]> git.ipfire.org Git - thirdparty/openssl.git/blob - crypto/ec/ec_mult.c
Pass through
[thirdparty/openssl.git] / crypto / ec / ec_mult.c
1 /*
2 * Copyright 2001-2017 The OpenSSL Project Authors. All Rights Reserved.
3 * Copyright (c) 2002, Oracle and/or its affiliates. All rights reserved
4 *
5 * Licensed under the OpenSSL license (the "License"). You may not use
6 * this file except in compliance with the License. You can obtain a copy
7 * in the file LICENSE in the source distribution or at
8 * https://www.openssl.org/source/license.html
9 */
10
11 #include <string.h>
12 #include <openssl/err.h>
13
14 #include "internal/cryptlib.h"
15 #include "internal/bn_int.h"
16 #include "ec_lcl.h"
17 #include "internal/refcount.h"
18
19 /*
20 * This file implements the wNAF-based interleaving multi-exponentiation method
21 * Formerly at:
22 * http://www.informatik.tu-darmstadt.de/TI/Mitarbeiter/moeller.html#multiexp
23 * You might now find it here:
24 * http://link.springer.com/chapter/10.1007%2F3-540-45537-X_13
25 * http://www.bmoeller.de/pdf/TI-01-08.multiexp.pdf
26 * For multiplication with precomputation, we use wNAF splitting, formerly at:
27 * http://www.informatik.tu-darmstadt.de/TI/Mitarbeiter/moeller.html#fastexp
28 */
29
30 /* structure for precomputed multiples of the generator */
31 struct ec_pre_comp_st {
32 const EC_GROUP *group; /* parent EC_GROUP object */
33 size_t blocksize; /* block size for wNAF splitting */
34 size_t numblocks; /* max. number of blocks for which we have
35 * precomputation */
36 size_t w; /* window size */
37 EC_POINT **points; /* array with pre-calculated multiples of
38 * generator: 'num' pointers to EC_POINT
39 * objects followed by a NULL */
40 size_t num; /* numblocks * 2^(w-1) */
41 CRYPTO_REF_COUNT references;
42 CRYPTO_RWLOCK *lock;
43 };
44
45 static EC_PRE_COMP *ec_pre_comp_new(const EC_GROUP *group)
46 {
47 EC_PRE_COMP *ret = NULL;
48
49 if (!group)
50 return NULL;
51
52 ret = OPENSSL_zalloc(sizeof(*ret));
53 if (ret == NULL) {
54 ECerr(EC_F_EC_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
55 return ret;
56 }
57
58 ret->group = group;
59 ret->blocksize = 8; /* default */
60 ret->w = 4; /* default */
61 ret->references = 1;
62
63 ret->lock = CRYPTO_THREAD_lock_new();
64 if (ret->lock == NULL) {
65 ECerr(EC_F_EC_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
66 OPENSSL_free(ret);
67 return NULL;
68 }
69 return ret;
70 }
71
72 EC_PRE_COMP *EC_ec_pre_comp_dup(EC_PRE_COMP *pre)
73 {
74 int i;
75 if (pre != NULL)
76 CRYPTO_UP_REF(&pre->references, &i, pre->lock);
77 return pre;
78 }
79
80 void EC_ec_pre_comp_free(EC_PRE_COMP *pre)
81 {
82 int i;
83
84 if (pre == NULL)
85 return;
86
87 CRYPTO_DOWN_REF(&pre->references, &i, pre->lock);
88 REF_PRINT_COUNT("EC_ec", pre);
89 if (i > 0)
90 return;
91 REF_ASSERT_ISNT(i < 0);
92
93 if (pre->points != NULL) {
94 EC_POINT **pts;
95
96 for (pts = pre->points; *pts != NULL; pts++)
97 EC_POINT_free(*pts);
98 OPENSSL_free(pre->points);
99 }
100 CRYPTO_THREAD_lock_free(pre->lock);
101 OPENSSL_free(pre);
102 }
103
104 #define EC_POINT_BN_set_flags(P, flags) do { \
105 BN_set_flags((P)->X, (flags)); \
106 BN_set_flags((P)->Y, (flags)); \
107 BN_set_flags((P)->Z, (flags)); \
108 } while(0)
109
110 /*-
111 * This functions computes (in constant time) a point multiplication over the
112 * EC group.
113 *
114 * At a high level, it is Montgomery ladder with conditional swaps.
115 *
116 * It performs either a fixed scalar point multiplication
117 * (scalar * generator)
118 * when point is NULL, or a generic scalar point multiplication
119 * (scalar * point)
120 * when point is not NULL.
121 *
122 * scalar should be in the range [0,n) otherwise all constant time bets are off.
123 *
124 * NB: This says nothing about EC_POINT_add and EC_POINT_dbl,
125 * which of course are not constant time themselves.
126 *
127 * The product is stored in r.
128 *
129 * Returns 1 on success, 0 otherwise.
130 */
131 static int ec_mul_consttime(const EC_GROUP *group, EC_POINT *r,
132 const BIGNUM *scalar, const EC_POINT *point,
133 BN_CTX *ctx)
134 {
135 int i, order_bits, group_top, kbit, pbit, Z_is_one;
136 EC_POINT *s = NULL;
137 BIGNUM *k = NULL;
138 BIGNUM *lambda = NULL;
139 BN_CTX *new_ctx = NULL;
140 int ret = 0;
141
142 if (ctx == NULL && (ctx = new_ctx = BN_CTX_secure_new()) == NULL)
143 goto err;
144
145 if ((group->order == NULL) || (group->field == NULL))
146 goto err;
147
148 order_bits = BN_num_bits(group->order);
149
150 s = EC_POINT_new(group);
151 if (s == NULL)
152 goto err;
153
154 if (point == NULL) {
155 if (group->generator == NULL)
156 goto err;
157 if (!EC_POINT_copy(s, group->generator))
158 goto err;
159 } else {
160 if (!EC_POINT_copy(s, point))
161 goto err;
162 }
163
164 EC_POINT_BN_set_flags(s, BN_FLG_CONSTTIME);
165
166 BN_CTX_start(ctx);
167 lambda = BN_CTX_get(ctx);
168 k = BN_CTX_get(ctx);
169 if (k == NULL)
170 goto err;
171
172 /*
173 * Group orders are often on a word boundary.
174 * So when we pad the scalar, some timing diff might
175 * pop if it needs to be expanded due to carries.
176 * So expand ahead of time.
177 */
178 group_top = bn_get_top(group->order);
179 if ((bn_wexpand(k, group_top + 1) == NULL)
180 || (bn_wexpand(lambda, group_top + 1) == NULL))
181 goto err;
182
183 if (!BN_copy(k, scalar))
184 goto err;
185
186 BN_set_flags(k, BN_FLG_CONSTTIME);
187
188 if ((BN_num_bits(k) > order_bits) || (BN_is_negative(k))) {
189 /*-
190 * this is an unusual input, and we don't guarantee
191 * constant-timeness
192 */
193 if (!BN_nnmod(k, k, group->order, ctx))
194 goto err;
195 }
196
197 if (!BN_add(lambda, k, group->order))
198 goto err;
199 BN_set_flags(lambda, BN_FLG_CONSTTIME);
200 if (!BN_add(k, lambda, group->order))
201 goto err;
202 /*
203 * lambda := scalar + order
204 * k := scalar + 2*order
205 */
206 kbit = BN_is_bit_set(lambda, order_bits);
207 BN_consttime_swap(kbit, k, lambda, group_top + 1);
208
209 group_top = bn_get_top(group->field);
210 if ((bn_wexpand(s->X, group_top) == NULL)
211 || (bn_wexpand(s->Y, group_top) == NULL)
212 || (bn_wexpand(s->Z, group_top) == NULL)
213 || (bn_wexpand(r->X, group_top) == NULL)
214 || (bn_wexpand(r->Y, group_top) == NULL)
215 || (bn_wexpand(r->Z, group_top) == NULL))
216 goto err;
217
218 /* top bit is a 1, in a fixed pos */
219 if (!EC_POINT_copy(r, s))
220 goto err;
221
222 EC_POINT_BN_set_flags(r, BN_FLG_CONSTTIME);
223
224 if (!EC_POINT_dbl(group, s, s, ctx))
225 goto err;
226
227 pbit = 0;
228
229 #define EC_POINT_CSWAP(c, a, b, w, t) do { \
230 BN_consttime_swap(c, (a)->X, (b)->X, w); \
231 BN_consttime_swap(c, (a)->Y, (b)->Y, w); \
232 BN_consttime_swap(c, (a)->Z, (b)->Z, w); \
233 t = ((a)->Z_is_one ^ (b)->Z_is_one) & (c); \
234 (a)->Z_is_one ^= (t); \
235 (b)->Z_is_one ^= (t); \
236 } while(0)
237
238 /*-
239 * The ladder step, with branches, is
240 *
241 * k[i] == 0: S = add(R, S), R = dbl(R)
242 * k[i] == 1: R = add(S, R), S = dbl(S)
243 *
244 * Swapping R, S conditionally on k[i] leaves you with state
245 *
246 * k[i] == 0: T, U = R, S
247 * k[i] == 1: T, U = S, R
248 *
249 * Then perform the ECC ops.
250 *
251 * U = add(T, U)
252 * T = dbl(T)
253 *
254 * Which leaves you with state
255 *
256 * k[i] == 0: U = add(R, S), T = dbl(R)
257 * k[i] == 1: U = add(S, R), T = dbl(S)
258 *
259 * Swapping T, U conditionally on k[i] leaves you with state
260 *
261 * k[i] == 0: R, S = T, U
262 * k[i] == 1: R, S = U, T
263 *
264 * Which leaves you with state
265 *
266 * k[i] == 0: S = add(R, S), R = dbl(R)
267 * k[i] == 1: R = add(S, R), S = dbl(S)
268 *
269 * So we get the same logic, but instead of a branch it's a
270 * conditional swap, followed by ECC ops, then another conditional swap.
271 *
272 * Optimization: The end of iteration i and start of i-1 looks like
273 *
274 * ...
275 * CSWAP(k[i], R, S)
276 * ECC
277 * CSWAP(k[i], R, S)
278 * (next iteration)
279 * CSWAP(k[i-1], R, S)
280 * ECC
281 * CSWAP(k[i-1], R, S)
282 * ...
283 *
284 * So instead of two contiguous swaps, you can merge the condition
285 * bits and do a single swap.
286 *
287 * k[i] k[i-1] Outcome
288 * 0 0 No Swap
289 * 0 1 Swap
290 * 1 0 Swap
291 * 1 1 No Swap
292 *
293 * This is XOR. pbit tracks the previous bit of k.
294 */
295
296 for (i = order_bits - 1; i >= 0; i--) {
297 kbit = BN_is_bit_set(k, i) ^ pbit;
298 EC_POINT_CSWAP(kbit, r, s, group_top, Z_is_one);
299 if (!EC_POINT_add(group, s, r, s, ctx))
300 goto err;
301 if (!EC_POINT_dbl(group, r, r, ctx))
302 goto err;
303 /*
304 * pbit logic merges this cswap with that of the
305 * next iteration
306 */
307 pbit ^= kbit;
308 }
309 /* one final cswap to move the right value into r */
310 EC_POINT_CSWAP(pbit, r, s, group_top, Z_is_one);
311 #undef EC_POINT_CSWAP
312
313 ret = 1;
314
315 err:
316 EC_POINT_free(s);
317 BN_CTX_end(ctx);
318 BN_CTX_free(new_ctx);
319
320 return ret;
321 }
322
323 #undef EC_POINT_BN_set_flags
324
325 /*
326 * TODO: table should be optimised for the wNAF-based implementation,
327 * sometimes smaller windows will give better performance (thus the
328 * boundaries should be increased)
329 */
330 #define EC_window_bits_for_scalar_size(b) \
331 ((size_t) \
332 ((b) >= 2000 ? 6 : \
333 (b) >= 800 ? 5 : \
334 (b) >= 300 ? 4 : \
335 (b) >= 70 ? 3 : \
336 (b) >= 20 ? 2 : \
337 1))
338
339 /*-
340 * Compute
341 * \sum scalars[i]*points[i],
342 * also including
343 * scalar*generator
344 * in the addition if scalar != NULL
345 */
346 int ec_wNAF_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar,
347 size_t num, const EC_POINT *points[], const BIGNUM *scalars[],
348 BN_CTX *ctx)
349 {
350 BN_CTX *new_ctx = NULL;
351 const EC_POINT *generator = NULL;
352 EC_POINT *tmp = NULL;
353 size_t totalnum;
354 size_t blocksize = 0, numblocks = 0; /* for wNAF splitting */
355 size_t pre_points_per_block = 0;
356 size_t i, j;
357 int k;
358 int r_is_inverted = 0;
359 int r_is_at_infinity = 1;
360 size_t *wsize = NULL; /* individual window sizes */
361 signed char **wNAF = NULL; /* individual wNAFs */
362 size_t *wNAF_len = NULL;
363 size_t max_len = 0;
364 size_t num_val;
365 EC_POINT **val = NULL; /* precomputation */
366 EC_POINT **v;
367 EC_POINT ***val_sub = NULL; /* pointers to sub-arrays of 'val' or
368 * 'pre_comp->points' */
369 const EC_PRE_COMP *pre_comp = NULL;
370 int num_scalar = 0; /* flag: will be set to 1 if 'scalar' must be
371 * treated like other scalars, i.e.
372 * precomputation is not available */
373 int ret = 0;
374
375 /*-
376 * Handle the common cases where the scalar is secret, enforcing a constant
377 * time scalar multiplication algorithm.
378 */
379 if ((scalar != NULL) && (num == 0)) {
380 /*-
381 * In this case we want to compute scalar * GeneratorPoint: this
382 * codepath is reached most prominently by (ephemeral) key generation
383 * of EC cryptosystems (i.e. ECDSA keygen and sign setup, ECDH
384 * keygen/first half), where the scalar is always secret. This is why
385 * we ignore if BN_FLG_CONSTTIME is actually set and we always call the
386 * constant time version.
387 */
388 return ec_mul_consttime(group, r, scalar, NULL, ctx);
389 }
390 if ((scalar == NULL) && (num == 1)) {
391 /*-
392 * In this case we want to compute scalar * GenericPoint: this codepath
393 * is reached most prominently by the second half of ECDH, where the
394 * secret scalar is multiplied by the peer's public point. To protect
395 * the secret scalar, we ignore if BN_FLG_CONSTTIME is actually set and
396 * we always call the constant time version.
397 */
398 return ec_mul_consttime(group, r, scalars[0], points[0], ctx);
399 }
400
401 if (group->meth != r->meth) {
402 ECerr(EC_F_EC_WNAF_MUL, EC_R_INCOMPATIBLE_OBJECTS);
403 return 0;
404 }
405
406 if ((scalar == NULL) && (num == 0)) {
407 return EC_POINT_set_to_infinity(group, r);
408 }
409
410 for (i = 0; i < num; i++) {
411 if (group->meth != points[i]->meth) {
412 ECerr(EC_F_EC_WNAF_MUL, EC_R_INCOMPATIBLE_OBJECTS);
413 return 0;
414 }
415 }
416
417 if (ctx == NULL) {
418 ctx = new_ctx = BN_CTX_new();
419 if (ctx == NULL)
420 goto err;
421 }
422
423 if (scalar != NULL) {
424 generator = EC_GROUP_get0_generator(group);
425 if (generator == NULL) {
426 ECerr(EC_F_EC_WNAF_MUL, EC_R_UNDEFINED_GENERATOR);
427 goto err;
428 }
429
430 /* look if we can use precomputed multiples of generator */
431
432 pre_comp = group->pre_comp.ec;
433 if (pre_comp && pre_comp->numblocks
434 && (EC_POINT_cmp(group, generator, pre_comp->points[0], ctx) ==
435 0)) {
436 blocksize = pre_comp->blocksize;
437
438 /*
439 * determine maximum number of blocks that wNAF splitting may
440 * yield (NB: maximum wNAF length is bit length plus one)
441 */
442 numblocks = (BN_num_bits(scalar) / blocksize) + 1;
443
444 /*
445 * we cannot use more blocks than we have precomputation for
446 */
447 if (numblocks > pre_comp->numblocks)
448 numblocks = pre_comp->numblocks;
449
450 pre_points_per_block = (size_t)1 << (pre_comp->w - 1);
451
452 /* check that pre_comp looks sane */
453 if (pre_comp->num != (pre_comp->numblocks * pre_points_per_block)) {
454 ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR);
455 goto err;
456 }
457 } else {
458 /* can't use precomputation */
459 pre_comp = NULL;
460 numblocks = 1;
461 num_scalar = 1; /* treat 'scalar' like 'num'-th element of
462 * 'scalars' */
463 }
464 }
465
466 totalnum = num + numblocks;
467
468 wsize = OPENSSL_malloc(totalnum * sizeof(wsize[0]));
469 wNAF_len = OPENSSL_malloc(totalnum * sizeof(wNAF_len[0]));
470 /* include space for pivot */
471 wNAF = OPENSSL_malloc((totalnum + 1) * sizeof(wNAF[0]));
472 val_sub = OPENSSL_malloc(totalnum * sizeof(val_sub[0]));
473
474 /* Ensure wNAF is initialised in case we end up going to err */
475 if (wNAF != NULL)
476 wNAF[0] = NULL; /* preliminary pivot */
477
478 if (wsize == NULL || wNAF_len == NULL || wNAF == NULL || val_sub == NULL) {
479 ECerr(EC_F_EC_WNAF_MUL, ERR_R_MALLOC_FAILURE);
480 goto err;
481 }
482
483 /*
484 * num_val will be the total number of temporarily precomputed points
485 */
486 num_val = 0;
487
488 for (i = 0; i < num + num_scalar; i++) {
489 size_t bits;
490
491 bits = i < num ? BN_num_bits(scalars[i]) : BN_num_bits(scalar);
492 wsize[i] = EC_window_bits_for_scalar_size(bits);
493 num_val += (size_t)1 << (wsize[i] - 1);
494 wNAF[i + 1] = NULL; /* make sure we always have a pivot */
495 wNAF[i] =
496 bn_compute_wNAF((i < num ? scalars[i] : scalar), wsize[i],
497 &wNAF_len[i]);
498 if (wNAF[i] == NULL)
499 goto err;
500 if (wNAF_len[i] > max_len)
501 max_len = wNAF_len[i];
502 }
503
504 if (numblocks) {
505 /* we go here iff scalar != NULL */
506
507 if (pre_comp == NULL) {
508 if (num_scalar != 1) {
509 ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR);
510 goto err;
511 }
512 /* we have already generated a wNAF for 'scalar' */
513 } else {
514 signed char *tmp_wNAF = NULL;
515 size_t tmp_len = 0;
516
517 if (num_scalar != 0) {
518 ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR);
519 goto err;
520 }
521
522 /*
523 * use the window size for which we have precomputation
524 */
525 wsize[num] = pre_comp->w;
526 tmp_wNAF = bn_compute_wNAF(scalar, wsize[num], &tmp_len);
527 if (!tmp_wNAF)
528 goto err;
529
530 if (tmp_len <= max_len) {
531 /*
532 * One of the other wNAFs is at least as long as the wNAF
533 * belonging to the generator, so wNAF splitting will not buy
534 * us anything.
535 */
536
537 numblocks = 1;
538 totalnum = num + 1; /* don't use wNAF splitting */
539 wNAF[num] = tmp_wNAF;
540 wNAF[num + 1] = NULL;
541 wNAF_len[num] = tmp_len;
542 /*
543 * pre_comp->points starts with the points that we need here:
544 */
545 val_sub[num] = pre_comp->points;
546 } else {
547 /*
548 * don't include tmp_wNAF directly into wNAF array - use wNAF
549 * splitting and include the blocks
550 */
551
552 signed char *pp;
553 EC_POINT **tmp_points;
554
555 if (tmp_len < numblocks * blocksize) {
556 /*
557 * possibly we can do with fewer blocks than estimated
558 */
559 numblocks = (tmp_len + blocksize - 1) / blocksize;
560 if (numblocks > pre_comp->numblocks) {
561 ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR);
562 OPENSSL_free(tmp_wNAF);
563 goto err;
564 }
565 totalnum = num + numblocks;
566 }
567
568 /* split wNAF in 'numblocks' parts */
569 pp = tmp_wNAF;
570 tmp_points = pre_comp->points;
571
572 for (i = num; i < totalnum; i++) {
573 if (i < totalnum - 1) {
574 wNAF_len[i] = blocksize;
575 if (tmp_len < blocksize) {
576 ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR);
577 OPENSSL_free(tmp_wNAF);
578 goto err;
579 }
580 tmp_len -= blocksize;
581 } else
582 /*
583 * last block gets whatever is left (this could be
584 * more or less than 'blocksize'!)
585 */
586 wNAF_len[i] = tmp_len;
587
588 wNAF[i + 1] = NULL;
589 wNAF[i] = OPENSSL_malloc(wNAF_len[i]);
590 if (wNAF[i] == NULL) {
591 ECerr(EC_F_EC_WNAF_MUL, ERR_R_MALLOC_FAILURE);
592 OPENSSL_free(tmp_wNAF);
593 goto err;
594 }
595 memcpy(wNAF[i], pp, wNAF_len[i]);
596 if (wNAF_len[i] > max_len)
597 max_len = wNAF_len[i];
598
599 if (*tmp_points == NULL) {
600 ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR);
601 OPENSSL_free(tmp_wNAF);
602 goto err;
603 }
604 val_sub[i] = tmp_points;
605 tmp_points += pre_points_per_block;
606 pp += blocksize;
607 }
608 OPENSSL_free(tmp_wNAF);
609 }
610 }
611 }
612
613 /*
614 * All points we precompute now go into a single array 'val'.
615 * 'val_sub[i]' is a pointer to the subarray for the i-th point, or to a
616 * subarray of 'pre_comp->points' if we already have precomputation.
617 */
618 val = OPENSSL_malloc((num_val + 1) * sizeof(val[0]));
619 if (val == NULL) {
620 ECerr(EC_F_EC_WNAF_MUL, ERR_R_MALLOC_FAILURE);
621 goto err;
622 }
623 val[num_val] = NULL; /* pivot element */
624
625 /* allocate points for precomputation */
626 v = val;
627 for (i = 0; i < num + num_scalar; i++) {
628 val_sub[i] = v;
629 for (j = 0; j < ((size_t)1 << (wsize[i] - 1)); j++) {
630 *v = EC_POINT_new(group);
631 if (*v == NULL)
632 goto err;
633 v++;
634 }
635 }
636 if (!(v == val + num_val)) {
637 ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR);
638 goto err;
639 }
640
641 if ((tmp = EC_POINT_new(group)) == NULL)
642 goto err;
643
644 /*-
645 * prepare precomputed values:
646 * val_sub[i][0] := points[i]
647 * val_sub[i][1] := 3 * points[i]
648 * val_sub[i][2] := 5 * points[i]
649 * ...
650 */
651 for (i = 0; i < num + num_scalar; i++) {
652 if (i < num) {
653 if (!EC_POINT_copy(val_sub[i][0], points[i]))
654 goto err;
655 } else {
656 if (!EC_POINT_copy(val_sub[i][0], generator))
657 goto err;
658 }
659
660 if (wsize[i] > 1) {
661 if (!EC_POINT_dbl(group, tmp, val_sub[i][0], ctx))
662 goto err;
663 for (j = 1; j < ((size_t)1 << (wsize[i] - 1)); j++) {
664 if (!EC_POINT_add
665 (group, val_sub[i][j], val_sub[i][j - 1], tmp, ctx))
666 goto err;
667 }
668 }
669 }
670
671 if (!EC_POINTs_make_affine(group, num_val, val, ctx))
672 goto err;
673
674 r_is_at_infinity = 1;
675
676 for (k = max_len - 1; k >= 0; k--) {
677 if (!r_is_at_infinity) {
678 if (!EC_POINT_dbl(group, r, r, ctx))
679 goto err;
680 }
681
682 for (i = 0; i < totalnum; i++) {
683 if (wNAF_len[i] > (size_t)k) {
684 int digit = wNAF[i][k];
685 int is_neg;
686
687 if (digit) {
688 is_neg = digit < 0;
689
690 if (is_neg)
691 digit = -digit;
692
693 if (is_neg != r_is_inverted) {
694 if (!r_is_at_infinity) {
695 if (!EC_POINT_invert(group, r, ctx))
696 goto err;
697 }
698 r_is_inverted = !r_is_inverted;
699 }
700
701 /* digit > 0 */
702
703 if (r_is_at_infinity) {
704 if (!EC_POINT_copy(r, val_sub[i][digit >> 1]))
705 goto err;
706 r_is_at_infinity = 0;
707 } else {
708 if (!EC_POINT_add
709 (group, r, r, val_sub[i][digit >> 1], ctx))
710 goto err;
711 }
712 }
713 }
714 }
715 }
716
717 if (r_is_at_infinity) {
718 if (!EC_POINT_set_to_infinity(group, r))
719 goto err;
720 } else {
721 if (r_is_inverted)
722 if (!EC_POINT_invert(group, r, ctx))
723 goto err;
724 }
725
726 ret = 1;
727
728 err:
729 BN_CTX_free(new_ctx);
730 EC_POINT_free(tmp);
731 OPENSSL_free(wsize);
732 OPENSSL_free(wNAF_len);
733 if (wNAF != NULL) {
734 signed char **w;
735
736 for (w = wNAF; *w != NULL; w++)
737 OPENSSL_free(*w);
738
739 OPENSSL_free(wNAF);
740 }
741 if (val != NULL) {
742 for (v = val; *v != NULL; v++)
743 EC_POINT_clear_free(*v);
744
745 OPENSSL_free(val);
746 }
747 OPENSSL_free(val_sub);
748 return ret;
749 }
750
751 /*-
752 * ec_wNAF_precompute_mult()
753 * creates an EC_PRE_COMP object with preprecomputed multiples of the generator
754 * for use with wNAF splitting as implemented in ec_wNAF_mul().
755 *
756 * 'pre_comp->points' is an array of multiples of the generator
757 * of the following form:
758 * points[0] = generator;
759 * points[1] = 3 * generator;
760 * ...
761 * points[2^(w-1)-1] = (2^(w-1)-1) * generator;
762 * points[2^(w-1)] = 2^blocksize * generator;
763 * points[2^(w-1)+1] = 3 * 2^blocksize * generator;
764 * ...
765 * points[2^(w-1)*(numblocks-1)-1] = (2^(w-1)) * 2^(blocksize*(numblocks-2)) * generator
766 * points[2^(w-1)*(numblocks-1)] = 2^(blocksize*(numblocks-1)) * generator
767 * ...
768 * points[2^(w-1)*numblocks-1] = (2^(w-1)) * 2^(blocksize*(numblocks-1)) * generator
769 * points[2^(w-1)*numblocks] = NULL
770 */
771 int ec_wNAF_precompute_mult(EC_GROUP *group, BN_CTX *ctx)
772 {
773 const EC_POINT *generator;
774 EC_POINT *tmp_point = NULL, *base = NULL, **var;
775 BN_CTX *new_ctx = NULL;
776 const BIGNUM *order;
777 size_t i, bits, w, pre_points_per_block, blocksize, numblocks, num;
778 EC_POINT **points = NULL;
779 EC_PRE_COMP *pre_comp;
780 int ret = 0;
781
782 /* if there is an old EC_PRE_COMP object, throw it away */
783 EC_pre_comp_free(group);
784 if ((pre_comp = ec_pre_comp_new(group)) == NULL)
785 return 0;
786
787 generator = EC_GROUP_get0_generator(group);
788 if (generator == NULL) {
789 ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, EC_R_UNDEFINED_GENERATOR);
790 goto err;
791 }
792
793 if (ctx == NULL) {
794 ctx = new_ctx = BN_CTX_new();
795 if (ctx == NULL)
796 goto err;
797 }
798
799 BN_CTX_start(ctx);
800
801 order = EC_GROUP_get0_order(group);
802 if (order == NULL)
803 goto err;
804 if (BN_is_zero(order)) {
805 ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, EC_R_UNKNOWN_ORDER);
806 goto err;
807 }
808
809 bits = BN_num_bits(order);
810 /*
811 * The following parameters mean we precompute (approximately) one point
812 * per bit. TBD: The combination 8, 4 is perfect for 160 bits; for other
813 * bit lengths, other parameter combinations might provide better
814 * efficiency.
815 */
816 blocksize = 8;
817 w = 4;
818 if (EC_window_bits_for_scalar_size(bits) > w) {
819 /* let's not make the window too small ... */
820 w = EC_window_bits_for_scalar_size(bits);
821 }
822
823 numblocks = (bits + blocksize - 1) / blocksize; /* max. number of blocks
824 * to use for wNAF
825 * splitting */
826
827 pre_points_per_block = (size_t)1 << (w - 1);
828 num = pre_points_per_block * numblocks; /* number of points to compute
829 * and store */
830
831 points = OPENSSL_malloc(sizeof(*points) * (num + 1));
832 if (points == NULL) {
833 ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, ERR_R_MALLOC_FAILURE);
834 goto err;
835 }
836
837 var = points;
838 var[num] = NULL; /* pivot */
839 for (i = 0; i < num; i++) {
840 if ((var[i] = EC_POINT_new(group)) == NULL) {
841 ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, ERR_R_MALLOC_FAILURE);
842 goto err;
843 }
844 }
845
846 if ((tmp_point = EC_POINT_new(group)) == NULL
847 || (base = EC_POINT_new(group)) == NULL) {
848 ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, ERR_R_MALLOC_FAILURE);
849 goto err;
850 }
851
852 if (!EC_POINT_copy(base, generator))
853 goto err;
854
855 /* do the precomputation */
856 for (i = 0; i < numblocks; i++) {
857 size_t j;
858
859 if (!EC_POINT_dbl(group, tmp_point, base, ctx))
860 goto err;
861
862 if (!EC_POINT_copy(*var++, base))
863 goto err;
864
865 for (j = 1; j < pre_points_per_block; j++, var++) {
866 /*
867 * calculate odd multiples of the current base point
868 */
869 if (!EC_POINT_add(group, *var, tmp_point, *(var - 1), ctx))
870 goto err;
871 }
872
873 if (i < numblocks - 1) {
874 /*
875 * get the next base (multiply current one by 2^blocksize)
876 */
877 size_t k;
878
879 if (blocksize <= 2) {
880 ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, ERR_R_INTERNAL_ERROR);
881 goto err;
882 }
883
884 if (!EC_POINT_dbl(group, base, tmp_point, ctx))
885 goto err;
886 for (k = 2; k < blocksize; k++) {
887 if (!EC_POINT_dbl(group, base, base, ctx))
888 goto err;
889 }
890 }
891 }
892
893 if (!EC_POINTs_make_affine(group, num, points, ctx))
894 goto err;
895
896 pre_comp->group = group;
897 pre_comp->blocksize = blocksize;
898 pre_comp->numblocks = numblocks;
899 pre_comp->w = w;
900 pre_comp->points = points;
901 points = NULL;
902 pre_comp->num = num;
903 SETPRECOMP(group, ec, pre_comp);
904 pre_comp = NULL;
905 ret = 1;
906
907 err:
908 if (ctx != NULL)
909 BN_CTX_end(ctx);
910 BN_CTX_free(new_ctx);
911 EC_ec_pre_comp_free(pre_comp);
912 if (points) {
913 EC_POINT **p;
914
915 for (p = points; *p != NULL; p++)
916 EC_POINT_free(*p);
917 OPENSSL_free(points);
918 }
919 EC_POINT_free(tmp_point);
920 EC_POINT_free(base);
921 return ret;
922 }
923
924 int ec_wNAF_have_precompute_mult(const EC_GROUP *group)
925 {
926 return HAVEPRECOMP(group, ec);
927 }