1 /* crypto/ec/ec_mult.c */
2 /* ====================================================================
3 * Copyright (c) 1998-2001 The OpenSSL Project. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in
14 * the documentation and/or other materials provided with the
17 * 3. All advertising materials mentioning features or use of this
18 * software must display the following acknowledgment:
19 * "This product includes software developed by the OpenSSL Project
20 * for use in the OpenSSL Toolkit. (http://www.openssl.org/)"
22 * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
23 * endorse or promote products derived from this software without
24 * prior written permission. For written permission, please contact
25 * openssl-core@openssl.org.
27 * 5. Products derived from this software may not be called "OpenSSL"
28 * nor may "OpenSSL" appear in their names without prior written
29 * permission of the OpenSSL Project.
31 * 6. Redistributions of any form whatsoever must retain the following
33 * "This product includes software developed by the OpenSSL Project
34 * for use in the OpenSSL Toolkit (http://www.openssl.org/)"
36 * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
37 * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
38 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
39 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE OpenSSL PROJECT OR
40 * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
41 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
42 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
43 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
45 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
46 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
47 * OF THE POSSIBILITY OF SUCH DAMAGE.
48 * ====================================================================
50 * This product includes cryptographic software written by Eric Young
51 * (eay@cryptsoft.com). This product includes software written by Tim
52 * Hudson (tjh@cryptsoft.com).
56 #include <openssl/err.h>
61 /* TODO: optional precomputation of multiples of the generator */
66 * wNAF-based interleaving multi-exponentation method
67 * (<URL:http://www.informatik.tu-darmstadt.de/TI/Mitarbeiter/moeller.html#multiexp>)
72 /* Determine the width-(w+1) Non-Adjacent Form (wNAF) of 'scalar'.
73 * This is an array r[] of values that are either zero or odd with an
74 * absolute value less than 2^w satisfying
75 * scalar = \sum_j r[j]*2^j
76 * where at most one of any w+1 consecutive digits is non-zero.
78 static signed char *compute_wNAF(const BIGNUM *scalar, int w, size_t *ret_len, BN_CTX *ctx)
82 signed char *r = NULL;
84 int bit, next_bit, mask;
89 if (c == NULL) goto err;
91 if (w <= 0 || w > 7) /* 'signed char' can represent integers with absolute values less than 2^7 */
93 ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR);
96 bit = 1 << w; /* at most 128 */
97 next_bit = bit << 1; /* at most 256 */
98 mask = next_bit - 1; /* at most 255 */
100 if (!BN_copy(c, scalar)) goto err;
107 len = BN_num_bits(c) + 1; /* wNAF may be one digit longer than binary representation */
108 r = OPENSSL_malloc(len);
109 if (r == NULL) goto err;
112 while (!BN_is_zero(c))
118 if (c->d == NULL || c->top == 0)
120 ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR);
128 if (!BN_add_word(c, -u)) goto err;
133 if (!BN_sub_word(c, u)) goto err;
136 if (u <= -bit || u >= bit || !(u & 1) || c->neg)
138 ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR);
147 ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR);
150 if (!BN_rshift1(c, c)) goto err;
155 ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR);
174 /* TODO: table should be optimised for the wNAF-based implementation */
175 #define EC_window_bits_for_scalar_size(b) \
184 * \sum scalars[i]*points[i],
187 * in the addition if scalar != NULL
189 int EC_POINTs_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar,
190 size_t num, const EC_POINT *points[], const BIGNUM *scalars[], BN_CTX *ctx)
192 BN_CTX *new_ctx = NULL;
193 EC_POINT *generator = NULL;
194 EC_POINT *tmp = NULL;
198 int r_is_inverted = 0;
199 int r_is_at_infinity = 1;
200 size_t *wsize = NULL; /* individual window sizes */
201 signed char **wNAF = NULL; /* individual wNAFs */
202 size_t *wNAF_len = NULL;
205 EC_POINT **val = NULL; /* precomputation */
207 EC_POINT ***val_sub = NULL; /* pointers to sub-arrays of 'val' */
212 generator = EC_GROUP_get0_generator(group);
213 if (generator == NULL)
215 ECerr(EC_F_EC_POINTS_MUL, EC_R_UNDEFINED_GENERATOR);
220 for (i = 0; i < num; i++)
222 if (group->meth != points[i]->meth)
224 ECerr(EC_F_EC_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS);
229 totalnum = num + (scalar != NULL);
231 wsize = OPENSSL_malloc(totalnum * sizeof wsize[0]);
232 wNAF_len = OPENSSL_malloc(totalnum * sizeof wNAF_len[0]);
233 wNAF = OPENSSL_malloc(totalnum * sizeof wNAF[0] + 1);
236 wNAF[0] = NULL; /* preliminary pivot */
238 if (wsize == NULL || wNAF_len == NULL || wNAF == NULL) goto err;
240 /* num_val := total number of points to precompute */
242 for (i = 0; i < totalnum; i++)
246 bits = i < num ? BN_num_bits(scalars[i]) : BN_num_bits(scalar);
247 wsize[i] = EC_window_bits_for_scalar_size(bits);
248 num_val += 1u << (wsize[i] - 1);
251 /* all precomputed points go into a single array 'val',
252 * 'val_sub[i]' is a pointer to the subarray for the i-th point */
253 val = OPENSSL_malloc((num_val + 1) * sizeof val[0]);
254 if (val == NULL) goto err;
255 val[num_val] = NULL; /* pivot element */
257 val_sub = OPENSSL_malloc(totalnum * sizeof val_sub[0]);
258 if (val_sub == NULL) goto err;
260 /* allocate points for precomputation */
262 for (i = 0; i < totalnum; i++)
265 for (j = 0; j < (1u << (wsize[i] - 1)); j++)
267 *v = EC_POINT_new(group);
268 if (*v == NULL) goto err;
272 if (!(v == val + num_val))
274 ECerr(EC_F_EC_POINTS_MUL, ERR_R_INTERNAL_ERROR);
280 ctx = new_ctx = BN_CTX_new();
285 tmp = EC_POINT_new(group);
286 if (tmp == NULL) goto err;
288 /* prepare precomputed values:
289 * val_sub[i][0] := points[i]
290 * val_sub[i][1] := 3 * points[i]
291 * val_sub[i][2] := 5 * points[i]
294 for (i = 0; i < totalnum; i++)
298 if (!EC_POINT_copy(val_sub[i][0], points[i])) goto err;
302 if (!EC_POINT_copy(val_sub[i][0], generator)) goto err;
307 if (!EC_POINT_dbl(group, tmp, val_sub[i][0], ctx)) goto err;
308 for (j = 1; j < (1u << (wsize[i] - 1)); j++)
310 if (!EC_POINT_add(group, val_sub[i][j], val_sub[i][j - 1], tmp, ctx)) goto err;
314 wNAF[i + 1] = NULL; /* make sure we always have a pivot */
315 wNAF[i] = compute_wNAF((i < num ? scalars[i] : scalar), wsize[i], &wNAF_len[i], ctx);
316 if (wNAF[i] == NULL) goto err;
317 if (wNAF_len[i] > max_len)
318 max_len = wNAF_len[i];
321 #if 1 /* optional; EC_window_bits_for_scalar_size assumes we do this step */
322 if (!EC_POINTs_make_affine(group, num_val, val, ctx)) goto err;
325 r_is_at_infinity = 1;
327 for (k = max_len - 1; k >= 0; k--)
329 if (!r_is_at_infinity)
331 if (!EC_POINT_dbl(group, r, r, ctx)) goto err;
334 for (i = 0; i < totalnum; i++)
338 int digit = wNAF[i][k];
348 if (is_neg != r_is_inverted)
350 if (!r_is_at_infinity)
352 if (!EC_POINT_invert(group, r, ctx)) goto err;
354 r_is_inverted = !r_is_inverted;
359 if (r_is_at_infinity)
361 if (!EC_POINT_copy(r, val_sub[i][digit >> 1])) goto err;
362 r_is_at_infinity = 0;
366 if (!EC_POINT_add(group, r, r, val_sub[i][digit >> 1], ctx)) goto err;
373 if (r_is_at_infinity)
375 if (!EC_POINT_set_to_infinity(group, r)) goto err;
380 if (!EC_POINT_invert(group, r, ctx)) goto err;
387 BN_CTX_free(new_ctx);
392 if (wNAF_len != NULL)
393 OPENSSL_free(wNAF_len);
398 for (w = wNAF; *w != NULL; w++)
405 for (v = val; *v != NULL; v++)
406 EC_POINT_clear_free(*v);
412 OPENSSL_free(val_sub);
420 * Basic interleaving multi-exponentation method
425 #define EC_window_bits_for_scalar_size(b) \
432 /* For window size 'w' (w >= 2), we compute the odd multiples
434 * This accounts for 2^(w-1) point additions (neglecting constants),
435 * each of which requires 16 field multiplications (4 squarings
436 * and 12 general multiplications) in the case of curves defined
437 * over GF(p), which are the only curves we have so far.
439 * Converting these precomputed points into affine form takes
440 * three field multiplications for inverting Z and one squaring
441 * and three multiplications for adjusting X and Y, i.e.
442 * 7 multiplications in total (1 squaring and 6 general multiplications),
443 * again except for constants.
445 * The average number of windows for a 'b' bit scalar is roughly
447 * Each of these windows (except possibly for the first one, but
448 * we are ignoring constants anyway) requires one point addition.
449 * As the precomputed table stores points in affine form, these
450 * additions take only 11 field multiplications each (3 squarings
451 * and 8 general multiplications).
453 * So the total workload, except for constants, is
455 * 2^(w-1)*[5 squarings + 18 multiplications]
456 * + (b/(w+1))*[3 squarings + 8 multiplications]
458 * If we assume that 10 squarings are as costly as 9 multiplications,
459 * our task is to find the 'w' that, given 'b', minimizes
461 * 2^(w-1)*(5*9 + 18*10) + (b/(w+1))*(3*9 + 8*10)
462 * = 2^(w-1)*225 + (b/(w+1))*107.
464 * Thus optimal window sizes should be roughly as follows:
466 * w >= 6 if b >= 1414
467 * w = 5 if 1413 >= b >= 505
468 * w = 4 if 504 >= b >= 169
469 * w = 3 if 168 >= b >= 51
470 * w = 2 if 50 >= b >= 13
473 * If we assume instead that squarings are exactly as costly as
474 * multiplications, we have to minimize
475 * 2^(w-1)*23 + (b/(w+1))*11.
477 * This gives us the following (nearly unchanged) table of optimal
480 * w >= 6 if b >= 1406
481 * w = 5 if 1405 >= b >= 502
482 * w = 4 if 501 >= b >= 168
483 * w = 3 if 167 >= b >= 51
484 * w = 2 if 50 >= b >= 13
487 * Note that neither table tries to take into account memory usage
488 * (allocation overhead, code locality etc.). Actual timings with
489 * NIST curves P-192, P-224, and P-256 with scalars of 192, 224,
490 * and 256 bits, respectively, show that w = 3 (instead of 4) is
491 * preferrable; timings with NIST curve P-384 and 384-bit scalars
492 * confirm that w = 4 is optimal for this case; and timings with
493 * NIST curve P-521 and 521-bit scalars show that w = 4 (instead
494 * of 5) is preferrable. So we generously round up all the
495 * boundaries and use the following table:
497 * w >= 6 if b >= 2000
498 * w = 5 if 1999 >= b >= 800
499 * w = 4 if 799 >= b >= 300
500 * w = 3 if 299 >= b >= 70
501 * w = 2 if 69 >= b >= 20
505 int EC_POINTs_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar,
506 size_t num, const EC_POINT *points[], const BIGNUM *scalars[], BN_CTX *ctx)
508 BN_CTX *new_ctx = NULL;
509 EC_POINT *generator = NULL;
510 EC_POINT *tmp = NULL;
514 int r_is_at_infinity = 1;
516 size_t *wsize = NULL; /* individual window sizes */
517 unsigned long *wbits = NULL; /* individual window contents */
518 int *wpos = NULL; /* position of bottom bit of current individual windows
519 * (wpos[i] is valid if wbits[i] != 0) */
521 EC_POINT **val = NULL; /* precomputation */
523 EC_POINT ***val_sub = NULL; /* pointers to sub-arrays of 'val' */
528 generator = EC_GROUP_get0_generator(group);
529 if (generator == NULL)
531 ECerr(EC_F_EC_POINTS_MUL, EC_R_UNDEFINED_GENERATOR);
536 for (i = 0; i < num; i++)
538 if (group->meth != points[i]->meth)
540 ECerr(EC_F_EC_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS);
545 totalnum = num + (scalar != NULL);
547 wsize = OPENSSL_malloc(totalnum * sizeof wsize[0]);
548 wbits = OPENSSL_malloc(totalnum * sizeof wbits[0]);
549 wpos = OPENSSL_malloc(totalnum * sizeof wpos[0]);
550 if (wsize == NULL || wbits == NULL || wpos == NULL) goto err;
552 /* num_val := total number of points to precompute */
554 for (i = 0; i < totalnum; i++)
558 bits = i < num ? BN_num_bits(scalars[i]) : BN_num_bits(scalar);
559 wsize[i] = EC_window_bits_for_scalar_size(bits);
560 num_val += 1u << (wsize[i] - 1);
567 /* all precomputed points go into a single array 'val',
568 * 'val_sub[i]' is a pointer to the subarray for the i-th point */
569 val = OPENSSL_malloc((num_val + 1) * sizeof val[0]);
570 if (val == NULL) goto err;
571 val[num_val] = NULL; /* pivot element */
573 val_sub = OPENSSL_malloc(totalnum * sizeof val_sub[0]);
574 if (val_sub == NULL) goto err;
576 /* allocate points for precomputation */
578 for (i = 0; i < totalnum; i++)
581 for (j = 0; j < (1u << (wsize[i] - 1)); j++)
583 *v = EC_POINT_new(group);
584 if (*v == NULL) goto err;
588 if (!(v == val + num_val))
590 ECerr(EC_F_EC_POINTS_MUL, ERR_R_INTERNAL_ERROR);
596 ctx = new_ctx = BN_CTX_new();
601 tmp = EC_POINT_new(group);
602 if (tmp == NULL) goto err;
604 /* prepare precomputed values:
605 * val_sub[i][0] := points[i]
606 * val_sub[i][1] := 3 * points[i]
607 * val_sub[i][2] := 5 * points[i]
610 for (i = 0; i < totalnum; i++)
614 if (!EC_POINT_copy(val_sub[i][0], points[i])) goto err;
617 if (!EC_POINT_invert(group, val_sub[i][0], ctx)) goto err;
622 if (!EC_POINT_copy(val_sub[i][0], generator)) goto err;
625 if (!EC_POINT_invert(group, val_sub[i][0], ctx)) goto err;
631 if (!EC_POINT_dbl(group, tmp, val_sub[i][0], ctx)) goto err;
632 for (j = 1; j < (1u << (wsize[i] - 1)); j++)
634 if (!EC_POINT_add(group, val_sub[i][j], val_sub[i][j - 1], tmp, ctx)) goto err;
639 #if 1 /* optional; EC_window_bits_for_scalar_size assumes we do this step */
640 if (!EC_POINTs_make_affine(group, num_val, val, ctx)) goto err;
643 r_is_at_infinity = 1;
645 for (k = max_bits - 1; k >= 0; k--)
647 if (!r_is_at_infinity)
649 if (!EC_POINT_dbl(group, r, r, ctx)) goto err;
652 for (i = 0; i < totalnum; i++)
658 s = i < num ? scalars[i] : scalar;
660 if (BN_is_bit_set(s, k))
662 /* look at bits k - wsize[i] + 1 .. k for this window */
663 t = k - wsize[i] + 1;
664 while (!BN_is_bit_set(s, t)) /* BN_is_bit_set is false for t < 0 */
668 for (t = k - 1; t >= wpos[i]; t--)
671 if (BN_is_bit_set(s, t))
674 /* now wbits[i] is the odd bit pattern at bits wpos[i] .. k */
678 if ((wbits[i] != 0) && (wpos[i] == k))
680 if (r_is_at_infinity)
682 if (!EC_POINT_copy(r, val_sub[i][wbits[i] >> 1])) goto err;
683 r_is_at_infinity = 0;
687 if (!EC_POINT_add(group, r, r, val_sub[i][wbits[i] >> 1], ctx)) goto err;
694 if (r_is_at_infinity)
695 if (!EC_POINT_set_to_infinity(group, r)) goto err;
701 BN_CTX_free(new_ctx);
712 for (v = val; *v != NULL; v++)
713 EC_POINT_clear_free(*v);
719 OPENSSL_free(val_sub);
726 int EC_POINT_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *g_scalar, const EC_POINT *point, const BIGNUM *p_scalar, BN_CTX *ctx)
728 const EC_POINT *points[1];
729 const BIGNUM *scalars[1];
732 scalars[0] = p_scalar;
734 return EC_POINTs_mul(group, r, g_scalar, (point != NULL && p_scalar != NULL), points, scalars, ctx);
738 int EC_GROUP_precompute_mult(EC_GROUP *group, BN_CTX *ctx)
740 const EC_POINT *generator;
741 BN_CTX *new_ctx = NULL;
745 generator = EC_GROUP_get0_generator(group);
746 if (generator == NULL)
748 ECerr(EC_F_EC_GROUP_PRECOMPUTE_MULT, EC_R_UNDEFINED_GENERATOR);
754 ctx = new_ctx = BN_CTX_new();
760 order = BN_CTX_get(ctx);
761 if (order == NULL) goto err;
763 if (!EC_GROUP_get_order(group, order, ctx)) return 0;
764 if (BN_is_zero(order))
766 ECerr(EC_F_EC_GROUP_PRECOMPUTE_MULT, EC_R_UNKNOWN_ORDER);
777 BN_CTX_free(new_ctx);