * 9 * 2]);
946
947 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
948 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
949 index += window_size;
950 booth_recode_w7(&sign0, &digit0, wvalue);
951 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
952 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
953 index += window_size;
954 booth_recode_w7(&sign1, &digit1, wvalue);
955 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
956 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
957 index += window_size;
958 booth_recode_w7(&sign2, &digit2, wvalue);
959 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
960 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
961 index += window_size;
962 booth_recode_w7(&sign3, &digit3, wvalue);
963
964 ecp_nistz256_avx2_multi_select_w7(point_arr, preComputedTable[4 * 1],
965 digit0, digit1, digit2, digit3);
966
967 ecp_nistz256_neg(tmp, point_arr[0].Y);
968 copy_conditional(point_arr[0].Y, tmp, sign0);
969 ecp_nistz256_neg(tmp, point_arr[1].Y);
970 copy_conditional(point_arr[1].Y, tmp, sign1);
971 ecp_nistz256_neg(tmp, point_arr[2].Y);
972 copy_conditional(point_arr[2].Y, tmp, sign2);
973 ecp_nistz256_neg(tmp, point_arr[3].Y);
974 copy_conditional(point_arr[3].Y, tmp, sign3);
975
976 ecp_nistz256_avx2_transpose_convert(bX4, point_arr);
977 ecp_nistz256_avx2_to_mont(bX4, bX4);
978 ecp_nistz256_avx2_to_mont(&bX4[4 * 9], &bX4[4 * 9]);
979 /* Optimized when both inputs are affine */
980 ecp_nistz256_avx2_point_add_affines_x4(aX4, aX4, bX4);
981
982 for (i = 2; i < 9; i++) {
983 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
984 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
985 index += window_size;
986 booth_recode_w7(&sign0, &digit0, wvalue);
987 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
988 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
989 index += window_size;
990 booth_recode_w7(&sign1, &digit1, wvalue);
991 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
992 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
993 index += window_size;
994 booth_recode_w7(&sign2, &digit2, wvalue);
995 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
996 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
997 index += window_size;
998 booth_recode_w7(&sign3, &digit3, wvalue);
999
1000 ecp_nistz256_avx2_multi_select_w7(point_arr,
1001 preComputedTable[4 * i],
1002 digit0, digit1, digit2, digit3);
1003
1004 ecp_nistz256_neg(tmp, point_arr[0].Y);
1005 copy_conditional(point_arr[0].Y, tmp, sign0);
1006 ecp_nistz256_neg(tmp, point_arr[1].Y);
1007 copy_conditional(point_arr[1].Y, tmp, sign1);
1008 ecp_nistz256_neg(tmp, point_arr[2].Y);
1009 copy_conditional(point_arr[2].Y, tmp, sign2);
1010 ecp_nistz256_neg(tmp, point_arr[3].Y);
1011 copy_conditional(point_arr[3].Y, tmp, sign3);
1012
1013 ecp_nistz256_avx2_transpose_convert(bX4, point_arr);
1014 ecp_nistz256_avx2_to_mont(bX4, bX4);
1015 ecp_nistz256_avx2_to_mont(&bX4[4 * 9], &bX4[4 * 9]);
1016
1017 ecp_nistz256_avx2_point_add_affine_x4(aX4, aX4, bX4);
1018 }
1019
1020 ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 0], &aX4[4 * 9 * 0]);
1021 ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 1], &aX4[4 * 9 * 1]);
1022 ecp_nistz256_avx2_from_mont(&aX4[4 * 9 * 2], &aX4[4 * 9 * 2]);
1023
1024 ecp_nistz256_avx2_convert_transpose_back(res_point_arr, aX4);
1025 /* Last window is performed serially */
1026 wvalue = *((u16 *) & p_str[(index - 1) / 8]);
1027 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1028 booth_recode_w7(&sign0, &digit0, wvalue);
1029 ecp_nistz256_avx2_select_w7((P256_POINT_AFFINE *) r,
1030 preComputedTable[36], digit0);
1031 ecp_nistz256_neg(tmp, r->Y);
1032 copy_conditional(r->Y, tmp, sign0);
1033 memcpy(r->Z, ONE, sizeof(ONE));
1034 /* Sum the four windows */
1035 ecp_nistz256_point_add(r, r, &res_point_arr[0]);
1036 ecp_nistz256_point_add(r, r, &res_point_arr[1]);
1037 ecp_nistz256_point_add(r, r, &res_point_arr[2]);
1038 ecp_nistz256_point_add(r, r, &res_point_arr[3]);
1039}
1040# endif
1041#endif
1042
1043static int ecp_nistz256_set_from_affine(EC_POINT * out, const EC_GROUP * group,
1044 const P256_POINT_AFFINE * in,
1045 BN_CTX * ctx)
1046{
1047 BIGNUM x, y;
1048 BN_ULONG d_x[P256_LIMBS], d_y[P256_LIMBS];
1049 int ret = 0;
1050
1051 memcpy(d_x, in->X, sizeof(d_x));
1052 x.d = d_x;
1053 x.dmax = x.top = P256_LIMBS;
1054 x.neg = 0;
1055 x.flags = BN_FLG_STATIC_DATA;
1056
1057 memcpy(d_y, in->Y, sizeof(d_y));
1058 y.d = d_y;
1059 y.dmax = y.top = P256_LIMBS;
1060 y.neg = 0;
1061 y.flags = BN_FLG_STATIC_DATA;
1062
1063 ret = EC_POINT_set_affine_coordinates_GFp(group, out, &x, &y, ctx);
1064
1065 return ret;
1066}
1067
1068/* r = scalar*G + sum(scalars[i]*points[i]) */
1069static int ecp_nistz256_points_mul(const EC_GROUP * group,
1070 EC_POINT * r,
1071 const BIGNUM * scalar,
1072 size_t num,
1073 const EC_POINT * points[],
1074 const BIGNUM * scalars[], BN_CTX * ctx)
1075{
1076 int i = 0, ret = 0, no_precomp_for_generator = 0, p_is_infinity = 0;
16e5b45f 1077 size_t j;
4d3fa06f
AP		 1078 unsigned char p_str[33] = { 0 };
1079 const PRECOMP256_ROW *preComputedTable = NULL;
1080 const EC_PRE_COMP *pre_comp = NULL;
1081 const EC_POINT *generator = NULL;
1082 unsigned int index = 0;
1083 const unsigned int window_size = 7;
1084 const unsigned int mask = (1 << (window_size + 1)) - 1;
1085 unsigned int wvalue;
1086 ALIGN32 union {
1087 P256_POINT p;
1088 P256_POINT_AFFINE a;
1089 } t, p;
1090 BIGNUM *tmp_scalar;
1091
1092 if (group->meth != r->meth) {
1093 ECerr(EC_F_NISTZ256_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS);
1094 return 0;
1095 }
1096 if ((scalar == NULL) && (num == 0))
1097 return EC_POINT_set_to_infinity(group, r);
1098
16e5b45f
DSH		 1099 for (j = 0; j < num; j++) {
1100 if (group->meth != points[j]->meth) {
4d3fa06f
AP		 1101 ECerr(EC_F_NISTZ256_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS);
1102 return 0;
1103 }
1104 }
1105
1106 /* Need 256 bits for space for all coordinates. */
1107 bn_wexpand(&r->X, P256_LIMBS);
1108 bn_wexpand(&r->Y, P256_LIMBS);
1109 bn_wexpand(&r->Z, P256_LIMBS);
1110 r->X.top = P256_LIMBS;
1111 r->Y.top = P256_LIMBS;
1112 r->Z.top = P256_LIMBS;
1113
1114 if (scalar) {
1115 generator = EC_GROUP_get0_generator(group);
1116 if (generator == NULL) {
1117 ECerr(EC_F_NISTZ256_POINTS_MUL, EC_R_UNDEFINED_GENERATOR);
1118 goto err;
1119 }
1120
1121 /* look if we can use precomputed multiples of generator */
1122 pre_comp =
1123 EC_EX_DATA_get_data(group->extra_data, ec_pre_comp_dup,
1124 ec_pre_comp_free, ec_pre_comp_clear_free);
1125
1126 if (pre_comp) {
1127 /* If there is a precomputed table for the generator,
1128 * check that it was generated with the same
1129 * generator. */
1130 EC_POINT *pre_comp_generator = EC_POINT_new(group);
1131 if (pre_comp_generator == NULL)
1132 goto err;
1133
1134 if (!ecp_nistz256_set_from_affine
1135 (pre_comp_generator, group, pre_comp->precomp[0], ctx))
1136 goto err;
1137
1138 if (0 == EC_POINT_cmp(group, generator, pre_comp_generator, ctx))
1139 preComputedTable = (const PRECOMP256_ROW *)pre_comp->precomp;
1140
1141 EC_POINT_free(pre_comp_generator);
1142 }
1143
1144 if (preComputedTable == NULL && ecp_nistz256_is_affine_G(generator)) {
1145 /* If there is no precomputed data, but the generator
1146 * is the default, a hardcoded table of precomputed
1147 * data is used. This is because applications, such as
1148 * Apache, do not use EC_KEY_precompute_mult. */
1149 preComputedTable = (const PRECOMP256_ROW *)ecp_nistz256_precomputed;
1150 }
1151
1152 if (preComputedTable) {
1153 if ((BN_num_bits(scalar) > 256)
1154 || BN_is_negative(scalar)) {
1155 if ((tmp_scalar = BN_CTX_get(ctx)) == NULL)
1156 goto err;
1157
1158 if (!BN_nnmod(tmp_scalar, scalar, &group->order, ctx)) {
1159 ECerr(EC_F_NISTZ256_POINTS_MUL, ERR_R_BN_LIB);
1160 goto err;
1161 }
1162 scalar = tmp_scalar;
1163 }
1164
1165 for (i = 0; i < scalar->top * BN_BYTES; i += BN_BYTES) {
1166 BN_ULONG d = scalar->d[i / BN_BYTES];
1167
1168 p_str[i + 0] = d & 0xff;
1169 p_str[i + 1] = (d >> 8) & 0xff;
1170 p_str[i + 2] = (d >> 16) & 0xff;
1171 p_str[i + 3] = (d >>= 24) & 0xff;
1172 if (BN_BYTES == 8) {
1173 d >>= 8;
1174 p_str[i + 4] = d & 0xff;
1175 p_str[i + 5] = (d >> 8) & 0xff;
1176 p_str[i + 6] = (d >> 16) & 0xff;
1177 p_str[i + 7] = (d >> 24) & 0xff;
1178 }
1179 }
1180
1181 for (; i < 33; i++)
1182 p_str[i] = 0;
1183
1184#if defined(ECP_NISTZ256_AVX2)
1185 if (ecp_nistz_avx2_eligible()) {
1186 ecp_nistz256_avx2_mul_g(&p.p, p_str, preComputedTable);
1187 } else
1188#endif
1189 {
1190 /* First window */
1191 wvalue = (p_str[0] << 1) & mask;
1192 index += window_size;
1193
1194 wvalue = _booth_recode_w7(wvalue);
1195
1196 ecp_nistz256_select_w7(&p.a, preComputedTable[0], wvalue >> 1);
1197
1198 ecp_nistz256_neg(p.p.Z, p.p.Y);
1199 copy_conditional(p.p.Y, p.p.Z, wvalue & 1);
1200
1201 memcpy(p.p.Z, ONE, sizeof(ONE));
1202
1203 for (i = 1; i < 37; i++) {
1204 unsigned int off = (index - 1) / 8;
1205 wvalue = p_str[off] | p_str[off + 1] << 8;
1206 wvalue = (wvalue >> ((index - 1) % 8)) & mask;
1207 index += window_size;
1208
1209 wvalue = _booth_recode_w7(wvalue);
1210
1211 ecp_nistz256_select_w7(&t.a,
1212 preComputedTable[i], wvalue >> 1);
1213
1214 ecp_nistz256_neg(t.p.Z, t.a.Y);
1215 copy_conditional(t.a.Y, t.p.Z, wvalue & 1);
1216
1217 ecp_nistz256_point_add_affine(&p.p, &p.p, &t.a);
1218 }
1219 }
1220 } else {
1221 p_is_infinity = 1;
1222 no_precomp_for_generator = 1;
1223 }
1224 } else
1225 p_is_infinity = 1;
1226
1227 if (no_precomp_for_generator) {
1228 /* Without a precomputed table for the generator, it has to be
1229 * handled like a normal point. */
1230 const BIGNUM **new_scalars;
1231 const EC_POINT **new_points;
1232
1233 new_scalars = OPENSSL_malloc((num + 1) * sizeof(BIGNUM *));
1234 if (!new_scalars) {
1235 ECerr(EC_F_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
1236 return 0;
1237 }
1238
1239 new_points = OPENSSL_malloc((num + 1) * sizeof(EC_POINT *));
1240 if (!new_points) {
1241 OPENSSL_free(new_scalars);
1242 ECerr(EC_F_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
1243 return 0;
1244 }
1245
1246 memcpy(new_scalars, scalars, num * sizeof(BIGNUM *));
1247 new_scalars[num] = scalar;
1248 memcpy(new_points, points, num * sizeof(EC_POINT *));
1249 new_points[num] = generator;
1250
1251 scalars = new_scalars;
1252 points = new_points;
1253 num++;
1254 }
1255
1256 if (num) {
1257 P256_POINT *out = &t.p;
1258 if (p_is_infinity)
1259 out = &p.p;
1260
1261 ecp_nistz256_windowed_mul(group, out, scalars, points, num, ctx);
1262
1263 if (!p_is_infinity)
1264 ecp_nistz256_point_add(&p.p, &p.p, out);
1265 }
1266
1267 if (no_precomp_for_generator) {
1268 OPENSSL_free(points);
1269 OPENSSL_free(scalars);
1270 }
1271
1272 memcpy(r->X.d, p.p.X, sizeof(p.p.X));
1273 memcpy(r->Y.d, p.p.Y, sizeof(p.p.Y));
1274 memcpy(r->Z.d, p.p.Z, sizeof(p.p.Z));
1275 bn_correct_top(&r->X);
1276 bn_correct_top(&r->Y);
1277 bn_correct_top(&r->Z);
1278
1279 ret = 1;
1280
1281err:
1282 return ret;
1283}
1284
1285static int ecp_nistz256_get_affine(const EC_GROUP * group,
1286 const EC_POINT * point,
1287 BIGNUM * x, BIGNUM * y, BN_CTX * ctx)
1288{
1289 BN_ULONG z_inv2[P256_LIMBS];
1290 BN_ULONG z_inv3[P256_LIMBS];
1291 BN_ULONG x_aff[P256_LIMBS];
1292 BN_ULONG y_aff[P256_LIMBS];
1293 BN_ULONG point_x[P256_LIMBS], point_y[P256_LIMBS], point_z[P256_LIMBS];
1294
1295 if (EC_POINT_is_at_infinity(group, point)) {
1296 ECerr(EC_F_NISTZ256_GET_AFFINE_COORDINATES, EC_R_POINT_AT_INFINITY);
1297 return 0;
1298 }
1299
1300 if (!ecp_nistz256_bignum_to_field_elem(point_x, &point->X) ||
1301 !ecp_nistz256_bignum_to_field_elem(point_y, &point->Y) ||
1302 !ecp_nistz256_bignum_to_field_elem(point_z, &point->Z)) {
1303 ECerr(EC_F_NISTZ256_GET_AFFINE_COORDINATES,
1304 EC_R_COORDINATES_OUT_OF_RANGE);
1305 return 0;
1306 }
1307
1308 ecp_nistz256_mod_inverse(z_inv3, point_z);
1309 ecp_nistz256_sqr_mont(z_inv2, z_inv3);
1310 ecp_nistz256_mul_mont(x_aff, z_inv2, point_x);
1311
1312 if (x != NULL) {
1313 bn_wexpand(x, P256_LIMBS);
1314 x->top = P256_LIMBS;
1315 ecp_nistz256_from_mont(x->d, x_aff);
1316 bn_correct_top(x);
1317 }
1318
1319 if (y != NULL) {
1320 ecp_nistz256_mul_mont(z_inv3, z_inv3, z_inv2);
1321 ecp_nistz256_mul_mont(y_aff, z_inv3, point_y);
1322 bn_wexpand(y, P256_LIMBS);
1323 y->top = P256_LIMBS;
1324 ecp_nistz256_from_mont(y->d, y_aff);
1325 bn_correct_top(y);
1326 }
1327
1328 return 1;
1329}
1330
1331static EC_PRE_COMP *ec_pre_comp_new(const EC_GROUP * group)
1332{
1333 EC_PRE_COMP *ret = NULL;
1334
1335 if (!group)
1336 return NULL;
1337
1338 ret = (EC_PRE_COMP *) OPENSSL_malloc(sizeof(EC_PRE_COMP));
1339
1340 if (!ret) {
1341 ECerr(EC_F_NISTZ256_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
1342 return ret;
1343 }
1344
1345 ret->group = group;
1346 ret->w = 6; /* default */
1347 ret->precomp = NULL;
1348 ret->precomp_storage = NULL;
1349 ret->references = 1;
1350 return ret;
1351}
1352
1353static void *ec_pre_comp_dup(void *src_)
1354{
1355 EC_PRE_COMP *src = src_;
1356
1357 /* no need to actually copy, these objects never change! */
1358 CRYPTO_add(&src->references, 1, CRYPTO_LOCK_EC_PRE_COMP);
1359
1360 return src_;
1361}
1362
1363static void ec_pre_comp_free(void *pre_)
1364{
1365 int i;
1366 EC_PRE_COMP *pre = pre_;
1367
1368 if (!pre)
1369 return;
1370
1371 i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP);
1372 if (i > 0)
1373 return;
1374
1375 if (pre->precomp_storage)
1376 OPENSSL_free(pre->precomp_storage);
1377
1378 OPENSSL_free(pre);
1379}
1380
1381static void ec_pre_comp_clear_free(void *pre_)
1382{
1383 int i;
1384 EC_PRE_COMP *pre = pre_;
1385
1386 if (!pre)
1387 return;
1388
1389 i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP);
1390 if (i > 0)
1391 return;
1392
1393 if (pre->precomp_storage) {
1394 OPENSSL_cleanse(pre->precomp,
1395 32 * sizeof(unsigned char) * (1 << pre->w) * 2 * 37);
1396 OPENSSL_free(pre->precomp_storage);
1397 }
1398 OPENSSL_cleanse(pre, sizeof *pre);
1399 OPENSSL_free(pre);
1400}
1401
1402static int ecp_nistz256_window_have_precompute_mult(const EC_GROUP * group)
1403{
1404 /* There is a hard-coded table for the default generator. */
1405 const EC_POINT *generator = EC_GROUP_get0_generator(group);
1406 if (generator != NULL && ecp_nistz256_is_affine_G(generator)) {
1407 /* There is a hard-coded table for the default generator. */
1408 return 1;
1409 }
1410
1411 return EC_EX_DATA_get_data(group->extra_data, ec_pre_comp_dup,
1412 ec_pre_comp_free,
1413 ec_pre_comp_clear_free) != NULL;
1414}
1415
1416const EC_METHOD *EC_GFp_nistz256_method(void)
1417{
1418 static const EC_METHOD ret = {
1419 EC_FLAGS_DEFAULT_OCT,
1420 NID_X9_62_prime_field,
1421 ec_GFp_mont_group_init,
1422 ec_GFp_mont_group_finish,
1423 ec_GFp_mont_group_clear_finish,
1424 ec_GFp_mont_group_copy,
1425 ec_GFp_mont_group_set_curve,
1426 ec_GFp_simple_group_get_curve,
1427 ec_GFp_simple_group_get_degree,
1428 ec_GFp_simple_group_check_discriminant,
1429 ec_GFp_simple_point_init,
1430 ec_GFp_simple_point_finish,
1431 ec_GFp_simple_point_clear_finish,
1432 ec_GFp_simple_point_copy,
1433 ec_GFp_simple_point_set_to_infinity,
1434 ec_GFp_simple_set_Jprojective_coordinates_GFp,
1435 ec_GFp_simple_get_Jprojective_coordinates_GFp,
1436 ec_GFp_simple_point_set_affine_coordinates,
1437 ecp_nistz256_get_affine,
1438 0, 0, 0,
1439 ec_GFp_simple_add,
1440 ec_GFp_simple_dbl,
1441 ec_GFp_simple_invert,
1442 ec_GFp_simple_is_at_infinity,
1443 ec_GFp_simple_is_on_curve,
1444 ec_GFp_simple_cmp,
1445 ec_GFp_simple_make_affine,
1446 ec_GFp_simple_points_make_affine,
1447 ecp_nistz256_points_mul, /* mul */
1448 ecp_nistz256_mult_precompute, /* precompute_mult */
1449 ecp_nistz256_window_have_precompute_mult, /* have_precompute_mult */
1450 ec_GFp_mont_field_mul,
1451 ec_GFp_mont_field_sqr,
1452 0, /* field_div */
1453 ec_GFp_mont_field_encode,
1454 ec_GFp_mont_field_decode,
1455 ec_GFp_mont_field_set_to_one
1456 };
1457
1458 return &ret;
1459}
A mirror of the official openssl repository