/src/openssl/crypto/bn/bn_local.h
Line | Count | Source (jump to first uncovered line) |
1 | | /* |
2 | | * Copyright 1995-2023 The OpenSSL Project Authors. All Rights Reserved. |
3 | | * |
4 | | * Licensed under the Apache License 2.0 (the "License"). You may not use |
5 | | * this file except in compliance with the License. You can obtain a copy |
6 | | * in the file LICENSE in the source distribution or at |
7 | | * https://www.openssl.org/source/license.html |
8 | | */ |
9 | | |
10 | | #ifndef OSSL_CRYPTO_BN_LOCAL_H |
11 | | # define OSSL_CRYPTO_BN_LOCAL_H |
12 | | |
13 | | /* |
14 | | * The EDK2 build doesn't use bn_conf.h; it sets THIRTY_TWO_BIT or |
15 | | * SIXTY_FOUR_BIT in its own environment since it doesn't re-run our |
16 | | * Configure script and needs to support both 32-bit and 64-bit. |
17 | | */ |
18 | | # include <openssl/opensslconf.h> |
19 | | |
20 | | # if !defined(OPENSSL_SYS_UEFI) |
21 | | # include "crypto/bn_conf.h" |
22 | | # endif |
23 | | |
24 | | # include "crypto/bn.h" |
25 | | # include "internal/cryptlib.h" |
26 | | # include "internal/numbers.h" |
27 | | |
28 | | /* |
29 | | * These preprocessor symbols control various aspects of the bignum headers |
30 | | * and library code. They're not defined by any "normal" configuration, as |
31 | | * they are intended for development and testing purposes. NB: defining |
32 | | * them can be useful for debugging application code as well as openssl |
33 | | * itself. BN_DEBUG - turn on various debugging alterations to the bignum |
34 | | * code BN_RAND_DEBUG - uses random poisoning of unused words to trip up |
35 | | * mismanagement of bignum internals. Enable BN_RAND_DEBUG is known to |
36 | | * break some of the OpenSSL tests. |
37 | | */ |
38 | | # if defined(BN_RAND_DEBUG) && !defined(BN_DEBUG) |
39 | | # define BN_DEBUG |
40 | | # endif |
41 | | # if defined(BN_RAND_DEBUG) |
42 | | # include <openssl/rand.h> |
43 | | # endif |
44 | | |
45 | | /* |
46 | | * This should limit the stack usage due to alloca to about 4K. |
47 | | * BN_SOFT_LIMIT is a soft limit equivalent to 2*OPENSSL_RSA_MAX_MODULUS_BITS. |
48 | | * Beyond that size bn_mul_mont is no longer used, and the constant time |
49 | | * assembler code is disabled, due to the blatant alloca and bn_mul_mont usage. |
50 | | * Note that bn_mul_mont does an alloca that is hidden away in assembly. |
51 | | * It is not recommended to do computations with numbers exceeding this limit, |
52 | | * since the result will be highly version dependent: |
53 | | * While the current OpenSSL version will use non-optimized, but safe code, |
54 | | * previous versions will use optimized code, that may crash due to unexpected |
55 | | * stack overflow, and future versions may very well turn this into a hard |
56 | | * limit. |
57 | | * Note however, that it is possible to override the size limit using |
58 | | * "./config -DBN_SOFT_LIMIT=<limit>" if necessary, and the O/S specific |
59 | | * stack limit is known and taken into consideration. |
60 | | */ |
61 | | # ifndef BN_SOFT_LIMIT |
62 | 73.0k | # define BN_SOFT_LIMIT (4096 / BN_BYTES) |
63 | | # endif |
64 | | |
65 | | # ifndef OPENSSL_SMALL_FOOTPRINT |
66 | | # define BN_MUL_COMBA |
67 | | # define BN_SQR_COMBA |
68 | | # define BN_RECURSION |
69 | | # endif |
70 | | |
71 | | /* |
72 | | * This next option uses the C libraries (2 word)/(1 word) function. If it is |
73 | | * not defined, I use my C version (which is slower). The reason for this |
74 | | * flag is that when the particular C compiler library routine is used, and |
75 | | * the library is linked with a different compiler, the library is missing. |
76 | | * This mostly happens when the library is built with gcc and then linked |
77 | | * using normal cc. This would be a common occurrence because gcc normally |
78 | | * produces code that is 2 times faster than system compilers for the big |
79 | | * number stuff. For machines with only one compiler (or shared libraries), |
80 | | * this should be on. Again this in only really a problem on machines using |
81 | | * "long long's", are 32bit, and are not using my assembler code. |
82 | | */ |
83 | | # if defined(OPENSSL_SYS_MSDOS) || defined(OPENSSL_SYS_WINDOWS) || \ |
84 | | defined(OPENSSL_SYS_WIN32) || defined(linux) |
85 | | # define BN_DIV2W |
86 | | # endif |
87 | | |
88 | | /* |
89 | | * 64-bit processor with LP64 ABI |
90 | | */ |
91 | | # ifdef SIXTY_FOUR_BIT_LONG |
92 | | # define BN_ULLONG unsigned long long |
93 | 19.3k | # define BN_BITS4 32 |
94 | 117k | # define BN_MASK2 (0xffffffffffffffffL) |
95 | 19.3k | # define BN_MASK2l (0xffffffffL) |
96 | 0 | # define BN_MASK2h (0xffffffff00000000L) |
97 | | # define BN_MASK2h1 (0xffffffff80000000L) |
98 | 0 | # define BN_DEC_CONV (10000000000000000000UL) |
99 | 0 | # define BN_DEC_NUM 19 |
100 | 0 | # define BN_DEC_FMT1 "%lu" |
101 | 0 | # define BN_DEC_FMT2 "%019lu" |
102 | | # endif |
103 | | |
104 | | /* |
105 | | * 64-bit processor other than LP64 ABI |
106 | | */ |
107 | | # ifdef SIXTY_FOUR_BIT |
108 | | # undef BN_LLONG |
109 | | # undef BN_ULLONG |
110 | | # define BN_BITS4 32 |
111 | | # define BN_MASK2 (0xffffffffffffffffLL) |
112 | | # define BN_MASK2l (0xffffffffL) |
113 | | # define BN_MASK2h (0xffffffff00000000LL) |
114 | | # define BN_MASK2h1 (0xffffffff80000000LL) |
115 | | # define BN_DEC_CONV (10000000000000000000ULL) |
116 | | # define BN_DEC_NUM 19 |
117 | | # define BN_DEC_FMT1 "%llu" |
118 | | # define BN_DEC_FMT2 "%019llu" |
119 | | # endif |
120 | | |
121 | | # ifdef THIRTY_TWO_BIT |
122 | | # ifdef BN_LLONG |
123 | | # if defined(_WIN32) && !defined(__GNUC__) |
124 | | # define BN_ULLONG unsigned __int64 |
125 | | # else |
126 | | # define BN_ULLONG unsigned long long |
127 | | # endif |
128 | | # endif |
129 | | # define BN_BITS4 16 |
130 | | # define BN_MASK2 (0xffffffffL) |
131 | | # define BN_MASK2l (0xffff) |
132 | | # define BN_MASK2h1 (0xffff8000L) |
133 | | # define BN_MASK2h (0xffff0000L) |
134 | | # define BN_DEC_CONV (1000000000L) |
135 | | # define BN_DEC_NUM 9 |
136 | | # define BN_DEC_FMT1 "%u" |
137 | | # define BN_DEC_FMT2 "%09u" |
138 | | # endif |
139 | | |
140 | | |
141 | | /*- |
142 | | * Bignum consistency macros |
143 | | * There is one "API" macro, bn_fix_top(), for stripping leading zeroes from |
144 | | * bignum data after direct manipulations on the data. There is also an |
145 | | * "internal" macro, bn_check_top(), for verifying that there are no leading |
146 | | * zeroes. Unfortunately, some auditing is required due to the fact that |
147 | | * bn_fix_top() has become an overabused duct-tape because bignum data is |
148 | | * occasionally passed around in an inconsistent state. So the following |
149 | | * changes have been made to sort this out; |
150 | | * - bn_fix_top()s implementation has been moved to bn_correct_top() |
151 | | * - if BN_DEBUG isn't defined, bn_fix_top() maps to bn_correct_top(), and |
152 | | * bn_check_top() is as before. |
153 | | * - if BN_DEBUG *is* defined; |
154 | | * - bn_check_top() tries to pollute unused words even if the bignum 'top' is |
155 | | * consistent. (ed: only if BN_RAND_DEBUG is defined) |
156 | | * - bn_fix_top() maps to bn_check_top() rather than "fixing" anything. |
157 | | * The idea is to have debug builds flag up inconsistent bignums when they |
158 | | * occur. If that occurs in a bn_fix_top(), we examine the code in question; if |
159 | | * the use of bn_fix_top() was appropriate (ie. it follows directly after code |
160 | | * that manipulates the bignum) it is converted to bn_correct_top(), and if it |
161 | | * was not appropriate, we convert it permanently to bn_check_top() and track |
162 | | * down the cause of the bug. Eventually, no internal code should be using the |
163 | | * bn_fix_top() macro. External applications and libraries should try this with |
164 | | * their own code too, both in terms of building against the openssl headers |
165 | | * with BN_DEBUG defined *and* linking with a version of OpenSSL built with it |
166 | | * defined. This not only improves external code, it provides more test |
167 | | * coverage for openssl's own code. |
168 | | */ |
169 | | |
170 | | # ifdef BN_DEBUG |
171 | | /* |
172 | | * The new BN_FLG_FIXED_TOP flag marks vectors that were not treated with |
173 | | * bn_correct_top, in other words such vectors are permitted to have zeros |
174 | | * in most significant limbs. Such vectors are used internally to achieve |
175 | | * execution time invariance for critical operations with private keys. |
176 | | * It's BN_DEBUG-only flag, because user application is not supposed to |
177 | | * observe it anyway. Moreover, optimizing compiler would actually remove |
178 | | * all operations manipulating the bit in question in non-BN_DEBUG build. |
179 | | */ |
180 | | # define BN_FLG_FIXED_TOP 0x10000 |
181 | | # ifdef BN_RAND_DEBUG |
182 | | # define bn_pollute(a) \ |
183 | | do { \ |
184 | | const BIGNUM *_bnum1 = (a); \ |
185 | | if (_bnum1->top < _bnum1->dmax) { \ |
186 | | unsigned char _tmp_char; \ |
187 | | /* We cast away const without the compiler knowing, any \ |
188 | | * *genuinely* constant variables that aren't mutable \ |
189 | | * wouldn't be constructed with top!=dmax. */ \ |
190 | | BN_ULONG *_not_const; \ |
191 | | memcpy(&_not_const, &_bnum1->d, sizeof(_not_const)); \ |
192 | | (void)RAND_bytes(&_tmp_char, 1); /* Debug only - safe to ignore error return */\ |
193 | | memset(_not_const + _bnum1->top, _tmp_char, \ |
194 | | sizeof(*_not_const) * (_bnum1->dmax - _bnum1->top)); \ |
195 | | } \ |
196 | | } while(0) |
197 | | # else |
198 | | # define bn_pollute(a) |
199 | | # endif |
200 | | # define bn_check_top(a) \ |
201 | | do { \ |
202 | | const BIGNUM *_bnum2 = (a); \ |
203 | | if (_bnum2 != NULL) { \ |
204 | | int _top = _bnum2->top; \ |
205 | | (void)ossl_assert((_top == 0 && !_bnum2->neg) || \ |
206 | | (_top && ((_bnum2->flags & BN_FLG_FIXED_TOP) \ |
207 | | || _bnum2->d[_top - 1] != 0))); \ |
208 | | bn_pollute(_bnum2); \ |
209 | | } \ |
210 | | } while(0) |
211 | | |
212 | | # define bn_fix_top(a) bn_check_top(a) |
213 | | |
214 | | # define bn_check_size(bn, bits) bn_wcheck_size(bn, ((bits+BN_BITS2-1))/BN_BITS2) |
215 | | # define bn_wcheck_size(bn, words) \ |
216 | | do { \ |
217 | | const BIGNUM *_bnum2 = (bn); \ |
218 | | assert((words) <= (_bnum2)->dmax && \ |
219 | | (words) >= (_bnum2)->top); \ |
220 | | /* avoid unused variable warning with NDEBUG */ \ |
221 | | (void)(_bnum2); \ |
222 | | } while(0) |
223 | | |
224 | | # else /* !BN_DEBUG */ |
225 | | |
226 | 65.9k | # define BN_FLG_FIXED_TOP 0 |
227 | | # define bn_pollute(a) |
228 | | # define bn_check_top(a) |
229 | | # define bn_fix_top(a) bn_correct_top(a) |
230 | | # define bn_check_size(bn, bits) |
231 | | # define bn_wcheck_size(bn, words) |
232 | | |
233 | | # endif |
234 | | |
235 | | BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num, |
236 | | BN_ULONG w); |
237 | | BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w); |
238 | | void bn_sqr_words(BN_ULONG *rp, const BN_ULONG *ap, int num); |
239 | | BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d); |
240 | | BN_ULONG bn_add_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, |
241 | | int num); |
242 | | BN_ULONG bn_sub_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, |
243 | | int num); |
244 | | |
245 | | struct bignum_st { |
246 | | BN_ULONG *d; /* |
247 | | * Pointer to an array of 'BN_BITS2' bit |
248 | | * chunks. These chunks are organised in |
249 | | * a least significant chunk first order. |
250 | | */ |
251 | | int top; /* Index of last used d +1. */ |
252 | | /* The next are internal book keeping for bn_expand. */ |
253 | | int dmax; /* Size of the d array. */ |
254 | | int neg; /* one if the number is negative */ |
255 | | int flags; |
256 | | }; |
257 | | |
258 | | /* Used for montgomery multiplication */ |
259 | | struct bn_mont_ctx_st { |
260 | | int ri; /* number of bits in R */ |
261 | | BIGNUM RR; /* used to convert to montgomery form, |
262 | | possibly zero-padded */ |
263 | | BIGNUM N; /* The modulus */ |
264 | | BIGNUM Ni; /* R*(1/R mod N) - N*Ni = 1 (Ni is only |
265 | | * stored for bignum algorithm) */ |
266 | | BN_ULONG n0[2]; /* least significant word(s) of Ni; (type |
267 | | * changed with 0.9.9, was "BN_ULONG n0;" |
268 | | * before) */ |
269 | | int flags; |
270 | | }; |
271 | | |
272 | | /* |
273 | | * Used for reciprocal division/mod functions It cannot be shared between |
274 | | * threads |
275 | | */ |
276 | | struct bn_recp_ctx_st { |
277 | | BIGNUM N; /* the divisor */ |
278 | | BIGNUM Nr; /* the reciprocal */ |
279 | | int num_bits; |
280 | | int shift; |
281 | | int flags; |
282 | | }; |
283 | | |
284 | | /* Used for slow "generation" functions. */ |
285 | | struct bn_gencb_st { |
286 | | unsigned int ver; /* To handle binary (in)compatibility */ |
287 | | void *arg; /* callback-specific data */ |
288 | | union { |
289 | | /* if (ver==1) - handles old style callbacks */ |
290 | | void (*cb_1) (int, int, void *); |
291 | | /* if (ver==2) - new callback style */ |
292 | | int (*cb_2) (int, int, BN_GENCB *); |
293 | | } cb; |
294 | | }; |
295 | | |
296 | | /*- |
297 | | * BN_window_bits_for_exponent_size -- macro for sliding window mod_exp functions |
298 | | * |
299 | | * |
300 | | * For window size 'w' (w >= 2) and a random 'b' bits exponent, |
301 | | * the number of multiplications is a constant plus on average |
302 | | * |
303 | | * 2^(w-1) + (b-w)/(w+1); |
304 | | * |
305 | | * here 2^(w-1) is for precomputing the table (we actually need |
306 | | * entries only for windows that have the lowest bit set), and |
307 | | * (b-w)/(w+1) is an approximation for the expected number of |
308 | | * w-bit windows, not counting the first one. |
309 | | * |
310 | | * Thus we should use |
311 | | * |
312 | | * w >= 6 if b > 671 |
313 | | * w = 5 if 671 > b > 239 |
314 | | * w = 4 if 239 > b > 79 |
315 | | * w = 3 if 79 > b > 23 |
316 | | * w <= 2 if 23 > b |
317 | | * |
318 | | * (with draws in between). Very small exponents are often selected |
319 | | * with low Hamming weight, so we use w = 1 for b <= 23. |
320 | | */ |
321 | | # define BN_window_bits_for_exponent_size(b) \ |
322 | 0 | ((b) > 671 ? 6 : \ |
323 | 0 | (b) > 239 ? 5 : \ |
324 | 0 | (b) > 79 ? 4 : \ |
325 | 0 | (b) > 23 ? 3 : 1) |
326 | | |
327 | | /* |
328 | | * BN_mod_exp_mont_consttime is based on the assumption that the L1 data cache |
329 | | * line width of the target processor is at least the following value. |
330 | | */ |
331 | 340 | # define MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH ( 64 ) |
332 | 170 | # define MOD_EXP_CTIME_MIN_CACHE_LINE_MASK (MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH - 1) |
333 | | |
334 | | /* |
335 | | * Window sizes optimized for fixed window size modular exponentiation |
336 | | * algorithm (BN_mod_exp_mont_consttime). To achieve the security goals of |
337 | | * BN_mode_exp_mont_consttime, the maximum size of the window must not exceed |
338 | | * log_2(MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH). Window size thresholds are |
339 | | * defined for cache line sizes of 32 and 64, cache line sizes where |
340 | | * log_2(32)=5 and log_2(64)=6 respectively. A window size of 7 should only be |
341 | | * used on processors that have a 128 byte or greater cache line size. |
342 | | */ |
343 | | # if MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 64 |
344 | | |
345 | | # define BN_window_bits_for_ctime_exponent_size(b) \ |
346 | 170 | ((b) > 937 ? 6 : \ |
347 | 170 | (b) > 306 ? 5 : \ |
348 | 170 | (b) > 89 ? 4 : \ |
349 | 110 | (b) > 22 ? 3 : 1) |
350 | | # define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE (6) |
351 | | |
352 | | # elif MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 32 |
353 | | |
354 | | # define BN_window_bits_for_ctime_exponent_size(b) \ |
355 | | ((b) > 306 ? 5 : \ |
356 | | (b) > 89 ? 4 : \ |
357 | | (b) > 22 ? 3 : 1) |
358 | | # define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE (5) |
359 | | |
360 | | # endif |
361 | | |
362 | | /* Pentium pro 16,16,16,32,64 */ |
363 | | /* Alpha 16,16,16,16.64 */ |
364 | 187 | # define BN_MULL_SIZE_NORMAL (16)/* 32 */ |
365 | 0 | # define BN_MUL_RECURSIVE_SIZE_NORMAL (16)/* 32 less than */ |
366 | 20 | # define BN_SQR_RECURSIVE_SIZE_NORMAL (16)/* 32 */ |
367 | 0 | # define BN_MUL_LOW_RECURSIVE_SIZE_NORMAL (32)/* 32 */ |
368 | | # define BN_MONT_CTX_SET_SIZE_WORD (64)/* 32 */ |
369 | | |
370 | | # if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) && !defined(PEDANTIC) |
371 | | /* |
372 | | * BN_UMULT_HIGH section. |
373 | | * If the compiler doesn't support 2*N integer type, then you have to |
374 | | * replace every N*N multiplication with 4 (N/2)*(N/2) accompanied by some |
375 | | * shifts and additions which unavoidably results in severe performance |
376 | | * penalties. Of course provided that the hardware is capable of producing |
377 | | * 2*N result... That's when you normally start considering assembler |
378 | | * implementation. However! It should be pointed out that some CPUs (e.g., |
379 | | * PowerPC, Alpha, and IA-64) provide *separate* instruction calculating |
380 | | * the upper half of the product placing the result into a general |
381 | | * purpose register. Now *if* the compiler supports inline assembler, |
382 | | * then it's not impossible to implement the "bignum" routines (and have |
383 | | * the compiler optimize 'em) exhibiting "native" performance in C. That's |
384 | | * what BN_UMULT_HIGH macro is about:-) Note that more recent compilers do |
385 | | * support 2*64 integer type, which is also used here. |
386 | | */ |
387 | | # if defined(__SIZEOF_INT128__) && __SIZEOF_INT128__==16 && \ |
388 | | (defined(SIXTY_FOUR_BIT) || defined(SIXTY_FOUR_BIT_LONG)) |
389 | | # define BN_UMULT_HIGH(a,b) (((uint128_t)(a)*(b))>>64) |
390 | | # define BN_UMULT_LOHI(low,high,a,b) ({ \ |
391 | | uint128_t ret=(uint128_t)(a)*(b); \ |
392 | | (high)=ret>>64; (low)=ret; }) |
393 | | # elif defined(__alpha) && (defined(SIXTY_FOUR_BIT_LONG) || defined(SIXTY_FOUR_BIT)) |
394 | | # if defined(__DECC) |
395 | | # include <c_asm.h> |
396 | | # define BN_UMULT_HIGH(a,b) (BN_ULONG)asm("umulh %a0,%a1,%v0",(a),(b)) |
397 | | # elif defined(__GNUC__) && __GNUC__>=2 |
398 | | # define BN_UMULT_HIGH(a,b) ({ \ |
399 | | register BN_ULONG ret; \ |
400 | | asm ("umulh %1,%2,%0" \ |
401 | | : "=r"(ret) \ |
402 | | : "r"(a), "r"(b)); \ |
403 | | ret; }) |
404 | | # endif /* compiler */ |
405 | | # elif defined(_ARCH_PPC64) && defined(SIXTY_FOUR_BIT_LONG) |
406 | | # if defined(__GNUC__) && __GNUC__>=2 |
407 | | # define BN_UMULT_HIGH(a,b) ({ \ |
408 | | register BN_ULONG ret; \ |
409 | | asm ("mulhdu %0,%1,%2" \ |
410 | | : "=r"(ret) \ |
411 | | : "r"(a), "r"(b)); \ |
412 | | ret; }) |
413 | | # endif /* compiler */ |
414 | | # elif (defined(__x86_64) || defined(__x86_64__)) && \ |
415 | | (defined(SIXTY_FOUR_BIT_LONG) || defined(SIXTY_FOUR_BIT)) |
416 | | # if defined(__GNUC__) && __GNUC__>=2 |
417 | | # define BN_UMULT_HIGH(a,b) ({ \ |
418 | | register BN_ULONG ret,discard; \ |
419 | | asm ("mulq %3" \ |
420 | | : "=a"(discard),"=d"(ret) \ |
421 | | : "a"(a), "g"(b) \ |
422 | | : "cc"); \ |
423 | | ret; }) |
424 | | # define BN_UMULT_LOHI(low,high,a,b) \ |
425 | | asm ("mulq %3" \ |
426 | | : "=a"(low),"=d"(high) \ |
427 | | : "a"(a),"g"(b) \ |
428 | | : "cc"); |
429 | | # endif |
430 | | # elif (defined(_M_AMD64) || defined(_M_X64)) && defined(SIXTY_FOUR_BIT) |
431 | | # if defined(_MSC_VER) && _MSC_VER>=1400 |
432 | | unsigned __int64 __umulh(unsigned __int64 a, unsigned __int64 b); |
433 | | unsigned __int64 _umul128(unsigned __int64 a, unsigned __int64 b, |
434 | | unsigned __int64 *h); |
435 | | # pragma intrinsic(__umulh,_umul128) |
436 | | # define BN_UMULT_HIGH(a,b) __umulh((a),(b)) |
437 | | # define BN_UMULT_LOHI(low,high,a,b) ((low)=_umul128((a),(b),&(high))) |
438 | | # endif |
439 | | # elif defined(__mips) && (defined(SIXTY_FOUR_BIT) || defined(SIXTY_FOUR_BIT_LONG)) |
440 | | # if defined(__GNUC__) && __GNUC__>=2 |
441 | | # define BN_UMULT_HIGH(a,b) ({ \ |
442 | | register BN_ULONG ret; \ |
443 | | asm ("dmultu %1,%2" \ |
444 | | : "=h"(ret) \ |
445 | | : "r"(a), "r"(b) : "l"); \ |
446 | | ret; }) |
447 | | # define BN_UMULT_LOHI(low,high,a,b) \ |
448 | | asm ("dmultu %2,%3" \ |
449 | | : "=l"(low),"=h"(high) \ |
450 | | : "r"(a), "r"(b)); |
451 | | # endif |
452 | | # elif defined(__aarch64__) && defined(SIXTY_FOUR_BIT_LONG) |
453 | | # if defined(__GNUC__) && __GNUC__>=2 |
454 | | # define BN_UMULT_HIGH(a,b) ({ \ |
455 | | register BN_ULONG ret; \ |
456 | | asm ("umulh %0,%1,%2" \ |
457 | | : "=r"(ret) \ |
458 | | : "r"(a), "r"(b)); \ |
459 | | ret; }) |
460 | | # endif |
461 | | # endif /* cpu */ |
462 | | # endif /* OPENSSL_NO_ASM */ |
463 | | |
464 | | # ifdef BN_RAND_DEBUG |
465 | | # define bn_clear_top2max(a) \ |
466 | | { \ |
467 | | int ind = (a)->dmax - (a)->top; \ |
468 | | BN_ULONG *ftl = &(a)->d[(a)->top-1]; \ |
469 | | for (; ind != 0; ind--) \ |
470 | | *(++ftl) = 0x0; \ |
471 | | } |
472 | | # else |
473 | | # define bn_clear_top2max(a) |
474 | | # endif |
475 | | |
476 | | # ifdef BN_LLONG |
477 | | /******************************************************************* |
478 | | * Using the long long type, has to be twice as wide as BN_ULONG... |
479 | | */ |
480 | | # define Lw(t) (((BN_ULONG)(t))&BN_MASK2) |
481 | | # define Hw(t) (((BN_ULONG)((t)>>BN_BITS2))&BN_MASK2) |
482 | | |
483 | | # define mul_add(r,a,w,c) { \ |
484 | | BN_ULLONG t; \ |
485 | | t=(BN_ULLONG)w * (a) + (r) + (c); \ |
486 | | (r)= Lw(t); \ |
487 | | (c)= Hw(t); \ |
488 | | } |
489 | | |
490 | | # define mul(r,a,w,c) { \ |
491 | | BN_ULLONG t; \ |
492 | | t=(BN_ULLONG)w * (a) + (c); \ |
493 | | (r)= Lw(t); \ |
494 | | (c)= Hw(t); \ |
495 | | } |
496 | | |
497 | | # define sqr(r0,r1,a) { \ |
498 | | BN_ULLONG t; \ |
499 | | t=(BN_ULLONG)(a)*(a); \ |
500 | | (r0)=Lw(t); \ |
501 | | (r1)=Hw(t); \ |
502 | | } |
503 | | |
504 | | # elif defined(BN_UMULT_LOHI) |
505 | | # define mul_add(r,a,w,c) { \ |
506 | | BN_ULONG high,low,ret,tmp=(a); \ |
507 | | ret = (r); \ |
508 | | BN_UMULT_LOHI(low,high,w,tmp); \ |
509 | | ret += (c); \ |
510 | | (c) = (ret<(c)); \ |
511 | | (c) += high; \ |
512 | | ret += low; \ |
513 | | (c) += (ret<low); \ |
514 | | (r) = ret; \ |
515 | | } |
516 | | |
517 | | # define mul(r,a,w,c) { \ |
518 | | BN_ULONG high,low,ret,ta=(a); \ |
519 | | BN_UMULT_LOHI(low,high,w,ta); \ |
520 | | ret = low + (c); \ |
521 | | (c) = high; \ |
522 | | (c) += (ret<low); \ |
523 | | (r) = ret; \ |
524 | | } |
525 | | |
526 | | # define sqr(r0,r1,a) { \ |
527 | | BN_ULONG tmp=(a); \ |
528 | | BN_UMULT_LOHI(r0,r1,tmp,tmp); \ |
529 | | } |
530 | | |
531 | | # elif defined(BN_UMULT_HIGH) |
532 | | # define mul_add(r,a,w,c) { \ |
533 | | BN_ULONG high,low,ret,tmp=(a); \ |
534 | | ret = (r); \ |
535 | | high= BN_UMULT_HIGH(w,tmp); \ |
536 | | ret += (c); \ |
537 | | low = (w) * tmp; \ |
538 | | (c) = (ret<(c)); \ |
539 | | (c) += high; \ |
540 | | ret += low; \ |
541 | | (c) += (ret<low); \ |
542 | | (r) = ret; \ |
543 | | } |
544 | | |
545 | | # define mul(r,a,w,c) { \ |
546 | | BN_ULONG high,low,ret,ta=(a); \ |
547 | | low = (w) * ta; \ |
548 | | high= BN_UMULT_HIGH(w,ta); \ |
549 | | ret = low + (c); \ |
550 | | (c) = high; \ |
551 | | (c) += (ret<low); \ |
552 | | (r) = ret; \ |
553 | | } |
554 | | |
555 | | # define sqr(r0,r1,a) { \ |
556 | | BN_ULONG tmp=(a); \ |
557 | | (r0) = tmp * tmp; \ |
558 | | (r1) = BN_UMULT_HIGH(tmp,tmp); \ |
559 | | } |
560 | | |
561 | | # else |
562 | | /************************************************************* |
563 | | * No long long type |
564 | | */ |
565 | | |
566 | 7.73k | # define LBITS(a) ((a)&BN_MASK2l) |
567 | 11.6k | # define HBITS(a) (((a)>>BN_BITS4)&BN_MASK2l) |
568 | 7.73k | # define L2HBITS(a) (((a)<<BN_BITS4)&BN_MASK2) |
569 | | |
570 | | # define LLBITS(a) ((a)&BN_MASKl) |
571 | | # define LHBITS(a) (((a)>>BN_BITS2)&BN_MASKl) |
572 | | # define LL2HBITS(a) ((BN_ULLONG)((a)&BN_MASKl)<<BN_BITS2) |
573 | | |
574 | | # define mul64(l,h,bl,bh) \ |
575 | 3.86k | { \ |
576 | 3.86k | BN_ULONG m,m1,lt,ht; \ |
577 | 3.86k | \ |
578 | 3.86k | lt=l; \ |
579 | 3.86k | ht=h; \ |
580 | 3.86k | m =(bh)*(lt); \ |
581 | 3.86k | lt=(bl)*(lt); \ |
582 | 3.86k | m1=(bl)*(ht); \ |
583 | 3.86k | ht =(bh)*(ht); \ |
584 | 3.86k | m=(m+m1)&BN_MASK2; ht += L2HBITS((BN_ULONG)(m < m1)); \ |
585 | 3.86k | ht+=HBITS(m); \ |
586 | 3.86k | m1=L2HBITS(m); \ |
587 | 3.86k | lt=(lt+m1)&BN_MASK2; ht += (lt < m1); \ |
588 | 3.86k | (l)=lt; \ |
589 | 3.86k | (h)=ht; \ |
590 | 3.86k | } |
591 | | |
592 | | # define sqr64(lo,ho,in) \ |
593 | | { \ |
594 | | BN_ULONG l,h,m; \ |
595 | | \ |
596 | | h=(in); \ |
597 | | l=LBITS(h); \ |
598 | | h=HBITS(h); \ |
599 | | m =(l)*(h); \ |
600 | | l*=l; \ |
601 | | h*=h; \ |
602 | | h+=(m&BN_MASK2h1)>>(BN_BITS4-1); \ |
603 | | m =(m&BN_MASK2l)<<(BN_BITS4+1); \ |
604 | | l=(l+m)&BN_MASK2; h += (l < m); \ |
605 | | (lo)=l; \ |
606 | | (ho)=h; \ |
607 | | } |
608 | | |
609 | | # define mul_add(r,a,bl,bh,c) { \ |
610 | | BN_ULONG l,h; \ |
611 | | \ |
612 | | h= (a); \ |
613 | | l=LBITS(h); \ |
614 | | h=HBITS(h); \ |
615 | | mul64(l,h,(bl),(bh)); \ |
616 | | \ |
617 | | /* non-multiply part */ \ |
618 | | l=(l+(c))&BN_MASK2; h += (l < (c)); \ |
619 | | (c)=(r); \ |
620 | | l=(l+(c))&BN_MASK2; h += (l < (c)); \ |
621 | | (c)=h&BN_MASK2; \ |
622 | | (r)=l; \ |
623 | | } |
624 | | |
625 | | # define mul(r,a,bl,bh,c) { \ |
626 | | BN_ULONG l,h; \ |
627 | | \ |
628 | | h= (a); \ |
629 | | l=LBITS(h); \ |
630 | | h=HBITS(h); \ |
631 | | mul64(l,h,(bl),(bh)); \ |
632 | | \ |
633 | | /* non-multiply part */ \ |
634 | | l+=(c); h += ((l&BN_MASK2) < (c)); \ |
635 | | (c)=h&BN_MASK2; \ |
636 | | (r)=l&BN_MASK2; \ |
637 | | } |
638 | | # endif /* !BN_LLONG */ |
639 | | |
640 | | void BN_RECP_CTX_init(BN_RECP_CTX *recp); |
641 | | void BN_MONT_CTX_init(BN_MONT_CTX *ctx); |
642 | | |
643 | | void bn_init(BIGNUM *a); |
644 | | void bn_mul_normal(BN_ULONG *r, BN_ULONG *a, int na, BN_ULONG *b, int nb); |
645 | | void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b); |
646 | | void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b); |
647 | | void bn_sqr_normal(BN_ULONG *r, const BN_ULONG *a, int n, BN_ULONG *tmp); |
648 | | void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a); |
649 | | void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a); |
650 | | int bn_cmp_words(const BN_ULONG *a, const BN_ULONG *b, int n); |
651 | | int bn_cmp_part_words(const BN_ULONG *a, const BN_ULONG *b, int cl, int dl); |
652 | | void bn_mul_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2, |
653 | | int dna, int dnb, BN_ULONG *t); |
654 | | void bn_mul_part_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, |
655 | | int n, int tna, int tnb, BN_ULONG *t); |
656 | | void bn_sqr_recursive(BN_ULONG *r, const BN_ULONG *a, int n2, BN_ULONG *t); |
657 | | void bn_mul_low_normal(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n); |
658 | | void bn_mul_low_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2, |
659 | | BN_ULONG *t); |
660 | | BN_ULONG bn_sub_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, |
661 | | int cl, int dl); |
662 | | int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, |
663 | | const BN_ULONG *np, const BN_ULONG *n0, int num); |
664 | | void bn_correct_top_consttime(BIGNUM *a); |
665 | | BIGNUM *int_bn_mod_inverse(BIGNUM *in, |
666 | | const BIGNUM *a, const BIGNUM *n, BN_CTX *ctx, |
667 | | int *noinv); |
668 | | |
669 | | static ossl_inline BIGNUM *bn_expand(BIGNUM *a, int bits) |
670 | 305 | { |
671 | 305 | if (bits > (INT_MAX - BN_BITS2 + 1)) |
672 | 0 | return NULL; |
673 | | |
674 | 305 | if (((bits+BN_BITS2-1)/BN_BITS2) <= (a)->dmax) |
675 | 17 | return a; |
676 | | |
677 | 288 | return bn_expand2((a),(bits+BN_BITS2-1)/BN_BITS2); |
678 | 305 | } Unexecuted instantiation: bn_dh.c:bn_expand Line | Count | Source | 670 | 305 | { | 671 | 305 | if (bits > (INT_MAX - BN_BITS2 + 1)) | 672 | 0 | return NULL; | 673 | | | 674 | 305 | if (((bits+BN_BITS2-1)/BN_BITS2) <= (a)->dmax) | 675 | 17 | return a; | 676 | | | 677 | 288 | return bn_expand2((a),(bits+BN_BITS2-1)/BN_BITS2); | 678 | 305 | } |
Unexecuted instantiation: bn_mont.c:bn_expand Unexecuted instantiation: bn_mul.c:bn_expand Unexecuted instantiation: bn_shift.c:bn_expand Unexecuted instantiation: bn_sqr.c:bn_expand Unexecuted instantiation: bn_word.c:bn_expand Unexecuted instantiation: x86_64-gcc.c:bn_expand Unexecuted instantiation: bn_add.c:bn_expand Unexecuted instantiation: bn_blind.c:bn_expand Unexecuted instantiation: bn_conv.c:bn_expand Unexecuted instantiation: bn_ctx.c:bn_expand Unexecuted instantiation: bn_div.c:bn_expand Unexecuted instantiation: bn_exp.c:bn_expand Unexecuted instantiation: bn_exp2.c:bn_expand Unexecuted instantiation: bn_gcd.c:bn_expand Unexecuted instantiation: bn_intern.c:bn_expand Unexecuted instantiation: bn_kron.c:bn_expand Unexecuted instantiation: bn_mod.c:bn_expand Unexecuted instantiation: bn_nist.c:bn_expand Unexecuted instantiation: bn_prime.c:bn_expand Unexecuted instantiation: bn_print.c:bn_expand Unexecuted instantiation: bn_rand.c:bn_expand Unexecuted instantiation: bn_recp.c:bn_expand Unexecuted instantiation: bn_rsa_fips186_4.c:bn_expand Unexecuted instantiation: bn_sqrt.c:bn_expand Unexecuted instantiation: bn_srp.c:bn_expand Unexecuted instantiation: rsaz_exp.c:bn_expand Unexecuted instantiation: rsaz_exp_x2.c:bn_expand Unexecuted instantiation: bn_gf2m.c:bn_expand |
679 | | |
680 | | int ossl_bn_check_prime(const BIGNUM *w, int checks, BN_CTX *ctx, |
681 | | int do_trial_division, BN_GENCB *cb); |
682 | | |
683 | | #endif |