diff options
Diffstat (limited to 'src/lib/libcrypto/ec/ec_mult.c')
-rw-r--r-- | src/lib/libcrypto/ec/ec_mult.c | 673 |
1 files changed, 563 insertions, 110 deletions
diff --git a/src/lib/libcrypto/ec/ec_mult.c b/src/lib/libcrypto/ec/ec_mult.c index 16822a73cf..2ba173ef36 100644 --- a/src/lib/libcrypto/ec/ec_mult.c +++ b/src/lib/libcrypto/ec/ec_mult.c | |||
@@ -1,6 +1,9 @@ | |||
1 | /* crypto/ec/ec_mult.c */ | 1 | /* crypto/ec/ec_mult.c */ |
2 | /* | ||
3 | * Originally written by Bodo Moeller and Nils Larsch for the OpenSSL project. | ||
4 | */ | ||
2 | /* ==================================================================== | 5 | /* ==================================================================== |
3 | * Copyright (c) 1998-2001 The OpenSSL Project. All rights reserved. | 6 | * Copyright (c) 1998-2007 The OpenSSL Project. All rights reserved. |
4 | * | 7 | * |
5 | * Redistribution and use in source and binary forms, with or without | 8 | * Redistribution and use in source and binary forms, with or without |
6 | * modification, are permitted provided that the following conditions | 9 | * modification, are permitted provided that the following conditions |
@@ -52,41 +55,161 @@ | |||
52 | * Hudson (tjh@cryptsoft.com). | 55 | * Hudson (tjh@cryptsoft.com). |
53 | * | 56 | * |
54 | */ | 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> | ||
55 | 65 | ||
56 | #include <openssl/err.h> | 66 | #include <openssl/err.h> |
57 | 67 | ||
58 | #include "ec_lcl.h" | 68 | #include "ec_lcl.h" |
59 | 69 | ||
60 | 70 | ||
61 | /* TODO: optional precomputation of multiples of the generator */ | 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 | */ | ||
62 | 77 | ||
63 | 78 | ||
64 | 79 | ||
65 | /* | 80 | |
66 | * wNAF-based interleaving multi-exponentation method | 81 | /* structure for precomputed multiples of the generator */ |
67 | * (<URL:http://www.informatik.tu-darmstadt.de/TI/Mitarbeiter/moeller.html#multiexp>) | 82 | typedef struct ec_pre_comp_st { |
68 | */ | 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 precomputation */ | ||
86 | size_t w; /* window size */ | ||
87 | EC_POINT **points; /* array with pre-calculated multiples of generator: | ||
88 | * 'num' pointers to EC_POINT objects followed by a NULL */ | ||
89 | size_t num; /* numblocks * 2^(w-1) */ | ||
90 | int references; | ||
91 | } EC_PRE_COMP; | ||
92 | |||
93 | /* functions to manage EC_PRE_COMP within the EC_GROUP extra_data framework */ | ||
94 | static void *ec_pre_comp_dup(void *); | ||
95 | static void ec_pre_comp_free(void *); | ||
96 | static void ec_pre_comp_clear_free(void *); | ||
97 | |||
98 | static EC_PRE_COMP *ec_pre_comp_new(const EC_GROUP *group) | ||
99 | { | ||
100 | EC_PRE_COMP *ret = NULL; | ||
101 | |||
102 | if (!group) | ||
103 | return NULL; | ||
104 | |||
105 | ret = (EC_PRE_COMP *)OPENSSL_malloc(sizeof(EC_PRE_COMP)); | ||
106 | if (!ret) | ||
107 | { | ||
108 | ECerr(EC_F_EC_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE); | ||
109 | return ret; | ||
110 | } | ||
111 | ret->group = group; | ||
112 | ret->blocksize = 8; /* default */ | ||
113 | ret->numblocks = 0; | ||
114 | ret->w = 4; /* default */ | ||
115 | ret->points = NULL; | ||
116 | ret->num = 0; | ||
117 | ret->references = 1; | ||
118 | return ret; | ||
119 | } | ||
120 | |||
121 | static void *ec_pre_comp_dup(void *src_) | ||
122 | { | ||
123 | EC_PRE_COMP *src = src_; | ||
124 | |||
125 | /* no need to actually copy, these objects never change! */ | ||
126 | |||
127 | CRYPTO_add(&src->references, 1, CRYPTO_LOCK_EC_PRE_COMP); | ||
128 | |||
129 | return src_; | ||
130 | } | ||
131 | |||
132 | static void ec_pre_comp_free(void *pre_) | ||
133 | { | ||
134 | int i; | ||
135 | EC_PRE_COMP *pre = pre_; | ||
136 | |||
137 | if (!pre) | ||
138 | return; | ||
139 | |||
140 | i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP); | ||
141 | if (i > 0) | ||
142 | return; | ||
143 | |||
144 | if (pre->points) | ||
145 | { | ||
146 | EC_POINT **p; | ||
147 | |||
148 | for (p = pre->points; *p != NULL; p++) | ||
149 | EC_POINT_free(*p); | ||
150 | OPENSSL_free(pre->points); | ||
151 | } | ||
152 | OPENSSL_free(pre); | ||
153 | } | ||
154 | |||
155 | static void ec_pre_comp_clear_free(void *pre_) | ||
156 | { | ||
157 | int i; | ||
158 | EC_PRE_COMP *pre = pre_; | ||
159 | |||
160 | if (!pre) | ||
161 | return; | ||
162 | |||
163 | i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP); | ||
164 | if (i > 0) | ||
165 | return; | ||
166 | |||
167 | if (pre->points) | ||
168 | { | ||
169 | EC_POINT **p; | ||
170 | |||
171 | for (p = pre->points; *p != NULL; p++) | ||
172 | EC_POINT_clear_free(*p); | ||
173 | OPENSSL_cleanse(pre->points, sizeof pre->points); | ||
174 | OPENSSL_free(pre->points); | ||
175 | } | ||
176 | OPENSSL_cleanse(pre, sizeof pre); | ||
177 | OPENSSL_free(pre); | ||
178 | } | ||
179 | |||
69 | 180 | ||
70 | 181 | ||
71 | /* Determine the width-(w+1) Non-Adjacent Form (wNAF) of 'scalar'. | 182 | |
183 | /* Determine the modified width-(w+1) Non-Adjacent Form (wNAF) of 'scalar'. | ||
72 | * This is an array r[] of values that are either zero or odd with an | 184 | * This is an array r[] of values that are either zero or odd with an |
73 | * absolute value less than 2^w satisfying | 185 | * absolute value less than 2^w satisfying |
74 | * scalar = \sum_j r[j]*2^j | 186 | * scalar = \sum_j r[j]*2^j |
75 | * where at most one of any w+1 consecutive digits is non-zero. | 187 | * where at most one of any w+1 consecutive digits is non-zero |
188 | * with the exception that the most significant digit may be only | ||
189 | * w-1 zeros away from that next non-zero digit. | ||
76 | */ | 190 | */ |
77 | static signed char *compute_wNAF(const BIGNUM *scalar, int w, size_t *ret_len, BN_CTX *ctx) | 191 | static signed char *compute_wNAF(const BIGNUM *scalar, int w, size_t *ret_len) |
78 | { | 192 | { |
79 | BIGNUM *c; | 193 | int window_val; |
80 | int ok = 0; | 194 | int ok = 0; |
81 | signed char *r = NULL; | 195 | signed char *r = NULL; |
82 | int sign = 1; | 196 | int sign = 1; |
83 | int bit, next_bit, mask; | 197 | int bit, next_bit, mask; |
84 | size_t len = 0, j; | 198 | size_t len = 0, j; |
85 | 199 | ||
86 | BN_CTX_start(ctx); | 200 | if (BN_is_zero(scalar)) |
87 | c = BN_CTX_get(ctx); | 201 | { |
88 | if (c == NULL) goto err; | 202 | r = OPENSSL_malloc(1); |
89 | 203 | if (!r) | |
204 | { | ||
205 | ECerr(EC_F_COMPUTE_WNAF, ERR_R_MALLOC_FAILURE); | ||
206 | goto err; | ||
207 | } | ||
208 | r[0] = 0; | ||
209 | *ret_len = 1; | ||
210 | return r; | ||
211 | } | ||
212 | |||
90 | if (w <= 0 || w > 7) /* 'signed char' can represent integers with absolute values less than 2^7 */ | 213 | if (w <= 0 || w > 7) /* 'signed char' can represent integers with absolute values less than 2^7 */ |
91 | { | 214 | { |
92 | ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR); | 215 | ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR); |
@@ -96,60 +219,90 @@ static signed char *compute_wNAF(const BIGNUM *scalar, int w, size_t *ret_len, B | |||
96 | next_bit = bit << 1; /* at most 256 */ | 219 | next_bit = bit << 1; /* at most 256 */ |
97 | mask = next_bit - 1; /* at most 255 */ | 220 | mask = next_bit - 1; /* at most 255 */ |
98 | 221 | ||
99 | if (!BN_copy(c, scalar)) goto err; | 222 | if (BN_is_negative(scalar)) |
100 | if (c->neg) | ||
101 | { | 223 | { |
102 | sign = -1; | 224 | sign = -1; |
103 | c->neg = 0; | ||
104 | } | 225 | } |
105 | 226 | ||
106 | len = BN_num_bits(c) + 1; /* wNAF may be one digit longer than binary representation */ | 227 | len = BN_num_bits(scalar); |
107 | r = OPENSSL_malloc(len); | 228 | r = OPENSSL_malloc(len + 1); /* modified wNAF may be one digit longer than binary representation |
108 | if (r == NULL) goto err; | 229 | * (*ret_len will be set to the actual length, i.e. at most |
230 | * BN_num_bits(scalar) + 1) */ | ||
231 | if (r == NULL) | ||
232 | { | ||
233 | ECerr(EC_F_COMPUTE_WNAF, ERR_R_MALLOC_FAILURE); | ||
234 | goto err; | ||
235 | } | ||
109 | 236 | ||
237 | if (scalar->d == NULL || scalar->top == 0) | ||
238 | { | ||
239 | ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR); | ||
240 | goto err; | ||
241 | } | ||
242 | window_val = scalar->d[0] & mask; | ||
110 | j = 0; | 243 | j = 0; |
111 | while (!BN_is_zero(c)) | 244 | while ((window_val != 0) || (j + w + 1 < len)) /* if j+w+1 >= len, window_val will not increase */ |
112 | { | 245 | { |
113 | int u = 0; | 246 | int digit = 0; |
114 | 247 | ||
115 | if (BN_is_odd(c)) | 248 | /* 0 <= window_val <= 2^(w+1) */ |
249 | |||
250 | if (window_val & 1) | ||
116 | { | 251 | { |
117 | if (c->d == NULL || c->top == 0) | 252 | /* 0 < window_val < 2^(w+1) */ |
253 | |||
254 | if (window_val & bit) | ||
118 | { | 255 | { |
119 | ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR); | 256 | digit = window_val - next_bit; /* -2^w < digit < 0 */ |
120 | goto err; | 257 | |
258 | #if 1 /* modified wNAF */ | ||
259 | if (j + w + 1 >= len) | ||
260 | { | ||
261 | /* special case for generating modified wNAFs: | ||
262 | * no new bits will be added into window_val, | ||
263 | * so using a positive digit here will decrease | ||
264 | * the total length of the representation */ | ||
265 | |||
266 | digit = window_val & (mask >> 1); /* 0 < digit < 2^w */ | ||
267 | } | ||
268 | #endif | ||
121 | } | 269 | } |
122 | u = c->d[0] & mask; | 270 | else |
123 | if (u & bit) | ||
124 | { | 271 | { |
125 | u -= next_bit; | 272 | digit = window_val; /* 0 < digit < 2^w */ |
126 | /* u < 0 */ | ||
127 | if (!BN_add_word(c, -u)) goto err; | ||
128 | } | 273 | } |
129 | else | 274 | |
275 | if (digit <= -bit || digit >= bit || !(digit & 1)) | ||
130 | { | 276 | { |
131 | /* u > 0 */ | 277 | ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR); |
132 | if (!BN_sub_word(c, u)) goto err; | 278 | goto err; |
133 | } | 279 | } |
134 | 280 | ||
135 | if (u <= -bit || u >= bit || !(u & 1) || c->neg) | 281 | window_val -= digit; |
282 | |||
283 | /* now window_val is 0 or 2^(w+1) in standard wNAF generation; | ||
284 | * for modified window NAFs, it may also be 2^w | ||
285 | */ | ||
286 | if (window_val != 0 && window_val != next_bit && window_val != bit) | ||
136 | { | 287 | { |
137 | ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR); | 288 | ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR); |
138 | goto err; | 289 | goto err; |
139 | } | 290 | } |
140 | } | 291 | } |
141 | 292 | ||
142 | r[j++] = sign * u; | 293 | r[j++] = sign * digit; |
143 | 294 | ||
144 | if (BN_is_odd(c)) | 295 | window_val >>= 1; |
296 | window_val += bit * BN_is_bit_set(scalar, j + w); | ||
297 | |||
298 | if (window_val > next_bit) | ||
145 | { | 299 | { |
146 | ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR); | 300 | ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR); |
147 | goto err; | 301 | goto err; |
148 | } | 302 | } |
149 | if (!BN_rshift1(c, c)) goto err; | ||
150 | } | 303 | } |
151 | 304 | ||
152 | if (j > len) | 305 | if (j > len + 1) |
153 | { | 306 | { |
154 | ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR); | 307 | ECerr(EC_F_COMPUTE_WNAF, ERR_R_INTERNAL_ERROR); |
155 | goto err; | 308 | goto err; |
@@ -158,7 +311,6 @@ static signed char *compute_wNAF(const BIGNUM *scalar, int w, size_t *ret_len, B | |||
158 | ok = 1; | 311 | ok = 1; |
159 | 312 | ||
160 | err: | 313 | err: |
161 | BN_CTX_end(ctx); | ||
162 | if (!ok) | 314 | if (!ok) |
163 | { | 315 | { |
164 | OPENSSL_free(r); | 316 | OPENSSL_free(r); |
@@ -181,7 +333,7 @@ static signed char *compute_wNAF(const BIGNUM *scalar, int w, size_t *ret_len, B | |||
181 | (b) >= 300 ? 4 : \ | 333 | (b) >= 300 ? 4 : \ |
182 | (b) >= 70 ? 3 : \ | 334 | (b) >= 70 ? 3 : \ |
183 | (b) >= 20 ? 2 : \ | 335 | (b) >= 20 ? 2 : \ |
184 | 1)) | 336 | 1)) |
185 | 337 | ||
186 | /* Compute | 338 | /* Compute |
187 | * \sum scalars[i]*points[i], | 339 | * \sum scalars[i]*points[i], |
@@ -189,13 +341,15 @@ static signed char *compute_wNAF(const BIGNUM *scalar, int w, size_t *ret_len, B | |||
189 | * scalar*generator | 341 | * scalar*generator |
190 | * in the addition if scalar != NULL | 342 | * in the addition if scalar != NULL |
191 | */ | 343 | */ |
192 | int EC_POINTs_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, | 344 | int ec_wNAF_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, |
193 | size_t num, const EC_POINT *points[], const BIGNUM *scalars[], BN_CTX *ctx) | 345 | size_t num, const EC_POINT *points[], const BIGNUM *scalars[], BN_CTX *ctx) |
194 | { | 346 | { |
195 | BN_CTX *new_ctx = NULL; | 347 | BN_CTX *new_ctx = NULL; |
196 | EC_POINT *generator = NULL; | 348 | const EC_POINT *generator = NULL; |
197 | EC_POINT *tmp = NULL; | 349 | EC_POINT *tmp = NULL; |
198 | size_t totalnum; | 350 | size_t totalnum; |
351 | size_t blocksize = 0, numblocks = 0; /* for wNAF splitting */ | ||
352 | size_t pre_points_per_block = 0; | ||
199 | size_t i, j; | 353 | size_t i, j; |
200 | int k; | 354 | int k; |
201 | int r_is_inverted = 0; | 355 | int r_is_inverted = 0; |
@@ -207,12 +361,15 @@ int EC_POINTs_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, | |||
207 | size_t num_val; | 361 | size_t num_val; |
208 | EC_POINT **val = NULL; /* precomputation */ | 362 | EC_POINT **val = NULL; /* precomputation */ |
209 | EC_POINT **v; | 363 | EC_POINT **v; |
210 | EC_POINT ***val_sub = NULL; /* pointers to sub-arrays of 'val' */ | 364 | EC_POINT ***val_sub = NULL; /* pointers to sub-arrays of 'val' or 'pre_comp->points' */ |
365 | const EC_PRE_COMP *pre_comp = NULL; | ||
366 | int num_scalar = 0; /* flag: will be set to 1 if 'scalar' must be treated like other scalars, | ||
367 | * i.e. precomputation is not available */ | ||
211 | int ret = 0; | 368 | int ret = 0; |
212 | 369 | ||
213 | if (group->meth != r->meth) | 370 | if (group->meth != r->meth) |
214 | { | 371 | { |
215 | ECerr(EC_F_EC_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS); | 372 | ECerr(EC_F_EC_WNAF_MUL, EC_R_INCOMPATIBLE_OBJECTS); |
216 | return 0; | 373 | return 0; |
217 | } | 374 | } |
218 | 375 | ||
@@ -221,59 +378,226 @@ int EC_POINTs_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, | |||
221 | return EC_POINT_set_to_infinity(group, r); | 378 | return EC_POINT_set_to_infinity(group, r); |
222 | } | 379 | } |
223 | 380 | ||
224 | if (scalar != NULL) | 381 | for (i = 0; i < num; i++) |
225 | { | 382 | { |
226 | generator = EC_GROUP_get0_generator(group); | 383 | if (group->meth != points[i]->meth) |
227 | if (generator == NULL) | ||
228 | { | 384 | { |
229 | ECerr(EC_F_EC_POINTS_MUL, EC_R_UNDEFINED_GENERATOR); | 385 | ECerr(EC_F_EC_WNAF_MUL, EC_R_INCOMPATIBLE_OBJECTS); |
230 | return 0; | 386 | return 0; |
231 | } | 387 | } |
232 | } | 388 | } |
233 | 389 | ||
234 | for (i = 0; i < num; i++) | 390 | if (ctx == NULL) |
235 | { | 391 | { |
236 | if (group->meth != points[i]->meth) | 392 | ctx = new_ctx = BN_CTX_new(); |
393 | if (ctx == NULL) | ||
394 | goto err; | ||
395 | } | ||
396 | |||
397 | if (scalar != NULL) | ||
398 | { | ||
399 | generator = EC_GROUP_get0_generator(group); | ||
400 | if (generator == NULL) | ||
237 | { | 401 | { |
238 | ECerr(EC_F_EC_POINTS_MUL, EC_R_INCOMPATIBLE_OBJECTS); | 402 | ECerr(EC_F_EC_WNAF_MUL, EC_R_UNDEFINED_GENERATOR); |
239 | return 0; | 403 | goto err; |
240 | } | 404 | } |
241 | } | 405 | |
406 | /* look if we can use precomputed multiples of generator */ | ||
407 | |||
408 | pre_comp = EC_EX_DATA_get_data(group->extra_data, ec_pre_comp_dup, ec_pre_comp_free, ec_pre_comp_clear_free); | ||
409 | |||
410 | if (pre_comp && pre_comp->numblocks && (EC_POINT_cmp(group, generator, pre_comp->points[0], ctx) == 0)) | ||
411 | { | ||
412 | blocksize = pre_comp->blocksize; | ||
242 | 413 | ||
243 | totalnum = num + (scalar != NULL); | 414 | /* determine maximum number of blocks that wNAF splitting may yield |
415 | * (NB: maximum wNAF length is bit length plus one) */ | ||
416 | numblocks = (BN_num_bits(scalar) / blocksize) + 1; | ||
244 | 417 | ||
245 | wsize = OPENSSL_malloc(totalnum * sizeof wsize[0]); | 418 | /* we cannot use more blocks than we have precomputation for */ |
419 | if (numblocks > pre_comp->numblocks) | ||
420 | numblocks = pre_comp->numblocks; | ||
421 | |||
422 | pre_points_per_block = 1u << (pre_comp->w - 1); | ||
423 | |||
424 | /* check that pre_comp looks sane */ | ||
425 | if (pre_comp->num != (pre_comp->numblocks * pre_points_per_block)) | ||
426 | { | ||
427 | ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR); | ||
428 | goto err; | ||
429 | } | ||
430 | } | ||
431 | else | ||
432 | { | ||
433 | /* can't use precomputation */ | ||
434 | pre_comp = NULL; | ||
435 | numblocks = 1; | ||
436 | num_scalar = 1; /* treat 'scalar' like 'num'-th element of 'scalars' */ | ||
437 | } | ||
438 | } | ||
439 | |||
440 | totalnum = num + numblocks; | ||
441 | |||
442 | wsize = OPENSSL_malloc(totalnum * sizeof wsize[0]); | ||
246 | wNAF_len = OPENSSL_malloc(totalnum * sizeof wNAF_len[0]); | 443 | wNAF_len = OPENSSL_malloc(totalnum * sizeof wNAF_len[0]); |
247 | wNAF = OPENSSL_malloc((totalnum + 1) * sizeof wNAF[0]); | 444 | wNAF = OPENSSL_malloc((totalnum + 1) * sizeof wNAF[0]); /* includes space for pivot */ |
248 | if (wNAF != NULL) | 445 | val_sub = OPENSSL_malloc(totalnum * sizeof val_sub[0]); |
446 | |||
447 | if (!wsize || !wNAF_len || !wNAF || !val_sub) | ||
249 | { | 448 | { |
250 | wNAF[0] = NULL; /* preliminary pivot */ | 449 | ECerr(EC_F_EC_WNAF_MUL, ERR_R_MALLOC_FAILURE); |
450 | goto err; | ||
251 | } | 451 | } |
252 | if (wsize == NULL || wNAF_len == NULL || wNAF == NULL) goto err; | ||
253 | 452 | ||
254 | /* num_val := total number of points to precompute */ | 453 | wNAF[0] = NULL; /* preliminary pivot */ |
454 | |||
455 | /* num_val will be the total number of temporarily precomputed points */ | ||
255 | num_val = 0; | 456 | num_val = 0; |
256 | for (i = 0; i < totalnum; i++) | 457 | |
458 | for (i = 0; i < num + num_scalar; i++) | ||
257 | { | 459 | { |
258 | size_t bits; | 460 | size_t bits; |
259 | 461 | ||
260 | bits = i < num ? BN_num_bits(scalars[i]) : BN_num_bits(scalar); | 462 | bits = i < num ? BN_num_bits(scalars[i]) : BN_num_bits(scalar); |
261 | wsize[i] = EC_window_bits_for_scalar_size(bits); | 463 | wsize[i] = EC_window_bits_for_scalar_size(bits); |
262 | num_val += 1u << (wsize[i] - 1); | 464 | num_val += 1u << (wsize[i] - 1); |
465 | wNAF[i + 1] = NULL; /* make sure we always have a pivot */ | ||
466 | wNAF[i] = compute_wNAF((i < num ? scalars[i] : scalar), wsize[i], &wNAF_len[i]); | ||
467 | if (wNAF[i] == NULL) | ||
468 | goto err; | ||
469 | if (wNAF_len[i] > max_len) | ||
470 | max_len = wNAF_len[i]; | ||
471 | } | ||
472 | |||
473 | if (numblocks) | ||
474 | { | ||
475 | /* we go here iff scalar != NULL */ | ||
476 | |||
477 | if (pre_comp == NULL) | ||
478 | { | ||
479 | if (num_scalar != 1) | ||
480 | { | ||
481 | ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR); | ||
482 | goto err; | ||
483 | } | ||
484 | /* we have already generated a wNAF for 'scalar' */ | ||
485 | } | ||
486 | else | ||
487 | { | ||
488 | signed char *tmp_wNAF = NULL; | ||
489 | size_t tmp_len = 0; | ||
490 | |||
491 | if (num_scalar != 0) | ||
492 | { | ||
493 | ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR); | ||
494 | goto err; | ||
495 | } | ||
496 | |||
497 | /* use the window size for which we have precomputation */ | ||
498 | wsize[num] = pre_comp->w; | ||
499 | tmp_wNAF = compute_wNAF(scalar, wsize[num], &tmp_len); | ||
500 | if (!tmp_wNAF) | ||
501 | goto err; | ||
502 | |||
503 | if (tmp_len <= max_len) | ||
504 | { | ||
505 | /* One of the other wNAFs is at least as long | ||
506 | * as the wNAF belonging to the generator, | ||
507 | * so wNAF splitting will not buy us anything. */ | ||
508 | |||
509 | numblocks = 1; | ||
510 | totalnum = num + 1; /* don't use wNAF splitting */ | ||
511 | wNAF[num] = tmp_wNAF; | ||
512 | wNAF[num + 1] = NULL; | ||
513 | wNAF_len[num] = tmp_len; | ||
514 | if (tmp_len > max_len) | ||
515 | max_len = tmp_len; | ||
516 | /* pre_comp->points starts with the points that we need here: */ | ||
517 | val_sub[num] = pre_comp->points; | ||
518 | } | ||
519 | else | ||
520 | { | ||
521 | /* don't include tmp_wNAF directly into wNAF array | ||
522 | * - use wNAF splitting and include the blocks */ | ||
523 | |||
524 | signed char *pp; | ||
525 | EC_POINT **tmp_points; | ||
526 | |||
527 | if (tmp_len < numblocks * blocksize) | ||
528 | { | ||
529 | /* possibly we can do with fewer blocks than estimated */ | ||
530 | numblocks = (tmp_len + blocksize - 1) / blocksize; | ||
531 | if (numblocks > pre_comp->numblocks) | ||
532 | { | ||
533 | ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR); | ||
534 | goto err; | ||
535 | } | ||
536 | totalnum = num + numblocks; | ||
537 | } | ||
538 | |||
539 | /* split wNAF in 'numblocks' parts */ | ||
540 | pp = tmp_wNAF; | ||
541 | tmp_points = pre_comp->points; | ||
542 | |||
543 | for (i = num; i < totalnum; i++) | ||
544 | { | ||
545 | if (i < totalnum - 1) | ||
546 | { | ||
547 | wNAF_len[i] = blocksize; | ||
548 | if (tmp_len < blocksize) | ||
549 | { | ||
550 | ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR); | ||
551 | goto err; | ||
552 | } | ||
553 | tmp_len -= blocksize; | ||
554 | } | ||
555 | else | ||
556 | /* last block gets whatever is left | ||
557 | * (this could be more or less than 'blocksize'!) */ | ||
558 | wNAF_len[i] = tmp_len; | ||
559 | |||
560 | wNAF[i + 1] = NULL; | ||
561 | wNAF[i] = OPENSSL_malloc(wNAF_len[i]); | ||
562 | if (wNAF[i] == NULL) | ||
563 | { | ||
564 | ECerr(EC_F_EC_WNAF_MUL, ERR_R_MALLOC_FAILURE); | ||
565 | OPENSSL_free(tmp_wNAF); | ||
566 | goto err; | ||
567 | } | ||
568 | memcpy(wNAF[i], pp, wNAF_len[i]); | ||
569 | if (wNAF_len[i] > max_len) | ||
570 | max_len = wNAF_len[i]; | ||
571 | |||
572 | if (*tmp_points == NULL) | ||
573 | { | ||
574 | ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR); | ||
575 | OPENSSL_free(tmp_wNAF); | ||
576 | goto err; | ||
577 | } | ||
578 | val_sub[i] = tmp_points; | ||
579 | tmp_points += pre_points_per_block; | ||
580 | pp += blocksize; | ||
581 | } | ||
582 | OPENSSL_free(tmp_wNAF); | ||
583 | } | ||
584 | } | ||
263 | } | 585 | } |
264 | 586 | ||
265 | /* all precomputed points go into a single array 'val', | 587 | /* All points we precompute now go into a single array 'val'. |
266 | * 'val_sub[i]' is a pointer to the subarray for the i-th point */ | 588 | * 'val_sub[i]' is a pointer to the subarray for the i-th point, |
589 | * or to a subarray of 'pre_comp->points' if we already have precomputation. */ | ||
267 | val = OPENSSL_malloc((num_val + 1) * sizeof val[0]); | 590 | val = OPENSSL_malloc((num_val + 1) * sizeof val[0]); |
268 | if (val == NULL) goto err; | 591 | if (val == NULL) |
592 | { | ||
593 | ECerr(EC_F_EC_WNAF_MUL, ERR_R_MALLOC_FAILURE); | ||
594 | goto err; | ||
595 | } | ||
269 | val[num_val] = NULL; /* pivot element */ | 596 | val[num_val] = NULL; /* pivot element */ |
270 | 597 | ||
271 | val_sub = OPENSSL_malloc(totalnum * sizeof val_sub[0]); | ||
272 | if (val_sub == NULL) goto err; | ||
273 | |||
274 | /* allocate points for precomputation */ | 598 | /* allocate points for precomputation */ |
275 | v = val; | 599 | v = val; |
276 | for (i = 0; i < totalnum; i++) | 600 | for (i = 0; i < num + num_scalar; i++) |
277 | { | 601 | { |
278 | val_sub[i] = v; | 602 | val_sub[i] = v; |
279 | for (j = 0; j < (1u << (wsize[i] - 1)); j++) | 603 | for (j = 0; j < (1u << (wsize[i] - 1)); j++) |
@@ -285,19 +609,12 @@ int EC_POINTs_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, | |||
285 | } | 609 | } |
286 | if (!(v == val + num_val)) | 610 | if (!(v == val + num_val)) |
287 | { | 611 | { |
288 | ECerr(EC_F_EC_POINTS_MUL, ERR_R_INTERNAL_ERROR); | 612 | ECerr(EC_F_EC_WNAF_MUL, ERR_R_INTERNAL_ERROR); |
289 | goto err; | 613 | goto err; |
290 | } | 614 | } |
291 | 615 | ||
292 | if (ctx == NULL) | 616 | if (!(tmp = EC_POINT_new(group))) |
293 | { | 617 | goto err; |
294 | ctx = new_ctx = BN_CTX_new(); | ||
295 | if (ctx == NULL) | ||
296 | goto err; | ||
297 | } | ||
298 | |||
299 | tmp = EC_POINT_new(group); | ||
300 | if (tmp == NULL) goto err; | ||
301 | 618 | ||
302 | /* prepare precomputed values: | 619 | /* prepare precomputed values: |
303 | * val_sub[i][0] := points[i] | 620 | * val_sub[i][0] := points[i] |
@@ -305,7 +622,7 @@ int EC_POINTs_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, | |||
305 | * val_sub[i][2] := 5 * points[i] | 622 | * val_sub[i][2] := 5 * points[i] |
306 | * ... | 623 | * ... |
307 | */ | 624 | */ |
308 | for (i = 0; i < totalnum; i++) | 625 | for (i = 0; i < num + num_scalar; i++) |
309 | { | 626 | { |
310 | if (i < num) | 627 | if (i < num) |
311 | { | 628 | { |
@@ -324,16 +641,11 @@ int EC_POINTs_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, | |||
324 | if (!EC_POINT_add(group, val_sub[i][j], val_sub[i][j - 1], tmp, ctx)) goto err; | 641 | if (!EC_POINT_add(group, val_sub[i][j], val_sub[i][j - 1], tmp, ctx)) goto err; |
325 | } | 642 | } |
326 | } | 643 | } |
327 | |||
328 | wNAF[i + 1] = NULL; /* make sure we always have a pivot */ | ||
329 | wNAF[i] = compute_wNAF((i < num ? scalars[i] : scalar), wsize[i], &wNAF_len[i], ctx); | ||
330 | if (wNAF[i] == NULL) goto err; | ||
331 | if (wNAF_len[i] > max_len) | ||
332 | max_len = wNAF_len[i]; | ||
333 | } | 644 | } |
334 | 645 | ||
335 | #if 1 /* optional; EC_window_bits_for_scalar_size assumes we do this step */ | 646 | #if 1 /* optional; EC_window_bits_for_scalar_size assumes we do this step */ |
336 | if (!EC_POINTs_make_affine(group, num_val, val, ctx)) goto err; | 647 | if (!EC_POINTs_make_affine(group, num_val, val, ctx)) |
648 | goto err; | ||
337 | #endif | 649 | #endif |
338 | 650 | ||
339 | r_is_at_infinity = 1; | 651 | r_is_at_infinity = 1; |
@@ -429,57 +741,198 @@ int EC_POINTs_mul(const EC_GROUP *group, EC_POINT *r, const BIGNUM *scalar, | |||
429 | } | 741 | } |
430 | 742 | ||
431 | 743 | ||
432 | 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) | 744 | /* ec_wNAF_precompute_mult() |
433 | { | 745 | * creates an EC_PRE_COMP object with preprecomputed multiples of the generator |
434 | const EC_POINT *points[1]; | 746 | * for use with wNAF splitting as implemented in ec_wNAF_mul(). |
435 | const BIGNUM *scalars[1]; | 747 | * |
436 | 748 | * 'pre_comp->points' is an array of multiples of the generator | |
437 | points[0] = point; | 749 | * of the following form: |
438 | scalars[0] = p_scalar; | 750 | * points[0] = generator; |
439 | 751 | * points[1] = 3 * generator; | |
440 | return EC_POINTs_mul(group, r, g_scalar, (point != NULL && p_scalar != NULL), points, scalars, ctx); | 752 | * ... |
441 | } | 753 | * points[2^(w-1)-1] = (2^(w-1)-1) * generator; |
442 | 754 | * points[2^(w-1)] = 2^blocksize * generator; | |
443 | 755 | * points[2^(w-1)+1] = 3 * 2^blocksize * generator; | |
444 | int EC_GROUP_precompute_mult(EC_GROUP *group, BN_CTX *ctx) | 756 | * ... |
757 | * points[2^(w-1)*(numblocks-1)-1] = (2^(w-1)) * 2^(blocksize*(numblocks-2)) * generator | ||
758 | * points[2^(w-1)*(numblocks-1)] = 2^(blocksize*(numblocks-1)) * generator | ||
759 | * ... | ||
760 | * points[2^(w-1)*numblocks-1] = (2^(w-1)) * 2^(blocksize*(numblocks-1)) * generator | ||
761 | * points[2^(w-1)*numblocks] = NULL | ||
762 | */ | ||
763 | int ec_wNAF_precompute_mult(EC_GROUP *group, BN_CTX *ctx) | ||
445 | { | 764 | { |
446 | const EC_POINT *generator; | 765 | const EC_POINT *generator; |
766 | EC_POINT *tmp_point = NULL, *base = NULL, **var; | ||
447 | BN_CTX *new_ctx = NULL; | 767 | BN_CTX *new_ctx = NULL; |
448 | BIGNUM *order; | 768 | BIGNUM *order; |
769 | size_t i, bits, w, pre_points_per_block, blocksize, numblocks, num; | ||
770 | EC_POINT **points = NULL; | ||
771 | EC_PRE_COMP *pre_comp; | ||
449 | int ret = 0; | 772 | int ret = 0; |
450 | 773 | ||
774 | /* if there is an old EC_PRE_COMP object, throw it away */ | ||
775 | EC_EX_DATA_free_data(&group->extra_data, ec_pre_comp_dup, ec_pre_comp_free, ec_pre_comp_clear_free); | ||
776 | |||
777 | if ((pre_comp = ec_pre_comp_new(group)) == NULL) | ||
778 | return 0; | ||
779 | |||
451 | generator = EC_GROUP_get0_generator(group); | 780 | generator = EC_GROUP_get0_generator(group); |
452 | if (generator == NULL) | 781 | if (generator == NULL) |
453 | { | 782 | { |
454 | ECerr(EC_F_EC_GROUP_PRECOMPUTE_MULT, EC_R_UNDEFINED_GENERATOR); | 783 | ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, EC_R_UNDEFINED_GENERATOR); |
455 | return 0; | 784 | goto err; |
456 | } | 785 | } |
457 | 786 | ||
458 | if (ctx == NULL) | 787 | if (ctx == NULL) |
459 | { | 788 | { |
460 | ctx = new_ctx = BN_CTX_new(); | 789 | ctx = new_ctx = BN_CTX_new(); |
461 | if (ctx == NULL) | 790 | if (ctx == NULL) |
462 | return 0; | 791 | goto err; |
463 | } | 792 | } |
464 | 793 | ||
465 | BN_CTX_start(ctx); | 794 | BN_CTX_start(ctx); |
466 | order = BN_CTX_get(ctx); | 795 | order = BN_CTX_get(ctx); |
467 | if (order == NULL) goto err; | 796 | if (order == NULL) goto err; |
468 | 797 | ||
469 | if (!EC_GROUP_get_order(group, order, ctx)) return 0; | 798 | if (!EC_GROUP_get_order(group, order, ctx)) goto err; |
470 | if (BN_is_zero(order)) | 799 | if (BN_is_zero(order)) |
471 | { | 800 | { |
472 | ECerr(EC_F_EC_GROUP_PRECOMPUTE_MULT, EC_R_UNKNOWN_ORDER); | 801 | ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, EC_R_UNKNOWN_ORDER); |
473 | goto err; | 802 | goto err; |
474 | } | 803 | } |
475 | 804 | ||
476 | /* TODO */ | 805 | bits = BN_num_bits(order); |
806 | /* The following parameters mean we precompute (approximately) | ||
807 | * one point per bit. | ||
808 | * | ||
809 | * TBD: The combination 8, 4 is perfect for 160 bits; for other | ||
810 | * bit lengths, other parameter combinations might provide better | ||
811 | * efficiency. | ||
812 | */ | ||
813 | blocksize = 8; | ||
814 | w = 4; | ||
815 | if (EC_window_bits_for_scalar_size(bits) > w) | ||
816 | { | ||
817 | /* let's not make the window too small ... */ | ||
818 | w = EC_window_bits_for_scalar_size(bits); | ||
819 | } | ||
820 | |||
821 | numblocks = (bits + blocksize - 1) / blocksize; /* max. number of blocks to use for wNAF splitting */ | ||
822 | |||
823 | pre_points_per_block = 1u << (w - 1); | ||
824 | num = pre_points_per_block * numblocks; /* number of points to compute and store */ | ||
477 | 825 | ||
478 | ret = 1; | 826 | points = OPENSSL_malloc(sizeof (EC_POINT*)*(num + 1)); |
827 | if (!points) | ||
828 | { | ||
829 | ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, ERR_R_MALLOC_FAILURE); | ||
830 | goto err; | ||
831 | } | ||
832 | |||
833 | var = points; | ||
834 | var[num] = NULL; /* pivot */ | ||
835 | for (i = 0; i < num; i++) | ||
836 | { | ||
837 | if ((var[i] = EC_POINT_new(group)) == NULL) | ||
838 | { | ||
839 | ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, ERR_R_MALLOC_FAILURE); | ||
840 | goto err; | ||
841 | } | ||
842 | } | ||
843 | |||
844 | if (!(tmp_point = EC_POINT_new(group)) || !(base = EC_POINT_new(group))) | ||
845 | { | ||
846 | ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, ERR_R_MALLOC_FAILURE); | ||
847 | goto err; | ||
848 | } | ||
849 | |||
850 | if (!EC_POINT_copy(base, generator)) | ||
851 | goto err; | ||
852 | |||
853 | /* do the precomputation */ | ||
854 | for (i = 0; i < numblocks; i++) | ||
855 | { | ||
856 | size_t j; | ||
857 | |||
858 | if (!EC_POINT_dbl(group, tmp_point, base, ctx)) | ||
859 | goto err; | ||
860 | |||
861 | if (!EC_POINT_copy(*var++, base)) | ||
862 | goto err; | ||
863 | |||
864 | for (j = 1; j < pre_points_per_block; j++, var++) | ||
865 | { | ||
866 | /* calculate odd multiples of the current base point */ | ||
867 | if (!EC_POINT_add(group, *var, tmp_point, *(var - 1), ctx)) | ||
868 | goto err; | ||
869 | } | ||
870 | |||
871 | if (i < numblocks - 1) | ||
872 | { | ||
873 | /* get the next base (multiply current one by 2^blocksize) */ | ||
874 | size_t k; | ||
875 | |||
876 | if (blocksize <= 2) | ||
877 | { | ||
878 | ECerr(EC_F_EC_WNAF_PRECOMPUTE_MULT, ERR_R_INTERNAL_ERROR); | ||
879 | goto err; | ||
880 | } | ||
881 | |||
882 | if (!EC_POINT_dbl(group, base, tmp_point, ctx)) | ||
883 | goto err; | ||
884 | for (k = 2; k < blocksize; k++) | ||
885 | { | ||
886 | if (!EC_POINT_dbl(group,base,base,ctx)) | ||
887 | goto err; | ||
888 | } | ||
889 | } | ||
890 | } | ||
891 | |||
892 | if (!EC_POINTs_make_affine(group, num, points, ctx)) | ||
893 | goto err; | ||
479 | 894 | ||
895 | pre_comp->group = group; | ||
896 | pre_comp->blocksize = blocksize; | ||
897 | pre_comp->numblocks = numblocks; | ||
898 | pre_comp->w = w; | ||
899 | pre_comp->points = points; | ||
900 | points = NULL; | ||
901 | pre_comp->num = num; | ||
902 | |||
903 | if (!EC_EX_DATA_set_data(&group->extra_data, pre_comp, | ||
904 | ec_pre_comp_dup, ec_pre_comp_free, ec_pre_comp_clear_free)) | ||
905 | goto err; | ||
906 | pre_comp = NULL; | ||
907 | |||
908 | ret = 1; | ||
480 | err: | 909 | err: |
481 | BN_CTX_end(ctx); | 910 | if (ctx != NULL) |
911 | BN_CTX_end(ctx); | ||
482 | if (new_ctx != NULL) | 912 | if (new_ctx != NULL) |
483 | BN_CTX_free(new_ctx); | 913 | BN_CTX_free(new_ctx); |
914 | if (pre_comp) | ||
915 | ec_pre_comp_free(pre_comp); | ||
916 | if (points) | ||
917 | { | ||
918 | EC_POINT **p; | ||
919 | |||
920 | for (p = points; *p != NULL; p++) | ||
921 | EC_POINT_free(*p); | ||
922 | OPENSSL_free(points); | ||
923 | } | ||
924 | if (tmp_point) | ||
925 | EC_POINT_free(tmp_point); | ||
926 | if (base) | ||
927 | EC_POINT_free(base); | ||
484 | return ret; | 928 | return ret; |
485 | } | 929 | } |
930 | |||
931 | |||
932 | int ec_wNAF_have_precompute_mult(const EC_GROUP *group) | ||
933 | { | ||
934 | if (EC_EX_DATA_get_data(group->extra_data, ec_pre_comp_dup, ec_pre_comp_free, ec_pre_comp_clear_free) != NULL) | ||
935 | return 1; | ||
936 | else | ||
937 | return 0; | ||
938 | } | ||