xref: /openbsd-src/lib/libcrypto/ec/ec_mult.c (revision 50b7afb2c2c0993b0894d4e34bf857cb13ed9c80)
1 /* $OpenBSD: ec_mult.c,v 1.14 2014/07/12 16:03:37 miod Exp $ */
2 /*
3  * Originally written by Bodo Moeller and Nils Larsch for the OpenSSL project.
4  */
5 /* ====================================================================
6  * Copyright (c) 1998-2007 The OpenSSL Project.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  *
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in
17  *    the documentation and/or other materials provided with the
18  *    distribution.
19  *
20  * 3. All advertising materials mentioning features or use of this
21  *    software must display the following acknowledgment:
22  *    "This product includes software developed by the OpenSSL Project
23  *    for use in the OpenSSL Toolkit. (http://www.openssl.org/)"
24  *
25  * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
26  *    endorse or promote products derived from this software without
27  *    prior written permission. For written permission, please contact
28  *    openssl-core@openssl.org.
29  *
30  * 5. Products derived from this software may not be called "OpenSSL"
31  *    nor may "OpenSSL" appear in their names without prior written
32  *    permission of the OpenSSL Project.
33  *
34  * 6. Redistributions of any form whatsoever must retain the following
35  *    acknowledgment:
36  *    "This product includes software developed by the OpenSSL Project
37  *    for use in the OpenSSL Toolkit (http://www.openssl.org/)"
38  *
39  * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
40  * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
41  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
42  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE OpenSSL PROJECT OR
43  * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
44  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
45  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
46  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
47  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
48  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
49  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
50  * OF THE POSSIBILITY OF SUCH DAMAGE.
51  * ====================================================================
52  *
53  * This product includes cryptographic software written by Eric Young
54  * (eay@cryptsoft.com).  This product includes software written by Tim
55  * Hudson (tjh@cryptsoft.com).
56  *
57  */
58 /* ====================================================================
59  * Copyright 2002 Sun Microsystems, Inc. ALL RIGHTS RESERVED.
60  * Portions of this software developed by SUN MICROSYSTEMS, INC.,
61  * and contributed to the OpenSSL project.
62  */
63 
64 #include <string.h>
65 
66 #include <openssl/err.h>
67 
68 #include "ec_lcl.h"
69 
70 
71 /*
72  * This file implements the wNAF-based interleaving multi-exponentation method
73  * (<URL:http://www.informatik.tu-darmstadt.de/TI/Mitarbeiter/moeller.html#multiexp>);
74  * for multiplication with precomputation, we use wNAF splitting
75  * (<URL:http://www.informatik.tu-darmstadt.de/TI/Mitarbeiter/moeller.html#fastexp>).
76  */
77 
78 
79 
80 
81 /* structure for precomputed multiples of the generator */
82 typedef struct ec_pre_comp_st {
83 	const EC_GROUP *group;	/* parent EC_GROUP object */
84 	size_t blocksize;	/* block size for wNAF splitting */
85 	size_t numblocks;	/* max. number of blocks for which we have
86 				 * precomputation */
87 	size_t w;		/* window size */
88 	EC_POINT **points;	/* array with pre-calculated multiples of
89 				 * generator: 'num' pointers to EC_POINT
90 				 * objects followed by a NULL */
91 	size_t num;		/* numblocks * 2^(w-1) */
92 	int references;
93 } EC_PRE_COMP;
94 
95 /* functions to manage EC_PRE_COMP within the EC_GROUP extra_data framework */
96 static void *ec_pre_comp_dup(void *);
97 static void ec_pre_comp_free(void *);
98 static void ec_pre_comp_clear_free(void *);
99 
100 static EC_PRE_COMP *
101 ec_pre_comp_new(const EC_GROUP * group)
102 {
103 	EC_PRE_COMP *ret = NULL;
104 
105 	if (!group)
106 		return NULL;
107 
108 	ret = malloc(sizeof(EC_PRE_COMP));
109 	if (!ret) {
110 		ECerr(EC_F_EC_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
111 		return ret;
112 	}
113 	ret->group = group;
114 	ret->blocksize = 8;	/* default */
115 	ret->numblocks = 0;
116 	ret->w = 4;		/* default */
117 	ret->points = NULL;
118 	ret->num = 0;
119 	ret->references = 1;
120 	return ret;
121 }
122 
123 static void *
124 ec_pre_comp_dup(void *src_)
125 {
126 	EC_PRE_COMP *src = src_;
127 
128 	/* no need to actually copy, these objects never change! */
129 
130 	CRYPTO_add(&src->references, 1, CRYPTO_LOCK_EC_PRE_COMP);
131 
132 	return src_;
133 }
134 
135 static void
136 ec_pre_comp_free(void *pre_)
137 {
138 	int i;
139 	EC_PRE_COMP *pre = pre_;
140 
141 	if (!pre)
142 		return;
143 
144 	i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP);
145 	if (i > 0)
146 		return;
147 
148 	if (pre->points) {
149 		EC_POINT **p;
150 
151 		for (p = pre->points; *p != NULL; p++)
152 			EC_POINT_free(*p);
153 		free(pre->points);
154 	}
155 	free(pre);
156 }
157 
158 static void
159 ec_pre_comp_clear_free(void *pre_)
160 {
161 	int i;
162 	EC_PRE_COMP *pre = pre_;
163 
164 	if (!pre)
165 		return;
166 
167 	i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP);
168 	if (i > 0)
169 		return;
170 
171 	if (pre->points) {
172 		EC_POINT **p;
173 
174 		for (p = pre->points; *p != NULL; p++) {
175 			EC_POINT_clear_free(*p);
176 			OPENSSL_cleanse(p, sizeof *p);
177 		}
178 		free(pre->points);
179 	}
180 	OPENSSL_cleanse(pre, sizeof *pre);
181 	free(pre);
182 }
183 
184 
185 
186 
187 /* Determine the modified width-(w+1) Non-Adjacent Form (wNAF) of 'scalar'.
188  * This is an array  r[]  of values that are either zero or odd with an
189  * absolute value less than  2^w  satisfying
190  *     scalar = \sum_j r[j]*2^j
191  * where at most one of any  w+1  consecutive digits is non-zero
192  * with the exception that the most significant digit may be only
193  * w-1 zeros away from that next non-zero digit.
194  */
195 static signed char *
196 compute_wNAF(const BIGNUM * scalar, int w, size_t * ret_len)
197 {
198 	int window_val;
199 	int ok = 0;
200 	signed char *r = NULL;
201 	int sign = 1;
202 	int bit, next_bit, mask;
203 	size_t len = 0, j;
204 
205 	if (BN_is_zero(scalar)) {
206 		r = malloc(1);
207 		if (!r) {
208 			ECerr(EC_F_COMPUTE_WNAF, ERR_R_MALLOC_FAILURE);
209 			goto err;
210 		}
211 		r[0] = 0;
212 		*ret_len = 1;
213 		return r;
214 	}
215 	if (w <= 0 || w > 7) {
216 		/* 'signed char' can represent integers with
217 		 * absolute values less than 2^7 */
218 		ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR);
219 		goto err;
220 	}
221 	bit = 1 << w;		/* at most 128 */
222 	next_bit = bit << 1;	/* at most 256 */
223 	mask = next_bit - 1;	/* at most 255 */
224 
225 	if (BN_is_negative(scalar)) {
226 		sign = -1;
227 	}
228 	if (scalar->d == NULL || scalar->top == 0) {
229 		ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR);
230 		goto err;
231 	}
232 	len = BN_num_bits(scalar);
233 	r = malloc(len + 1);	/* modified wNAF may be one digit longer than
234 				 * binary representation (*ret_len will be
235 				 * set to the actual length, i.e. at most
236 				 * BN_num_bits(scalar) + 1) */
237 	if (r == NULL) {
238 		ECerr(EC_F_COMPUTE_WNAF, ERR_R_MALLOC_FAILURE);
239 		goto err;
240 	}
241 	window_val = scalar->d[0] & mask;
242 	j = 0;
243 	while ((window_val != 0) || (j + w + 1 < len)) {
244 		/* if j+w+1 >= len, window_val will not increase */
245 		int digit = 0;
246 
247 		/* 0 <= window_val <= 2^(w+1) */
248 		if (window_val & 1) {
249 			/* 0 < window_val < 2^(w+1) */
250 			if (window_val & bit) {
251 				digit = window_val - next_bit;	/* -2^w < digit < 0 */
252 
253 #if 1				/* modified wNAF */
254 				if (j + w + 1 >= len) {
255 					/*
256 					 * special case for generating
257 					 * modified wNAFs: no new bits will
258 					 * be added into window_val, so using
259 					 * a positive digit here will
260 					 * decrease the total length of the
261 					 * representation
262 					 */
263 
264 					digit = window_val & (mask >> 1);	/* 0 < digit < 2^w */
265 				}
266 #endif
267 			} else {
268 				digit = window_val;	/* 0 < digit < 2^w */
269 			}
270 
271 			if (digit <= -bit || digit >= bit || !(digit & 1)) {
272 				ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR);
273 				goto err;
274 			}
275 			window_val -= digit;
276 
277 			/*
278 			 * now window_val is 0 or 2^(w+1) in standard wNAF
279 			 * generation; for modified window NAFs, it may also
280 			 * be 2^w
281 			 */
282 			if (window_val != 0 && window_val != next_bit && window_val != bit) {
283 				ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR);
284 				goto err;
285 			}
286 		}
287 		r[j++] = sign * digit;
288 
289 		window_val >>= 1;
290 		window_val += bit * BN_is_bit_set(scalar, j + w);
291 
292 		if (window_val > next_bit) {
293 			ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR);
294 			goto err;
295 		}
296 	}
297 
298 	if (j > len + 1) {
299 		ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR);
300 		goto err;
301 	}
302 	len = j;
303 	ok = 1;
304 
305 err:
306 	if (!ok) {
307 		free(r);
308 		r = NULL;
309 	}
310 	if (ok)
311 		*ret_len = len;
312 	return r;
313 }
314 
315 
316 /* TODO: table should be optimised for the wNAF-based implementation,
317  *       sometimes smaller windows will give better performance
318  *       (thus the boundaries should be increased)
319  */
320 #define EC_window_bits_for_scalar_size(b) \
321 		((size_t) \
322 		 ((b) >= 2000 ? 6 : \
323 		  (b) >=  800 ? 5 : \
324 		  (b) >=  300 ? 4 : \
325 		  (b) >=   70 ? 3 : \
326 		  (b) >=   20 ? 2 : \
327 		  1))
328 
329 /* Compute
330  *      \sum scalars[i]*points[i],
331  * also including
332  *      scalar*generator
333  * in the addition if scalar != NULL
334  */
335 int
336 ec_wNAF_mul(const EC_GROUP * group, EC_POINT * r, const BIGNUM * scalar,
337     size_t num, const EC_POINT * points[], const BIGNUM * scalars[], BN_CTX * ctx)
338 {
339 	BN_CTX *new_ctx = NULL;
340 	const EC_POINT *generator = NULL;
341 	EC_POINT *tmp = NULL;
342 	size_t totalnum;
343 	size_t blocksize = 0, numblocks = 0;	/* for wNAF splitting */
344 	size_t pre_points_per_block = 0;
345 	size_t i, j;
346 	int k;
347 	int r_is_inverted = 0;
348 	int r_is_at_infinity = 1;
349 	size_t *wsize = NULL;	/* individual window sizes */
350 	signed char **wNAF = NULL;	/* individual wNAFs */
351 	size_t *wNAF_len = NULL;
352 	size_t max_len = 0;
353 	size_t num_val;
354 	EC_POINT **val = NULL;	/* precomputation */
355 	EC_POINT **v;
356 	EC_POINT ***val_sub = NULL;	/* pointers to sub-arrays of 'val' or
357 					 * 'pre_comp->points' */
358 	const EC_PRE_COMP *pre_comp = NULL;
359 	int num_scalar = 0;	/* flag: will be set to 1 if 'scalar' must be
360 				 * treated like other scalars, i.e.
361 				 * precomputation is not available */
362 	int ret = 0;
363 
364 	if (group->meth != r->meth) {
365 		ECerr(EC_F_EC_WNAF_MUL, EC_R_INCOMPATIBLE_OBJECTS);
366 		return 0;
367 	}
368 	if ((scalar == NULL) && (num == 0)) {
369 		return EC_POINT_set_to_infinity(group, r);
370 	}
371 	for (i = 0; i < num; i++) {
372 		if (group->meth != points[i]->meth) {
373 			ECerr(EC_F_EC_WNAF_MUL, EC_R_INCOMPATIBLE_OBJECTS);
374 			return 0;
375 		}
376 	}
377 
378 	if (ctx == NULL) {
379 		ctx = new_ctx = BN_CTX_new();
380 		if (ctx == NULL)
381 			goto err;
382 	}
383 	if (scalar != NULL) {
384 		generator = EC_GROUP_get0_generator(group);
385 		if (generator == NULL) {
386 			ECerr(EC_F_EC_WNAF_MUL, EC_R_UNDEFINED_GENERATOR);
387 			goto err;
388 		}
389 		/* look if we can use precomputed multiples of generator */
390 
391 		pre_comp = EC_EX_DATA_get_data(group->extra_data, ec_pre_comp_dup, ec_pre_comp_free, ec_pre_comp_clear_free);
392 
393 		if (pre_comp && pre_comp->numblocks &&
394 		    (EC_POINT_cmp(group, generator, pre_comp->points[0], ctx) == 0)) {
395 			blocksize = pre_comp->blocksize;
396 
397 			/*
398 			 * determine maximum number of blocks that wNAF
399 			 * splitting may yield (NB: maximum wNAF length is
400 			 * bit length plus one)
401 			 */
402 			numblocks = (BN_num_bits(scalar) / blocksize) + 1;
403 
404 			/*
405 			 * we cannot use more blocks than we have
406 			 * precomputation for
407 			 */
408 			if (numblocks > pre_comp->numblocks)
409 				numblocks = pre_comp->numblocks;
410 
411 			pre_points_per_block = (size_t) 1 << (pre_comp->w - 1);
412 
413 			/* check that pre_comp looks sane */
414 			if (pre_comp->num != (pre_comp->numblocks * pre_points_per_block)) {
415 				ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR);
416 				goto err;
417 			}
418 		} else {
419 			/* can't use precomputation */
420 			pre_comp = NULL;
421 			numblocks = 1;
422 			num_scalar = 1;	/* treat 'scalar' like 'num'-th
423 					 * element of 'scalars' */
424 		}
425 	}
426 	totalnum = num + numblocks;
427 
428 	wsize = reallocarray(NULL, totalnum, sizeof wsize[0]);
429 	wNAF_len = reallocarray(NULL, totalnum, sizeof wNAF_len[0]);
430 	/* includes space for pivot */
431 	wNAF = reallocarray(NULL, (totalnum + 1), sizeof wNAF[0]);
432 	val_sub = reallocarray(NULL, totalnum, sizeof val_sub[0]);
433 
434 	if (!wsize || !wNAF_len || !wNAF || !val_sub) {
435 		ECerr(EC_F_EC_WNAF_MUL, ERR_R_MALLOC_FAILURE);
436 		goto err;
437 	}
438 	wNAF[0] = NULL;		/* preliminary pivot */
439 
440 	/* num_val will be the total number of temporarily precomputed points */
441 	num_val = 0;
442 
443 	for (i = 0; i < num + num_scalar; i++) {
444 		size_t bits;
445 
446 		bits = i < num ? BN_num_bits(scalars[i]) : BN_num_bits(scalar);
447 		wsize[i] = EC_window_bits_for_scalar_size(bits);
448 		num_val += (size_t) 1 << (wsize[i] - 1);
449 		wNAF[i + 1] = NULL;	/* make sure we always have a pivot */
450 		wNAF[i] = compute_wNAF((i < num ? scalars[i] : scalar), wsize[i], &wNAF_len[i]);
451 		if (wNAF[i] == NULL)
452 			goto err;
453 		if (wNAF_len[i] > max_len)
454 			max_len = wNAF_len[i];
455 	}
456 
457 	if (numblocks) {
458 		/* we go here iff scalar != NULL */
459 
460 		if (pre_comp == NULL) {
461 			if (num_scalar != 1) {
462 				ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR);
463 				goto err;
464 			}
465 			/* we have already generated a wNAF for 'scalar' */
466 		} else {
467 			signed char *tmp_wNAF = NULL;
468 			size_t tmp_len = 0;
469 
470 			if (num_scalar != 0) {
471 				ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR);
472 				goto err;
473 			}
474 			/*
475 			 * use the window size for which we have
476 			 * precomputation
477 			 */
478 			wsize[num] = pre_comp->w;
479 			tmp_wNAF = compute_wNAF(scalar, wsize[num], &tmp_len);
480 			if (!tmp_wNAF)
481 				goto err;
482 
483 			if (tmp_len <= max_len) {
484 				/*
485 				 * One of the other wNAFs is at least as long
486 				 * as the wNAF belonging to the generator, so
487 				 * wNAF splitting will not buy us anything.
488 				 */
489 
490 				numblocks = 1;
491 				totalnum = num + 1;	/* don't use wNAF
492 							 * splitting */
493 				wNAF[num] = tmp_wNAF;
494 				wNAF[num + 1] = NULL;
495 				wNAF_len[num] = tmp_len;
496 				if (tmp_len > max_len)
497 					max_len = tmp_len;
498 				/*
499 				 * pre_comp->points starts with the points
500 				 * that we need here:
501 				 */
502 				val_sub[num] = pre_comp->points;
503 			} else {
504 				/*
505 				 * don't include tmp_wNAF directly into wNAF
506 				 * array - use wNAF splitting and include the
507 				 * blocks
508 				 */
509 
510 				signed char *pp;
511 				EC_POINT **tmp_points;
512 
513 				if (tmp_len < numblocks * blocksize) {
514 					/*
515 					 * possibly we can do with fewer
516 					 * blocks than estimated
517 					 */
518 					numblocks = (tmp_len + blocksize - 1) / blocksize;
519 					if (numblocks > pre_comp->numblocks) {
520 						ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR);
521 						goto err;
522 					}
523 					totalnum = num + numblocks;
524 				}
525 				/* split wNAF in 'numblocks' parts */
526 				pp = tmp_wNAF;
527 				tmp_points = pre_comp->points;
528 
529 				for (i = num; i < totalnum; i++) {
530 					if (i < totalnum - 1) {
531 						wNAF_len[i] = blocksize;
532 						if (tmp_len < blocksize) {
533 							ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR);
534 							goto err;
535 						}
536 						tmp_len -= blocksize;
537 					} else
538 						/*
539 						 * last block gets whatever
540 						 * is left (this could be
541 						 * more or less than
542 						 * 'blocksize'!)
543 						 */
544 						wNAF_len[i] = tmp_len;
545 
546 					wNAF[i + 1] = NULL;
547 					wNAF[i] = malloc(wNAF_len[i]);
548 					if (wNAF[i] == NULL) {
549 						ECerr(EC_F_EC_WNAF_MUL, ERR_R_MALLOC_FAILURE);
550 						free(tmp_wNAF);
551 						goto err;
552 					}
553 					memcpy(wNAF[i], pp, wNAF_len[i]);
554 					if (wNAF_len[i] > max_len)
555 						max_len = wNAF_len[i];
556 
557 					if (*tmp_points == NULL) {
558 						ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR);
559 						free(tmp_wNAF);
560 						goto err;
561 					}
562 					val_sub[i] = tmp_points;
563 					tmp_points += pre_points_per_block;
564 					pp += blocksize;
565 				}
566 				free(tmp_wNAF);
567 			}
568 		}
569 	}
570 	/*
571 	 * All points we precompute now go into a single array 'val'.
572 	 * 'val_sub[i]' is a pointer to the subarray for the i-th point, or
573 	 * to a subarray of 'pre_comp->points' if we already have
574 	 * precomputation.
575 	 */
576 	val = reallocarray(NULL, (num_val + 1), sizeof val[0]);
577 	if (val == NULL) {
578 		ECerr(EC_F_EC_WNAF_MUL, ERR_R_MALLOC_FAILURE);
579 		goto err;
580 	}
581 	val[num_val] = NULL;	/* pivot element */
582 
583 	/* allocate points for precomputation */
584 	v = val;
585 	for (i = 0; i < num + num_scalar; i++) {
586 		val_sub[i] = v;
587 		for (j = 0; j < ((size_t) 1 << (wsize[i] - 1)); j++) {
588 			*v = EC_POINT_new(group);
589 			if (*v == NULL)
590 				goto err;
591 			v++;
592 		}
593 	}
594 	if (!(v == val + num_val)) {
595 		ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR);
596 		goto err;
597 	}
598 	if (!(tmp = EC_POINT_new(group)))
599 		goto err;
600 
601 	/*
602 	 * prepare precomputed values: val_sub[i][0] :=     points[i]
603 	 * val_sub[i][1] := 3 * points[i] val_sub[i][2] := 5 * points[i] ...
604 	 */
605 	for (i = 0; i < num + num_scalar; i++) {
606 		if (i < num) {
607 			if (!EC_POINT_copy(val_sub[i][0], points[i]))
608 				goto err;
609 		} else {
610 			if (!EC_POINT_copy(val_sub[i][0], generator))
611 				goto err;
612 		}
613 
614 		if (wsize[i] > 1) {
615 			if (!EC_POINT_dbl(group, tmp, val_sub[i][0], ctx))
616 				goto err;
617 			for (j = 1; j < ((size_t) 1 << (wsize[i] - 1)); j++) {
618 				if (!EC_POINT_add(group, val_sub[i][j], val_sub[i][j - 1], tmp, ctx))
619 					goto err;
620 			}
621 		}
622 	}
623 
624 #if 1				/* optional; EC_window_bits_for_scalar_size
625 				 * assumes we do this step */
626 	if (!EC_POINTs_make_affine(group, num_val, val, ctx))
627 		goto err;
628 #endif
629 
630 	r_is_at_infinity = 1;
631 
632 	for (k = max_len - 1; k >= 0; k--) {
633 		if (!r_is_at_infinity) {
634 			if (!EC_POINT_dbl(group, r, r, ctx))
635 				goto err;
636 		}
637 		for (i = 0; i < totalnum; i++) {
638 			if (wNAF_len[i] > (size_t) k) {
639 				int digit = wNAF[i][k];
640 				int is_neg;
641 
642 				if (digit) {
643 					is_neg = digit < 0;
644 
645 					if (is_neg)
646 						digit = -digit;
647 
648 					if (is_neg != r_is_inverted) {
649 						if (!r_is_at_infinity) {
650 							if (!EC_POINT_invert(group, r, ctx))
651 								goto err;
652 						}
653 						r_is_inverted = !r_is_inverted;
654 					}
655 					/* digit > 0 */
656 
657 					if (r_is_at_infinity) {
658 						if (!EC_POINT_copy(r, val_sub[i][digit >> 1]))
659 							goto err;
660 						r_is_at_infinity = 0;
661 					} else {
662 						if (!EC_POINT_add(group, r, r, val_sub[i][digit >> 1], ctx))
663 							goto err;
664 					}
665 				}
666 			}
667 		}
668 	}
669 
670 	if (r_is_at_infinity) {
671 		if (!EC_POINT_set_to_infinity(group, r))
672 			goto err;
673 	} else {
674 		if (r_is_inverted)
675 			if (!EC_POINT_invert(group, r, ctx))
676 				goto err;
677 	}
678 
679 	ret = 1;
680 
681 err:
682 	BN_CTX_free(new_ctx);
683 	EC_POINT_free(tmp);
684 	free(wsize);
685 	free(wNAF_len);
686 	if (wNAF != NULL) {
687 		signed char **w;
688 
689 		for (w = wNAF; *w != NULL; w++)
690 			free(*w);
691 
692 		free(wNAF);
693 	}
694 	if (val != NULL) {
695 		for (v = val; *v != NULL; v++)
696 			EC_POINT_clear_free(*v);
697 		free(val);
698 	}
699 	free(val_sub);
700 	return ret;
701 }
702 
703 
704 /* ec_wNAF_precompute_mult()
705  * creates an EC_PRE_COMP object with preprecomputed multiples of the generator
706  * for use with wNAF splitting as implemented in ec_wNAF_mul().
707  *
708  * 'pre_comp->points' is an array of multiples of the generator
709  * of the following form:
710  * points[0] =     generator;
711  * points[1] = 3 * generator;
712  * ...
713  * points[2^(w-1)-1] =     (2^(w-1)-1) * generator;
714  * points[2^(w-1)]   =     2^blocksize * generator;
715  * points[2^(w-1)+1] = 3 * 2^blocksize * generator;
716  * ...
717  * points[2^(w-1)*(numblocks-1)-1] = (2^(w-1)) *  2^(blocksize*(numblocks-2)) * generator
718  * points[2^(w-1)*(numblocks-1)]   =              2^(blocksize*(numblocks-1)) * generator
719  * ...
720  * points[2^(w-1)*numblocks-1]     = (2^(w-1)) *  2^(blocksize*(numblocks-1)) * generator
721  * points[2^(w-1)*numblocks]       = NULL
722  */
723 int
724 ec_wNAF_precompute_mult(EC_GROUP * group, BN_CTX * ctx)
725 {
726 	const EC_POINT *generator;
727 	EC_POINT *tmp_point = NULL, *base = NULL, **var;
728 	BN_CTX *new_ctx = NULL;
729 	BIGNUM *order;
730 	size_t i, bits, w, pre_points_per_block, blocksize, numblocks,
731 	 num;
732 	EC_POINT **points = NULL;
733 	EC_PRE_COMP *pre_comp;
734 	int ret = 0;
735 
736 	/* if there is an old EC_PRE_COMP object, throw it away */
737 	EC_EX_DATA_free_data(&group->extra_data, ec_pre_comp_dup, ec_pre_comp_free, ec_pre_comp_clear_free);
738 
739 	if ((pre_comp = ec_pre_comp_new(group)) == NULL)
740 		return 0;
741 
742 	generator = EC_GROUP_get0_generator(group);
743 	if (generator == NULL) {
744 		ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, EC_R_UNDEFINED_GENERATOR);
745 		goto err;
746 	}
747 	if (ctx == NULL) {
748 		ctx = new_ctx = BN_CTX_new();
749 		if (ctx == NULL)
750 			goto err;
751 	}
752 	BN_CTX_start(ctx);
753 	order = BN_CTX_get(ctx);
754 	if (order == NULL)
755 		goto err;
756 
757 	if (!EC_GROUP_get_order(group, order, ctx))
758 		goto err;
759 	if (BN_is_zero(order)) {
760 		ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, EC_R_UNKNOWN_ORDER);
761 		goto err;
762 	}
763 	bits = BN_num_bits(order);
764 	/*
765 	 * The following parameters mean we precompute (approximately) one
766 	 * point per bit.
767 	 *
768 	 * TBD: The combination  8, 4  is perfect for 160 bits; for other bit
769 	 * lengths, other parameter combinations might provide better
770 	 * efficiency.
771 	 */
772 	blocksize = 8;
773 	w = 4;
774 	if (EC_window_bits_for_scalar_size(bits) > w) {
775 		/* let's not make the window too small ... */
776 		w = EC_window_bits_for_scalar_size(bits);
777 	}
778 	numblocks = (bits + blocksize - 1) / blocksize;	/* max. number of blocks
779 							 * to use for wNAF
780 							 * splitting */
781 
782 	pre_points_per_block = (size_t) 1 << (w - 1);
783 	num = pre_points_per_block * numblocks;	/* number of points to
784 						 * compute and store */
785 
786 	points = reallocarray(NULL, (num + 1), sizeof(EC_POINT *));
787 	if (!points) {
788 		ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, ERR_R_MALLOC_FAILURE);
789 		goto err;
790 	}
791 	var = points;
792 	var[num] = NULL;	/* pivot */
793 	for (i = 0; i < num; i++) {
794 		if ((var[i] = EC_POINT_new(group)) == NULL) {
795 			ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, ERR_R_MALLOC_FAILURE);
796 			goto err;
797 		}
798 	}
799 
800 	if (!(tmp_point = EC_POINT_new(group)) || !(base = EC_POINT_new(group))) {
801 		ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, ERR_R_MALLOC_FAILURE);
802 		goto err;
803 	}
804 	if (!EC_POINT_copy(base, generator))
805 		goto err;
806 
807 	/* do the precomputation */
808 	for (i = 0; i < numblocks; i++) {
809 		size_t j;
810 
811 		if (!EC_POINT_dbl(group, tmp_point, base, ctx))
812 			goto err;
813 
814 		if (!EC_POINT_copy(*var++, base))
815 			goto err;
816 
817 		for (j = 1; j < pre_points_per_block; j++, var++) {
818 			/* calculate odd multiples of the current base point */
819 			if (!EC_POINT_add(group, *var, tmp_point, *(var - 1), ctx))
820 				goto err;
821 		}
822 
823 		if (i < numblocks - 1) {
824 			/*
825 			 * get the next base (multiply current one by
826 			 * 2^blocksize)
827 			 */
828 			size_t k;
829 
830 			if (blocksize <= 2) {
831 				ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, ERR_R_INTERNAL_ERROR);
832 				goto err;
833 			}
834 			if (!EC_POINT_dbl(group, base, tmp_point, ctx))
835 				goto err;
836 			for (k = 2; k < blocksize; k++) {
837 				if (!EC_POINT_dbl(group, base, base, ctx))
838 					goto err;
839 			}
840 		}
841 	}
842 
843 	if (!EC_POINTs_make_affine(group, num, points, ctx))
844 		goto err;
845 
846 	pre_comp->group = group;
847 	pre_comp->blocksize = blocksize;
848 	pre_comp->numblocks = numblocks;
849 	pre_comp->w = w;
850 	pre_comp->points = points;
851 	points = NULL;
852 	pre_comp->num = num;
853 
854 	if (!EC_EX_DATA_set_data(&group->extra_data, pre_comp,
855 		ec_pre_comp_dup, ec_pre_comp_free, ec_pre_comp_clear_free))
856 		goto err;
857 	pre_comp = NULL;
858 
859 	ret = 1;
860 err:
861 	if (ctx != NULL)
862 		BN_CTX_end(ctx);
863 	BN_CTX_free(new_ctx);
864 	ec_pre_comp_free(pre_comp);
865 	if (points) {
866 		EC_POINT **p;
867 
868 		for (p = points; *p != NULL; p++)
869 			EC_POINT_free(*p);
870 		free(points);
871 	}
872 	EC_POINT_free(tmp_point);
873 	EC_POINT_free(base);
874 	return ret;
875 }
876 
877 
878 int
879 ec_wNAF_have_precompute_mult(const EC_GROUP * group)
880 {
881 	if (EC_EX_DATA_get_data(group->extra_data, ec_pre_comp_dup, ec_pre_comp_free, ec_pre_comp_clear_free) != NULL)
882 		return 1;
883 	else
884 		return 0;
885 }
886