/src/libressl/crypto/ec/ec_mult.c
Line | Count | Source (jump to first uncovered line) |
1 | | /* $OpenBSD: ec_mult.c,v 1.24 2018/07/15 16:27:39 tb 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 | 0 | { |
103 | 0 | EC_PRE_COMP *ret = NULL; |
104 | |
|
105 | 0 | if (!group) |
106 | 0 | return NULL; |
107 | | |
108 | 0 | ret = malloc(sizeof(EC_PRE_COMP)); |
109 | 0 | if (!ret) { |
110 | 0 | ECerror(ERR_R_MALLOC_FAILURE); |
111 | 0 | return ret; |
112 | 0 | } |
113 | 0 | ret->group = group; |
114 | 0 | ret->blocksize = 8; /* default */ |
115 | 0 | ret->numblocks = 0; |
116 | 0 | ret->w = 4; /* default */ |
117 | 0 | ret->points = NULL; |
118 | 0 | ret->num = 0; |
119 | 0 | ret->references = 1; |
120 | 0 | return ret; |
121 | 0 | } |
122 | | |
123 | | static void * |
124 | | ec_pre_comp_dup(void *src_) |
125 | 0 | { |
126 | 0 | EC_PRE_COMP *src = src_; |
127 | | |
128 | | /* no need to actually copy, these objects never change! */ |
129 | |
|
130 | 0 | CRYPTO_add(&src->references, 1, CRYPTO_LOCK_EC_PRE_COMP); |
131 | |
|
132 | 0 | return src_; |
133 | 0 | } |
134 | | |
135 | | static void |
136 | | ec_pre_comp_free(void *pre_) |
137 | 0 | { |
138 | 0 | int i; |
139 | 0 | EC_PRE_COMP *pre = pre_; |
140 | |
|
141 | 0 | if (!pre) |
142 | 0 | return; |
143 | | |
144 | 0 | i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP); |
145 | 0 | if (i > 0) |
146 | 0 | return; |
147 | | |
148 | 0 | if (pre->points) { |
149 | 0 | EC_POINT **p; |
150 | |
|
151 | 0 | for (p = pre->points; *p != NULL; p++) |
152 | 0 | EC_POINT_free(*p); |
153 | 0 | free(pre->points); |
154 | 0 | } |
155 | 0 | free(pre); |
156 | 0 | } |
157 | | |
158 | | static void |
159 | | ec_pre_comp_clear_free(void *pre_) |
160 | 0 | { |
161 | 0 | int i; |
162 | 0 | EC_PRE_COMP *pre = pre_; |
163 | |
|
164 | 0 | if (!pre) |
165 | 0 | return; |
166 | | |
167 | 0 | i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP); |
168 | 0 | if (i > 0) |
169 | 0 | return; |
170 | | |
171 | 0 | if (pre->points) { |
172 | 0 | EC_POINT **p; |
173 | |
|
174 | 0 | for (p = pre->points; *p != NULL; p++) { |
175 | 0 | EC_POINT_clear_free(*p); |
176 | 0 | explicit_bzero(p, sizeof *p); |
177 | 0 | } |
178 | 0 | free(pre->points); |
179 | 0 | } |
180 | 0 | freezero(pre, sizeof *pre); |
181 | 0 | } |
182 | | |
183 | | |
184 | | |
185 | | |
186 | | /* Determine the modified width-(w+1) Non-Adjacent Form (wNAF) of 'scalar'. |
187 | | * This is an array r[] of values that are either zero or odd with an |
188 | | * absolute value less than 2^w satisfying |
189 | | * scalar = \sum_j r[j]*2^j |
190 | | * where at most one of any w+1 consecutive digits is non-zero |
191 | | * with the exception that the most significant digit may be only |
192 | | * w-1 zeros away from that next non-zero digit. |
193 | | */ |
194 | | static signed char * |
195 | | compute_wNAF(const BIGNUM * scalar, int w, size_t * ret_len) |
196 | 0 | { |
197 | 0 | int window_val; |
198 | 0 | int ok = 0; |
199 | 0 | signed char *r = NULL; |
200 | 0 | int sign = 1; |
201 | 0 | int bit, next_bit, mask; |
202 | 0 | size_t len = 0, j; |
203 | |
|
204 | 0 | if (BN_is_zero(scalar)) { |
205 | 0 | r = malloc(1); |
206 | 0 | if (!r) { |
207 | 0 | ECerror(ERR_R_MALLOC_FAILURE); |
208 | 0 | goto err; |
209 | 0 | } |
210 | 0 | r[0] = 0; |
211 | 0 | *ret_len = 1; |
212 | 0 | return r; |
213 | 0 | } |
214 | 0 | if (w <= 0 || w > 7) { |
215 | | /* 'signed char' can represent integers with |
216 | | * absolute values less than 2^7 */ |
217 | 0 | ECerror(ERR_R_INTERNAL_ERROR); |
218 | 0 | goto err; |
219 | 0 | } |
220 | 0 | bit = 1 << w; /* at most 128 */ |
221 | 0 | next_bit = bit << 1; /* at most 256 */ |
222 | 0 | mask = next_bit - 1; /* at most 255 */ |
223 | |
|
224 | 0 | if (BN_is_negative(scalar)) { |
225 | 0 | sign = -1; |
226 | 0 | } |
227 | 0 | if (scalar->d == NULL || scalar->top == 0) { |
228 | 0 | ECerror(ERR_R_INTERNAL_ERROR); |
229 | 0 | goto err; |
230 | 0 | } |
231 | 0 | len = BN_num_bits(scalar); |
232 | 0 | r = malloc(len + 1); /* modified wNAF may be one digit longer than |
233 | | * binary representation (*ret_len will be |
234 | | * set to the actual length, i.e. at most |
235 | | * BN_num_bits(scalar) + 1) */ |
236 | 0 | if (r == NULL) { |
237 | 0 | ECerror(ERR_R_MALLOC_FAILURE); |
238 | 0 | goto err; |
239 | 0 | } |
240 | 0 | window_val = scalar->d[0] & mask; |
241 | 0 | j = 0; |
242 | 0 | while ((window_val != 0) || (j + w + 1 < len)) { |
243 | | /* if j+w+1 >= len, window_val will not increase */ |
244 | 0 | int digit = 0; |
245 | | |
246 | | /* 0 <= window_val <= 2^(w+1) */ |
247 | 0 | if (window_val & 1) { |
248 | | /* 0 < window_val < 2^(w+1) */ |
249 | 0 | if (window_val & bit) { |
250 | 0 | digit = window_val - next_bit; /* -2^w < digit < 0 */ |
251 | |
|
252 | 0 | #if 1 /* modified wNAF */ |
253 | 0 | if (j + w + 1 >= len) { |
254 | | /* |
255 | | * special case for generating |
256 | | * modified wNAFs: no new bits will |
257 | | * be added into window_val, so using |
258 | | * a positive digit here will |
259 | | * decrease the total length of the |
260 | | * representation |
261 | | */ |
262 | |
|
263 | 0 | digit = window_val & (mask >> 1); /* 0 < digit < 2^w */ |
264 | 0 | } |
265 | 0 | #endif |
266 | 0 | } else { |
267 | 0 | digit = window_val; /* 0 < digit < 2^w */ |
268 | 0 | } |
269 | |
|
270 | 0 | if (digit <= -bit || digit >= bit || !(digit & 1)) { |
271 | 0 | ECerror(ERR_R_INTERNAL_ERROR); |
272 | 0 | goto err; |
273 | 0 | } |
274 | 0 | window_val -= digit; |
275 | | |
276 | | /* |
277 | | * now window_val is 0 or 2^(w+1) in standard wNAF |
278 | | * generation; for modified window NAFs, it may also |
279 | | * be 2^w |
280 | | */ |
281 | 0 | if (window_val != 0 && window_val != next_bit && window_val != bit) { |
282 | 0 | ECerror(ERR_R_INTERNAL_ERROR); |
283 | 0 | goto err; |
284 | 0 | } |
285 | 0 | } |
286 | 0 | r[j++] = sign * digit; |
287 | |
|
288 | 0 | window_val >>= 1; |
289 | 0 | window_val += bit * BN_is_bit_set(scalar, j + w); |
290 | |
|
291 | 0 | if (window_val > next_bit) { |
292 | 0 | ECerror(ERR_R_INTERNAL_ERROR); |
293 | 0 | goto err; |
294 | 0 | } |
295 | 0 | } |
296 | | |
297 | 0 | if (j > len + 1) { |
298 | 0 | ECerror(ERR_R_INTERNAL_ERROR); |
299 | 0 | goto err; |
300 | 0 | } |
301 | 0 | len = j; |
302 | 0 | ok = 1; |
303 | |
|
304 | 0 | err: |
305 | 0 | if (!ok) { |
306 | 0 | free(r); |
307 | 0 | r = NULL; |
308 | 0 | } |
309 | 0 | if (ok) |
310 | 0 | *ret_len = len; |
311 | 0 | return r; |
312 | 0 | } |
313 | | |
314 | | |
315 | | /* TODO: table should be optimised for the wNAF-based implementation, |
316 | | * sometimes smaller windows will give better performance |
317 | | * (thus the boundaries should be increased) |
318 | | */ |
319 | | #define EC_window_bits_for_scalar_size(b) \ |
320 | 0 | ((size_t) \ |
321 | 0 | ((b) >= 2000 ? 6 : \ |
322 | 0 | (b) >= 800 ? 5 : \ |
323 | 0 | (b) >= 300 ? 4 : \ |
324 | 0 | (b) >= 70 ? 3 : \ |
325 | 0 | (b) >= 20 ? 2 : \ |
326 | 0 | 1)) |
327 | | |
328 | | /* Compute |
329 | | * \sum scalars[i]*points[i], |
330 | | * also including |
331 | | * scalar*generator |
332 | | * in the addition if scalar != NULL |
333 | | */ |
334 | | int |
335 | | ec_wNAF_mul(const EC_GROUP * group, EC_POINT * r, const BIGNUM * scalar, |
336 | | size_t num, const EC_POINT * points[], const BIGNUM * scalars[], BN_CTX * ctx) |
337 | 0 | { |
338 | 0 | BN_CTX *new_ctx = NULL; |
339 | 0 | const EC_POINT *generator = NULL; |
340 | 0 | EC_POINT *tmp = NULL; |
341 | 0 | size_t totalnum; |
342 | 0 | size_t blocksize = 0, numblocks = 0; /* for wNAF splitting */ |
343 | 0 | size_t pre_points_per_block = 0; |
344 | 0 | size_t i, j; |
345 | 0 | int k; |
346 | 0 | int r_is_inverted = 0; |
347 | 0 | int r_is_at_infinity = 1; |
348 | 0 | size_t *wsize = NULL; /* individual window sizes */ |
349 | 0 | signed char **wNAF = NULL; /* individual wNAFs */ |
350 | 0 | signed char *tmp_wNAF = NULL; |
351 | 0 | size_t *wNAF_len = NULL; |
352 | 0 | size_t max_len = 0; |
353 | 0 | size_t num_val; |
354 | 0 | EC_POINT **val = NULL; /* precomputation */ |
355 | 0 | EC_POINT **v; |
356 | 0 | EC_POINT ***val_sub = NULL; /* pointers to sub-arrays of 'val' or |
357 | | * 'pre_comp->points' */ |
358 | 0 | const EC_PRE_COMP *pre_comp = NULL; |
359 | 0 | 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 | 0 | int ret = 0; |
363 | |
|
364 | 0 | if (group->meth != r->meth) { |
365 | 0 | ECerror(EC_R_INCOMPATIBLE_OBJECTS); |
366 | 0 | return 0; |
367 | 0 | } |
368 | 0 | if ((scalar == NULL) && (num == 0)) { |
369 | 0 | return EC_POINT_set_to_infinity(group, r); |
370 | 0 | } |
371 | 0 | for (i = 0; i < num; i++) { |
372 | 0 | if (group->meth != points[i]->meth) { |
373 | 0 | ECerror(EC_R_INCOMPATIBLE_OBJECTS); |
374 | 0 | return 0; |
375 | 0 | } |
376 | 0 | } |
377 | | |
378 | 0 | if (ctx == NULL) { |
379 | 0 | ctx = new_ctx = BN_CTX_new(); |
380 | 0 | if (ctx == NULL) |
381 | 0 | goto err; |
382 | 0 | } |
383 | 0 | if (scalar != NULL) { |
384 | 0 | generator = EC_GROUP_get0_generator(group); |
385 | 0 | if (generator == NULL) { |
386 | 0 | ECerror(EC_R_UNDEFINED_GENERATOR); |
387 | 0 | goto err; |
388 | 0 | } |
389 | | /* look if we can use precomputed multiples of generator */ |
390 | | |
391 | 0 | 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 | 0 | if (pre_comp && pre_comp->numblocks && |
394 | 0 | (EC_POINT_cmp(group, generator, pre_comp->points[0], ctx) == 0)) { |
395 | 0 | 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 | 0 | numblocks = (BN_num_bits(scalar) / blocksize) + 1; |
403 | | |
404 | | /* |
405 | | * we cannot use more blocks than we have |
406 | | * precomputation for |
407 | | */ |
408 | 0 | if (numblocks > pre_comp->numblocks) |
409 | 0 | numblocks = pre_comp->numblocks; |
410 | |
|
411 | 0 | pre_points_per_block = (size_t) 1 << (pre_comp->w - 1); |
412 | | |
413 | | /* check that pre_comp looks sane */ |
414 | 0 | if (pre_comp->num != (pre_comp->numblocks * pre_points_per_block)) { |
415 | 0 | ECerror(ERR_R_INTERNAL_ERROR); |
416 | 0 | goto err; |
417 | 0 | } |
418 | 0 | } else { |
419 | | /* can't use precomputation */ |
420 | 0 | pre_comp = NULL; |
421 | 0 | numblocks = 1; |
422 | 0 | num_scalar = 1; /* treat 'scalar' like 'num'-th |
423 | | * element of 'scalars' */ |
424 | 0 | } |
425 | 0 | } |
426 | 0 | totalnum = num + numblocks; |
427 | | |
428 | | /* includes space for pivot */ |
429 | 0 | wNAF = reallocarray(NULL, (totalnum + 1), sizeof wNAF[0]); |
430 | 0 | if (wNAF == NULL) { |
431 | 0 | ECerror(ERR_R_MALLOC_FAILURE); |
432 | 0 | goto err; |
433 | 0 | } |
434 | | |
435 | 0 | wNAF[0] = NULL; /* preliminary pivot */ |
436 | |
|
437 | 0 | wsize = reallocarray(NULL, totalnum, sizeof wsize[0]); |
438 | 0 | wNAF_len = reallocarray(NULL, totalnum, sizeof wNAF_len[0]); |
439 | 0 | val_sub = reallocarray(NULL, totalnum, sizeof val_sub[0]); |
440 | |
|
441 | 0 | if (wsize == NULL || wNAF_len == NULL || val_sub == NULL) { |
442 | 0 | ECerror(ERR_R_MALLOC_FAILURE); |
443 | 0 | goto err; |
444 | 0 | } |
445 | | |
446 | | /* num_val will be the total number of temporarily precomputed points */ |
447 | 0 | num_val = 0; |
448 | |
|
449 | 0 | for (i = 0; i < num + num_scalar; i++) { |
450 | 0 | size_t bits; |
451 | |
|
452 | 0 | bits = i < num ? BN_num_bits(scalars[i]) : BN_num_bits(scalar); |
453 | 0 | wsize[i] = EC_window_bits_for_scalar_size(bits); |
454 | 0 | num_val += (size_t) 1 << (wsize[i] - 1); |
455 | 0 | wNAF[i + 1] = NULL; /* make sure we always have a pivot */ |
456 | 0 | wNAF[i] = compute_wNAF((i < num ? scalars[i] : scalar), wsize[i], &wNAF_len[i]); |
457 | 0 | if (wNAF[i] == NULL) |
458 | 0 | goto err; |
459 | 0 | if (wNAF_len[i] > max_len) |
460 | 0 | max_len = wNAF_len[i]; |
461 | 0 | } |
462 | | |
463 | 0 | if (numblocks) { |
464 | | /* we go here iff scalar != NULL */ |
465 | |
|
466 | 0 | if (pre_comp == NULL) { |
467 | 0 | if (num_scalar != 1) { |
468 | 0 | ECerror(ERR_R_INTERNAL_ERROR); |
469 | 0 | goto err; |
470 | 0 | } |
471 | | /* we have already generated a wNAF for 'scalar' */ |
472 | 0 | } else { |
473 | 0 | size_t tmp_len = 0; |
474 | |
|
475 | 0 | if (num_scalar != 0) { |
476 | 0 | ECerror(ERR_R_INTERNAL_ERROR); |
477 | 0 | goto err; |
478 | 0 | } |
479 | | /* |
480 | | * use the window size for which we have |
481 | | * precomputation |
482 | | */ |
483 | 0 | wsize[num] = pre_comp->w; |
484 | 0 | tmp_wNAF = compute_wNAF(scalar, wsize[num], &tmp_len); |
485 | 0 | if (tmp_wNAF == NULL) |
486 | 0 | goto err; |
487 | | |
488 | 0 | if (tmp_len <= max_len) { |
489 | | /* |
490 | | * One of the other wNAFs is at least as long |
491 | | * as the wNAF belonging to the generator, so |
492 | | * wNAF splitting will not buy us anything. |
493 | | */ |
494 | |
|
495 | 0 | numblocks = 1; |
496 | 0 | totalnum = num + 1; /* don't use wNAF |
497 | | * splitting */ |
498 | 0 | wNAF[num] = tmp_wNAF; |
499 | 0 | tmp_wNAF = NULL; |
500 | 0 | wNAF[num + 1] = NULL; |
501 | 0 | wNAF_len[num] = tmp_len; |
502 | 0 | if (tmp_len > max_len) |
503 | 0 | max_len = tmp_len; |
504 | | /* |
505 | | * pre_comp->points starts with the points |
506 | | * that we need here: |
507 | | */ |
508 | 0 | val_sub[num] = pre_comp->points; |
509 | 0 | } else { |
510 | | /* |
511 | | * don't include tmp_wNAF directly into wNAF |
512 | | * array - use wNAF splitting and include the |
513 | | * blocks |
514 | | */ |
515 | |
|
516 | 0 | signed char *pp; |
517 | 0 | EC_POINT **tmp_points; |
518 | |
|
519 | 0 | if (tmp_len < numblocks * blocksize) { |
520 | | /* |
521 | | * possibly we can do with fewer |
522 | | * blocks than estimated |
523 | | */ |
524 | 0 | numblocks = (tmp_len + blocksize - 1) / blocksize; |
525 | 0 | if (numblocks > pre_comp->numblocks) { |
526 | 0 | ECerror(ERR_R_INTERNAL_ERROR); |
527 | 0 | goto err; |
528 | 0 | } |
529 | 0 | totalnum = num + numblocks; |
530 | 0 | } |
531 | | /* split wNAF in 'numblocks' parts */ |
532 | 0 | pp = tmp_wNAF; |
533 | 0 | tmp_points = pre_comp->points; |
534 | |
|
535 | 0 | for (i = num; i < totalnum; i++) { |
536 | 0 | if (i < totalnum - 1) { |
537 | 0 | wNAF_len[i] = blocksize; |
538 | 0 | if (tmp_len < blocksize) { |
539 | 0 | ECerror(ERR_R_INTERNAL_ERROR); |
540 | 0 | goto err; |
541 | 0 | } |
542 | 0 | tmp_len -= blocksize; |
543 | 0 | } else |
544 | | /* |
545 | | * last block gets whatever |
546 | | * is left (this could be |
547 | | * more or less than |
548 | | * 'blocksize'!) |
549 | | */ |
550 | 0 | wNAF_len[i] = tmp_len; |
551 | | |
552 | 0 | wNAF[i + 1] = NULL; |
553 | 0 | wNAF[i] = malloc(wNAF_len[i]); |
554 | 0 | if (wNAF[i] == NULL) { |
555 | 0 | ECerror(ERR_R_MALLOC_FAILURE); |
556 | 0 | goto err; |
557 | 0 | } |
558 | 0 | memcpy(wNAF[i], pp, wNAF_len[i]); |
559 | 0 | if (wNAF_len[i] > max_len) |
560 | 0 | max_len = wNAF_len[i]; |
561 | |
|
562 | 0 | if (*tmp_points == NULL) { |
563 | 0 | ECerror(ERR_R_INTERNAL_ERROR); |
564 | 0 | goto err; |
565 | 0 | } |
566 | 0 | val_sub[i] = tmp_points; |
567 | 0 | tmp_points += pre_points_per_block; |
568 | 0 | pp += blocksize; |
569 | 0 | } |
570 | 0 | } |
571 | 0 | } |
572 | 0 | } |
573 | | /* |
574 | | * All points we precompute now go into a single array 'val'. |
575 | | * 'val_sub[i]' is a pointer to the subarray for the i-th point, or |
576 | | * to a subarray of 'pre_comp->points' if we already have |
577 | | * precomputation. |
578 | | */ |
579 | 0 | val = reallocarray(NULL, (num_val + 1), sizeof val[0]); |
580 | 0 | if (val == NULL) { |
581 | 0 | ECerror(ERR_R_MALLOC_FAILURE); |
582 | 0 | goto err; |
583 | 0 | } |
584 | 0 | val[num_val] = NULL; /* pivot element */ |
585 | | |
586 | | /* allocate points for precomputation */ |
587 | 0 | v = val; |
588 | 0 | for (i = 0; i < num + num_scalar; i++) { |
589 | 0 | val_sub[i] = v; |
590 | 0 | for (j = 0; j < ((size_t) 1 << (wsize[i] - 1)); j++) { |
591 | 0 | *v = EC_POINT_new(group); |
592 | 0 | if (*v == NULL) |
593 | 0 | goto err; |
594 | 0 | v++; |
595 | 0 | } |
596 | 0 | } |
597 | 0 | if (!(v == val + num_val)) { |
598 | 0 | ECerror(ERR_R_INTERNAL_ERROR); |
599 | 0 | goto err; |
600 | 0 | } |
601 | 0 | if (!(tmp = EC_POINT_new(group))) |
602 | 0 | goto err; |
603 | | |
604 | | /* |
605 | | * prepare precomputed values: val_sub[i][0] := points[i] |
606 | | * val_sub[i][1] := 3 * points[i] val_sub[i][2] := 5 * points[i] ... |
607 | | */ |
608 | 0 | for (i = 0; i < num + num_scalar; i++) { |
609 | 0 | if (i < num) { |
610 | 0 | if (!EC_POINT_copy(val_sub[i][0], points[i])) |
611 | 0 | goto err; |
612 | 0 | } else { |
613 | 0 | if (!EC_POINT_copy(val_sub[i][0], generator)) |
614 | 0 | goto err; |
615 | 0 | } |
616 | | |
617 | 0 | if (wsize[i] > 1) { |
618 | 0 | if (!EC_POINT_dbl(group, tmp, val_sub[i][0], ctx)) |
619 | 0 | goto err; |
620 | 0 | for (j = 1; j < ((size_t) 1 << (wsize[i] - 1)); j++) { |
621 | 0 | if (!EC_POINT_add(group, val_sub[i][j], val_sub[i][j - 1], tmp, ctx)) |
622 | 0 | goto err; |
623 | 0 | } |
624 | 0 | } |
625 | 0 | } |
626 | | |
627 | 0 | if (!EC_POINTs_make_affine(group, num_val, val, ctx)) |
628 | 0 | goto err; |
629 | | |
630 | 0 | r_is_at_infinity = 1; |
631 | |
|
632 | 0 | for (k = max_len - 1; k >= 0; k--) { |
633 | 0 | if (!r_is_at_infinity) { |
634 | 0 | if (!EC_POINT_dbl(group, r, r, ctx)) |
635 | 0 | goto err; |
636 | 0 | } |
637 | 0 | for (i = 0; i < totalnum; i++) { |
638 | 0 | if (wNAF_len[i] > (size_t) k) { |
639 | 0 | int digit = wNAF[i][k]; |
640 | 0 | int is_neg; |
641 | |
|
642 | 0 | if (digit) { |
643 | 0 | is_neg = digit < 0; |
644 | |
|
645 | 0 | if (is_neg) |
646 | 0 | digit = -digit; |
647 | |
|
648 | 0 | if (is_neg != r_is_inverted) { |
649 | 0 | if (!r_is_at_infinity) { |
650 | 0 | if (!EC_POINT_invert(group, r, ctx)) |
651 | 0 | goto err; |
652 | 0 | } |
653 | 0 | r_is_inverted = !r_is_inverted; |
654 | 0 | } |
655 | | /* digit > 0 */ |
656 | | |
657 | 0 | if (r_is_at_infinity) { |
658 | 0 | if (!EC_POINT_copy(r, val_sub[i][digit >> 1])) |
659 | 0 | goto err; |
660 | 0 | r_is_at_infinity = 0; |
661 | 0 | } else { |
662 | 0 | if (!EC_POINT_add(group, r, r, val_sub[i][digit >> 1], ctx)) |
663 | 0 | goto err; |
664 | 0 | } |
665 | 0 | } |
666 | 0 | } |
667 | 0 | } |
668 | 0 | } |
669 | | |
670 | 0 | if (r_is_at_infinity) { |
671 | 0 | if (!EC_POINT_set_to_infinity(group, r)) |
672 | 0 | goto err; |
673 | 0 | } else { |
674 | 0 | if (r_is_inverted) |
675 | 0 | if (!EC_POINT_invert(group, r, ctx)) |
676 | 0 | goto err; |
677 | 0 | } |
678 | | |
679 | 0 | ret = 1; |
680 | |
|
681 | 0 | err: |
682 | 0 | BN_CTX_free(new_ctx); |
683 | 0 | EC_POINT_free(tmp); |
684 | 0 | free(wsize); |
685 | 0 | free(wNAF_len); |
686 | 0 | free(tmp_wNAF); |
687 | 0 | if (wNAF != NULL) { |
688 | 0 | signed char **w; |
689 | |
|
690 | 0 | for (w = wNAF; *w != NULL; w++) |
691 | 0 | free(*w); |
692 | |
|
693 | 0 | free(wNAF); |
694 | 0 | } |
695 | 0 | if (val != NULL) { |
696 | 0 | for (v = val; *v != NULL; v++) |
697 | 0 | EC_POINT_clear_free(*v); |
698 | 0 | free(val); |
699 | 0 | } |
700 | 0 | free(val_sub); |
701 | 0 | return ret; |
702 | 0 | } |
703 | | |
704 | | |
705 | | /* ec_wNAF_precompute_mult() |
706 | | * creates an EC_PRE_COMP object with preprecomputed multiples of the generator |
707 | | * for use with wNAF splitting as implemented in ec_wNAF_mul(). |
708 | | * |
709 | | * 'pre_comp->points' is an array of multiples of the generator |
710 | | * of the following form: |
711 | | * points[0] = generator; |
712 | | * points[1] = 3 * generator; |
713 | | * ... |
714 | | * points[2^(w-1)-1] = (2^(w-1)-1) * generator; |
715 | | * points[2^(w-1)] = 2^blocksize * generator; |
716 | | * points[2^(w-1)+1] = 3 * 2^blocksize * generator; |
717 | | * ... |
718 | | * points[2^(w-1)*(numblocks-1)-1] = (2^(w-1)) * 2^(blocksize*(numblocks-2)) * generator |
719 | | * points[2^(w-1)*(numblocks-1)] = 2^(blocksize*(numblocks-1)) * generator |
720 | | * ... |
721 | | * points[2^(w-1)*numblocks-1] = (2^(w-1)) * 2^(blocksize*(numblocks-1)) * generator |
722 | | * points[2^(w-1)*numblocks] = NULL |
723 | | */ |
724 | | int |
725 | | ec_wNAF_precompute_mult(EC_GROUP * group, BN_CTX * ctx) |
726 | 0 | { |
727 | 0 | const EC_POINT *generator; |
728 | 0 | EC_POINT *tmp_point = NULL, *base = NULL, **var; |
729 | 0 | BN_CTX *new_ctx = NULL; |
730 | 0 | BIGNUM *order; |
731 | 0 | size_t i, bits, w, pre_points_per_block, blocksize, numblocks, |
732 | 0 | num; |
733 | 0 | EC_POINT **points = NULL; |
734 | 0 | EC_PRE_COMP *pre_comp; |
735 | 0 | int ret = 0; |
736 | | |
737 | | /* if there is an old EC_PRE_COMP object, throw it away */ |
738 | 0 | EC_EX_DATA_free_data(&group->extra_data, ec_pre_comp_dup, ec_pre_comp_free, ec_pre_comp_clear_free); |
739 | |
|
740 | 0 | if ((pre_comp = ec_pre_comp_new(group)) == NULL) |
741 | 0 | return 0; |
742 | | |
743 | 0 | generator = EC_GROUP_get0_generator(group); |
744 | 0 | if (generator == NULL) { |
745 | 0 | ECerror(EC_R_UNDEFINED_GENERATOR); |
746 | 0 | goto err; |
747 | 0 | } |
748 | 0 | if (ctx == NULL) { |
749 | 0 | ctx = new_ctx = BN_CTX_new(); |
750 | 0 | if (ctx == NULL) |
751 | 0 | goto err; |
752 | 0 | } |
753 | 0 | BN_CTX_start(ctx); |
754 | 0 | if ((order = BN_CTX_get(ctx)) == NULL) |
755 | 0 | goto err; |
756 | | |
757 | 0 | if (!EC_GROUP_get_order(group, order, ctx)) |
758 | 0 | goto err; |
759 | 0 | if (BN_is_zero(order)) { |
760 | 0 | ECerror(EC_R_UNKNOWN_ORDER); |
761 | 0 | goto err; |
762 | 0 | } |
763 | 0 | 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 | 0 | blocksize = 8; |
773 | 0 | w = 4; |
774 | 0 | if (EC_window_bits_for_scalar_size(bits) > w) { |
775 | | /* let's not make the window too small ... */ |
776 | 0 | w = EC_window_bits_for_scalar_size(bits); |
777 | 0 | } |
778 | 0 | numblocks = (bits + blocksize - 1) / blocksize; /* max. number of blocks |
779 | | * to use for wNAF |
780 | | * splitting */ |
781 | |
|
782 | 0 | pre_points_per_block = (size_t) 1 << (w - 1); |
783 | 0 | num = pre_points_per_block * numblocks; /* number of points to |
784 | | * compute and store */ |
785 | |
|
786 | 0 | points = reallocarray(NULL, (num + 1), sizeof(EC_POINT *)); |
787 | 0 | if (!points) { |
788 | 0 | ECerror(ERR_R_MALLOC_FAILURE); |
789 | 0 | goto err; |
790 | 0 | } |
791 | 0 | var = points; |
792 | 0 | var[num] = NULL; /* pivot */ |
793 | 0 | for (i = 0; i < num; i++) { |
794 | 0 | if ((var[i] = EC_POINT_new(group)) == NULL) { |
795 | 0 | ECerror(ERR_R_MALLOC_FAILURE); |
796 | 0 | goto err; |
797 | 0 | } |
798 | 0 | } |
799 | | |
800 | 0 | if (!(tmp_point = EC_POINT_new(group)) || !(base = EC_POINT_new(group))) { |
801 | 0 | ECerror(ERR_R_MALLOC_FAILURE); |
802 | 0 | goto err; |
803 | 0 | } |
804 | 0 | if (!EC_POINT_copy(base, generator)) |
805 | 0 | goto err; |
806 | | |
807 | | /* do the precomputation */ |
808 | 0 | for (i = 0; i < numblocks; i++) { |
809 | 0 | size_t j; |
810 | |
|
811 | 0 | if (!EC_POINT_dbl(group, tmp_point, base, ctx)) |
812 | 0 | goto err; |
813 | | |
814 | 0 | if (!EC_POINT_copy(*var++, base)) |
815 | 0 | goto err; |
816 | | |
817 | 0 | for (j = 1; j < pre_points_per_block; j++, var++) { |
818 | | /* calculate odd multiples of the current base point */ |
819 | 0 | if (!EC_POINT_add(group, *var, tmp_point, *(var - 1), ctx)) |
820 | 0 | goto err; |
821 | 0 | } |
822 | | |
823 | 0 | if (i < numblocks - 1) { |
824 | | /* |
825 | | * get the next base (multiply current one by |
826 | | * 2^blocksize) |
827 | | */ |
828 | 0 | size_t k; |
829 | |
|
830 | 0 | if (blocksize <= 2) { |
831 | 0 | ECerror(ERR_R_INTERNAL_ERROR); |
832 | 0 | goto err; |
833 | 0 | } |
834 | 0 | if (!EC_POINT_dbl(group, base, tmp_point, ctx)) |
835 | 0 | goto err; |
836 | 0 | for (k = 2; k < blocksize; k++) { |
837 | 0 | if (!EC_POINT_dbl(group, base, base, ctx)) |
838 | 0 | goto err; |
839 | 0 | } |
840 | 0 | } |
841 | 0 | } |
842 | | |
843 | 0 | if (!EC_POINTs_make_affine(group, num, points, ctx)) |
844 | 0 | goto err; |
845 | | |
846 | 0 | pre_comp->group = group; |
847 | 0 | pre_comp->blocksize = blocksize; |
848 | 0 | pre_comp->numblocks = numblocks; |
849 | 0 | pre_comp->w = w; |
850 | 0 | pre_comp->points = points; |
851 | 0 | points = NULL; |
852 | 0 | pre_comp->num = num; |
853 | |
|
854 | 0 | if (!EC_EX_DATA_set_data(&group->extra_data, pre_comp, |
855 | 0 | ec_pre_comp_dup, ec_pre_comp_free, ec_pre_comp_clear_free)) |
856 | 0 | goto err; |
857 | 0 | pre_comp = NULL; |
858 | |
|
859 | 0 | ret = 1; |
860 | 0 | err: |
861 | 0 | if (ctx != NULL) |
862 | 0 | BN_CTX_end(ctx); |
863 | 0 | BN_CTX_free(new_ctx); |
864 | 0 | ec_pre_comp_free(pre_comp); |
865 | 0 | if (points) { |
866 | 0 | EC_POINT **p; |
867 | |
|
868 | 0 | for (p = points; *p != NULL; p++) |
869 | 0 | EC_POINT_free(*p); |
870 | 0 | free(points); |
871 | 0 | } |
872 | 0 | EC_POINT_free(tmp_point); |
873 | 0 | EC_POINT_free(base); |
874 | 0 | return ret; |
875 | 0 | } |
876 | | |
877 | | |
878 | | int |
879 | | ec_wNAF_have_precompute_mult(const EC_GROUP * group) |
880 | 0 | { |
881 | 0 | if (EC_EX_DATA_get_data(group->extra_data, ec_pre_comp_dup, ec_pre_comp_free, ec_pre_comp_clear_free) != NULL) |
882 | 0 | return 1; |
883 | 0 | else |
884 | 0 | return 0; |
885 | 0 | } |