1 | /*
|
---|
2 | * Copyright 1995-2016 The OpenSSL Project Authors. All Rights Reserved.
|
---|
3 | *
|
---|
4 | * Licensed under the OpenSSL license (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 HEADER_BN_LCL_H
|
---|
11 | # define HEADER_BN_LCL_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 "internal/bn_conf.h"
|
---|
22 | # endif
|
---|
23 |
|
---|
24 | # include "internal/bn_int.h"
|
---|
25 |
|
---|
26 | #ifdef __cplusplus
|
---|
27 | extern "C" {
|
---|
28 | #endif
|
---|
29 |
|
---|
30 | /*
|
---|
31 | * These preprocessor symbols control various aspects of the bignum headers
|
---|
32 | * and library code. They're not defined by any "normal" configuration, as
|
---|
33 | * they are intended for development and testing purposes. NB: defining all
|
---|
34 | * three can be useful for debugging application code as well as openssl
|
---|
35 | * itself. BN_DEBUG - turn on various debugging alterations to the bignum
|
---|
36 | * code BN_DEBUG_RAND - uses random poisoning of unused words to trip up
|
---|
37 | * mismanagement of bignum internals. You must also define BN_DEBUG.
|
---|
38 | */
|
---|
39 | /* #define BN_DEBUG */
|
---|
40 | /* #define BN_DEBUG_RAND */
|
---|
41 |
|
---|
42 | # ifndef OPENSSL_SMALL_FOOTPRINT
|
---|
43 | # define BN_MUL_COMBA
|
---|
44 | # define BN_SQR_COMBA
|
---|
45 | # define BN_RECURSION
|
---|
46 | # endif
|
---|
47 |
|
---|
48 | /*
|
---|
49 | * This next option uses the C libraries (2 word)/(1 word) function. If it is
|
---|
50 | * not defined, I use my C version (which is slower). The reason for this
|
---|
51 | * flag is that when the particular C compiler library routine is used, and
|
---|
52 | * the library is linked with a different compiler, the library is missing.
|
---|
53 | * This mostly happens when the library is built with gcc and then linked
|
---|
54 | * using normal cc. This would be a common occurrence because gcc normally
|
---|
55 | * produces code that is 2 times faster than system compilers for the big
|
---|
56 | * number stuff. For machines with only one compiler (or shared libraries),
|
---|
57 | * this should be on. Again this in only really a problem on machines using
|
---|
58 | * "long long's", are 32bit, and are not using my assembler code.
|
---|
59 | */
|
---|
60 | # if defined(OPENSSL_SYS_MSDOS) || defined(OPENSSL_SYS_WINDOWS) || \
|
---|
61 | defined(OPENSSL_SYS_WIN32) || defined(linux)
|
---|
62 | # define BN_DIV2W
|
---|
63 | # endif
|
---|
64 |
|
---|
65 | /*
|
---|
66 | * 64-bit processor with LP64 ABI
|
---|
67 | */
|
---|
68 | # ifdef SIXTY_FOUR_BIT_LONG
|
---|
69 | # define BN_ULLONG unsigned long long
|
---|
70 | # define BN_BITS4 32
|
---|
71 | # define BN_MASK2 (0xffffffffffffffffL)
|
---|
72 | # define BN_MASK2l (0xffffffffL)
|
---|
73 | # define BN_MASK2h (0xffffffff00000000L)
|
---|
74 | # define BN_MASK2h1 (0xffffffff80000000L)
|
---|
75 | # define BN_DEC_CONV (10000000000000000000UL)
|
---|
76 | # define BN_DEC_NUM 19
|
---|
77 | # define BN_DEC_FMT1 "%lu"
|
---|
78 | # define BN_DEC_FMT2 "%019lu"
|
---|
79 | # endif
|
---|
80 |
|
---|
81 | /*
|
---|
82 | * 64-bit processor other than LP64 ABI
|
---|
83 | */
|
---|
84 | # ifdef SIXTY_FOUR_BIT
|
---|
85 | # undef BN_LLONG
|
---|
86 | # undef BN_ULLONG
|
---|
87 | # define BN_BITS4 32
|
---|
88 | # define BN_MASK2 (0xffffffffffffffffLL)
|
---|
89 | # define BN_MASK2l (0xffffffffL)
|
---|
90 | # define BN_MASK2h (0xffffffff00000000LL)
|
---|
91 | # define BN_MASK2h1 (0xffffffff80000000LL)
|
---|
92 | # define BN_DEC_CONV (10000000000000000000ULL)
|
---|
93 | # define BN_DEC_NUM 19
|
---|
94 | # define BN_DEC_FMT1 "%llu"
|
---|
95 | # define BN_DEC_FMT2 "%019llu"
|
---|
96 | # endif
|
---|
97 |
|
---|
98 | # ifdef THIRTY_TWO_BIT
|
---|
99 | # ifdef BN_LLONG
|
---|
100 | # if defined(_WIN32) && !defined(__GNUC__)
|
---|
101 | # define BN_ULLONG unsigned __int64
|
---|
102 | # else
|
---|
103 | # define BN_ULLONG unsigned long long
|
---|
104 | # endif
|
---|
105 | # endif
|
---|
106 | # define BN_BITS4 16
|
---|
107 | # define BN_MASK2 (0xffffffffL)
|
---|
108 | # define BN_MASK2l (0xffff)
|
---|
109 | # define BN_MASK2h1 (0xffff8000L)
|
---|
110 | # define BN_MASK2h (0xffff0000L)
|
---|
111 | # define BN_DEC_CONV (1000000000L)
|
---|
112 | # define BN_DEC_NUM 9
|
---|
113 | # define BN_DEC_FMT1 "%u"
|
---|
114 | # define BN_DEC_FMT2 "%09u"
|
---|
115 | # endif
|
---|
116 |
|
---|
117 |
|
---|
118 | /*-
|
---|
119 | * Bignum consistency macros
|
---|
120 | * There is one "API" macro, bn_fix_top(), for stripping leading zeroes from
|
---|
121 | * bignum data after direct manipulations on the data. There is also an
|
---|
122 | * "internal" macro, bn_check_top(), for verifying that there are no leading
|
---|
123 | * zeroes. Unfortunately, some auditing is required due to the fact that
|
---|
124 | * bn_fix_top() has become an overabused duct-tape because bignum data is
|
---|
125 | * occasionally passed around in an inconsistent state. So the following
|
---|
126 | * changes have been made to sort this out;
|
---|
127 | * - bn_fix_top()s implementation has been moved to bn_correct_top()
|
---|
128 | * - if BN_DEBUG isn't defined, bn_fix_top() maps to bn_correct_top(), and
|
---|
129 | * bn_check_top() is as before.
|
---|
130 | * - if BN_DEBUG *is* defined;
|
---|
131 | * - bn_check_top() tries to pollute unused words even if the bignum 'top' is
|
---|
132 | * consistent. (ed: only if BN_DEBUG_RAND is defined)
|
---|
133 | * - bn_fix_top() maps to bn_check_top() rather than "fixing" anything.
|
---|
134 | * The idea is to have debug builds flag up inconsistent bignums when they
|
---|
135 | * occur. If that occurs in a bn_fix_top(), we examine the code in question; if
|
---|
136 | * the use of bn_fix_top() was appropriate (ie. it follows directly after code
|
---|
137 | * that manipulates the bignum) it is converted to bn_correct_top(), and if it
|
---|
138 | * was not appropriate, we convert it permanently to bn_check_top() and track
|
---|
139 | * down the cause of the bug. Eventually, no internal code should be using the
|
---|
140 | * bn_fix_top() macro. External applications and libraries should try this with
|
---|
141 | * their own code too, both in terms of building against the openssl headers
|
---|
142 | * with BN_DEBUG defined *and* linking with a version of OpenSSL built with it
|
---|
143 | * defined. This not only improves external code, it provides more test
|
---|
144 | * coverage for openssl's own code.
|
---|
145 | */
|
---|
146 |
|
---|
147 | # ifdef BN_DEBUG
|
---|
148 |
|
---|
149 | # ifdef BN_DEBUG_RAND
|
---|
150 | /* To avoid "make update" cvs wars due to BN_DEBUG, use some tricks */
|
---|
151 | # ifndef RAND_bytes
|
---|
152 | int RAND_bytes(unsigned char *buf, int num);
|
---|
153 | # define BN_DEBUG_TRIX
|
---|
154 | # endif
|
---|
155 | # define bn_pollute(a) \
|
---|
156 | do { \
|
---|
157 | const BIGNUM *_bnum1 = (a); \
|
---|
158 | if (_bnum1->top < _bnum1->dmax) { \
|
---|
159 | unsigned char _tmp_char; \
|
---|
160 | /* We cast away const without the compiler knowing, any \
|
---|
161 | * *genuinely* constant variables that aren't mutable \
|
---|
162 | * wouldn't be constructed with top!=dmax. */ \
|
---|
163 | BN_ULONG *_not_const; \
|
---|
164 | memcpy(&_not_const, &_bnum1->d, sizeof(_not_const)); \
|
---|
165 | RAND_bytes(&_tmp_char, 1); /* Debug only - safe to ignore error return */\
|
---|
166 | memset(_not_const + _bnum1->top, _tmp_char, \
|
---|
167 | sizeof(*_not_const) * (_bnum1->dmax - _bnum1->top)); \
|
---|
168 | } \
|
---|
169 | } while(0)
|
---|
170 | # ifdef BN_DEBUG_TRIX
|
---|
171 | # undef RAND_bytes
|
---|
172 | # endif
|
---|
173 | # else
|
---|
174 | # define bn_pollute(a)
|
---|
175 | # endif
|
---|
176 | # define bn_check_top(a) \
|
---|
177 | do { \
|
---|
178 | const BIGNUM *_bnum2 = (a); \
|
---|
179 | if (_bnum2 != NULL) { \
|
---|
180 | OPENSSL_assert(((_bnum2->top == 0) && !_bnum2->neg) || \
|
---|
181 | (_bnum2->top && (_bnum2->d[_bnum2->top - 1] != 0))); \
|
---|
182 | bn_pollute(_bnum2); \
|
---|
183 | } \
|
---|
184 | } while(0)
|
---|
185 |
|
---|
186 | # define bn_fix_top(a) bn_check_top(a)
|
---|
187 |
|
---|
188 | # define bn_check_size(bn, bits) bn_wcheck_size(bn, ((bits+BN_BITS2-1))/BN_BITS2)
|
---|
189 | # define bn_wcheck_size(bn, words) \
|
---|
190 | do { \
|
---|
191 | const BIGNUM *_bnum2 = (bn); \
|
---|
192 | OPENSSL_assert((words) <= (_bnum2)->dmax && \
|
---|
193 | (words) >= (_bnum2)->top); \
|
---|
194 | /* avoid unused variable warning with NDEBUG */ \
|
---|
195 | (void)(_bnum2); \
|
---|
196 | } while(0)
|
---|
197 |
|
---|
198 | # else /* !BN_DEBUG */
|
---|
199 |
|
---|
200 | # define bn_pollute(a)
|
---|
201 | # define bn_check_top(a)
|
---|
202 | # define bn_fix_top(a) bn_correct_top(a)
|
---|
203 | # define bn_check_size(bn, bits)
|
---|
204 | # define bn_wcheck_size(bn, words)
|
---|
205 |
|
---|
206 | # endif
|
---|
207 |
|
---|
208 | BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num,
|
---|
209 | BN_ULONG w);
|
---|
210 | BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w);
|
---|
211 | void bn_sqr_words(BN_ULONG *rp, const BN_ULONG *ap, int num);
|
---|
212 | BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d);
|
---|
213 | BN_ULONG bn_add_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
|
---|
214 | int num);
|
---|
215 | BN_ULONG bn_sub_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
|
---|
216 | int num);
|
---|
217 |
|
---|
218 | struct bignum_st {
|
---|
219 | BN_ULONG *d; /* Pointer to an array of 'BN_BITS2' bit
|
---|
220 | * chunks. */
|
---|
221 | int top; /* Index of last used d +1. */
|
---|
222 | /* The next are internal book keeping for bn_expand. */
|
---|
223 | int dmax; /* Size of the d array. */
|
---|
224 | int neg; /* one if the number is negative */
|
---|
225 | int flags;
|
---|
226 | };
|
---|
227 |
|
---|
228 | /* Used for montgomery multiplication */
|
---|
229 | struct bn_mont_ctx_st {
|
---|
230 | int ri; /* number of bits in R */
|
---|
231 | BIGNUM RR; /* used to convert to montgomery form */
|
---|
232 | BIGNUM N; /* The modulus */
|
---|
233 | BIGNUM Ni; /* R*(1/R mod N) - N*Ni = 1 (Ni is only
|
---|
234 | * stored for bignum algorithm) */
|
---|
235 | BN_ULONG n0[2]; /* least significant word(s) of Ni; (type
|
---|
236 | * changed with 0.9.9, was "BN_ULONG n0;"
|
---|
237 | * before) */
|
---|
238 | int flags;
|
---|
239 | };
|
---|
240 |
|
---|
241 | /*
|
---|
242 | * Used for reciprocal division/mod functions It cannot be shared between
|
---|
243 | * threads
|
---|
244 | */
|
---|
245 | struct bn_recp_ctx_st {
|
---|
246 | BIGNUM N; /* the divisor */
|
---|
247 | BIGNUM Nr; /* the reciprocal */
|
---|
248 | int num_bits;
|
---|
249 | int shift;
|
---|
250 | int flags;
|
---|
251 | };
|
---|
252 |
|
---|
253 | /* Used for slow "generation" functions. */
|
---|
254 | struct bn_gencb_st {
|
---|
255 | unsigned int ver; /* To handle binary (in)compatibility */
|
---|
256 | void *arg; /* callback-specific data */
|
---|
257 | union {
|
---|
258 | /* if (ver==1) - handles old style callbacks */
|
---|
259 | void (*cb_1) (int, int, void *);
|
---|
260 | /* if (ver==2) - new callback style */
|
---|
261 | int (*cb_2) (int, int, BN_GENCB *);
|
---|
262 | } cb;
|
---|
263 | };
|
---|
264 |
|
---|
265 | /*-
|
---|
266 | * BN_window_bits_for_exponent_size -- macro for sliding window mod_exp functions
|
---|
267 | *
|
---|
268 | *
|
---|
269 | * For window size 'w' (w >= 2) and a random 'b' bits exponent,
|
---|
270 | * the number of multiplications is a constant plus on average
|
---|
271 | *
|
---|
272 | * 2^(w-1) + (b-w)/(w+1);
|
---|
273 | *
|
---|
274 | * here 2^(w-1) is for precomputing the table (we actually need
|
---|
275 | * entries only for windows that have the lowest bit set), and
|
---|
276 | * (b-w)/(w+1) is an approximation for the expected number of
|
---|
277 | * w-bit windows, not counting the first one.
|
---|
278 | *
|
---|
279 | * Thus we should use
|
---|
280 | *
|
---|
281 | * w >= 6 if b > 671
|
---|
282 | * w = 5 if 671 > b > 239
|
---|
283 | * w = 4 if 239 > b > 79
|
---|
284 | * w = 3 if 79 > b > 23
|
---|
285 | * w <= 2 if 23 > b
|
---|
286 | *
|
---|
287 | * (with draws in between). Very small exponents are often selected
|
---|
288 | * with low Hamming weight, so we use w = 1 for b <= 23.
|
---|
289 | */
|
---|
290 | # define BN_window_bits_for_exponent_size(b) \
|
---|
291 | ((b) > 671 ? 6 : \
|
---|
292 | (b) > 239 ? 5 : \
|
---|
293 | (b) > 79 ? 4 : \
|
---|
294 | (b) > 23 ? 3 : 1)
|
---|
295 |
|
---|
296 | /*
|
---|
297 | * BN_mod_exp_mont_conttime is based on the assumption that the L1 data cache
|
---|
298 | * line width of the target processor is at least the following value.
|
---|
299 | */
|
---|
300 | # define MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH ( 64 )
|
---|
301 | # define MOD_EXP_CTIME_MIN_CACHE_LINE_MASK (MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH - 1)
|
---|
302 |
|
---|
303 | /*
|
---|
304 | * Window sizes optimized for fixed window size modular exponentiation
|
---|
305 | * algorithm (BN_mod_exp_mont_consttime). To achieve the security goals of
|
---|
306 | * BN_mode_exp_mont_consttime, the maximum size of the window must not exceed
|
---|
307 | * log_2(MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH). Window size thresholds are
|
---|
308 | * defined for cache line sizes of 32 and 64, cache line sizes where
|
---|
309 | * log_2(32)=5 and log_2(64)=6 respectively. A window size of 7 should only be
|
---|
310 | * used on processors that have a 128 byte or greater cache line size.
|
---|
311 | */
|
---|
312 | # if MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 64
|
---|
313 |
|
---|
314 | # define BN_window_bits_for_ctime_exponent_size(b) \
|
---|
315 | ((b) > 937 ? 6 : \
|
---|
316 | (b) > 306 ? 5 : \
|
---|
317 | (b) > 89 ? 4 : \
|
---|
318 | (b) > 22 ? 3 : 1)
|
---|
319 | # define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE (6)
|
---|
320 |
|
---|
321 | # elif MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 32
|
---|
322 |
|
---|
323 | # define BN_window_bits_for_ctime_exponent_size(b) \
|
---|
324 | ((b) > 306 ? 5 : \
|
---|
325 | (b) > 89 ? 4 : \
|
---|
326 | (b) > 22 ? 3 : 1)
|
---|
327 | # define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE (5)
|
---|
328 |
|
---|
329 | # endif
|
---|
330 |
|
---|
331 | /* Pentium pro 16,16,16,32,64 */
|
---|
332 | /* Alpha 16,16,16,16.64 */
|
---|
333 | # define BN_MULL_SIZE_NORMAL (16)/* 32 */
|
---|
334 | # define BN_MUL_RECURSIVE_SIZE_NORMAL (16)/* 32 less than */
|
---|
335 | # define BN_SQR_RECURSIVE_SIZE_NORMAL (16)/* 32 */
|
---|
336 | # define BN_MUL_LOW_RECURSIVE_SIZE_NORMAL (32)/* 32 */
|
---|
337 | # define BN_MONT_CTX_SET_SIZE_WORD (64)/* 32 */
|
---|
338 |
|
---|
339 | /*
|
---|
340 | * 2011-02-22 SMS. In various places, a size_t variable or a type cast to
|
---|
341 | * size_t was used to perform integer-only operations on pointers. This
|
---|
342 | * failed on VMS with 64-bit pointers (CC /POINTER_SIZE = 64) because size_t
|
---|
343 | * is still only 32 bits. What's needed in these cases is an integer type
|
---|
344 | * with the same size as a pointer, which size_t is not certain to be. The
|
---|
345 | * only fix here is VMS-specific.
|
---|
346 | */
|
---|
347 | # if defined(OPENSSL_SYS_VMS)
|
---|
348 | # if __INITIAL_POINTER_SIZE == 64
|
---|
349 | # define PTR_SIZE_INT long long
|
---|
350 | # else /* __INITIAL_POINTER_SIZE == 64 */
|
---|
351 | # define PTR_SIZE_INT int
|
---|
352 | # endif /* __INITIAL_POINTER_SIZE == 64 [else] */
|
---|
353 | # elif !defined(PTR_SIZE_INT) /* defined(OPENSSL_SYS_VMS) */
|
---|
354 | # define PTR_SIZE_INT size_t
|
---|
355 | # endif /* defined(OPENSSL_SYS_VMS) [else] */
|
---|
356 |
|
---|
357 | # if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) && !defined(PEDANTIC)
|
---|
358 | /*
|
---|
359 | * BN_UMULT_HIGH section.
|
---|
360 | *
|
---|
361 | * No, I'm not trying to overwhelm you when stating that the
|
---|
362 | * product of N-bit numbers is 2*N bits wide:-) No, I don't expect
|
---|
363 | * you to be impressed when I say that if the compiler doesn't
|
---|
364 | * support 2*N integer type, then you have to replace every N*N
|
---|
365 | * multiplication with 4 (N/2)*(N/2) accompanied by some shifts
|
---|
366 | * and additions which unavoidably results in severe performance
|
---|
367 | * penalties. Of course provided that the hardware is capable of
|
---|
368 | * producing 2*N result... That's when you normally start
|
---|
369 | * considering assembler implementation. However! It should be
|
---|
370 | * pointed out that some CPUs (most notably Alpha, PowerPC and
|
---|
371 | * upcoming IA-64 family:-) provide *separate* instruction
|
---|
372 | * calculating the upper half of the product placing the result
|
---|
373 | * into a general purpose register. Now *if* the compiler supports
|
---|
374 | * inline assembler, then it's not impossible to implement the
|
---|
375 | * "bignum" routines (and have the compiler optimize 'em)
|
---|
376 | * exhibiting "native" performance in C. That's what BN_UMULT_HIGH
|
---|
377 | * macro is about:-)
|
---|
378 | *
|
---|
379 | * <appro@fy.chalmers.se>
|
---|
380 | */
|
---|
381 | # if defined(__alpha) && (defined(SIXTY_FOUR_BIT_LONG) || defined(SIXTY_FOUR_BIT))
|
---|
382 | # if defined(__DECC)
|
---|
383 | # include <c_asm.h>
|
---|
384 | # define BN_UMULT_HIGH(a,b) (BN_ULONG)asm("umulh %a0,%a1,%v0",(a),(b))
|
---|
385 | # elif defined(__GNUC__) && __GNUC__>=2
|
---|
386 | # define BN_UMULT_HIGH(a,b) ({ \
|
---|
387 | register BN_ULONG ret; \
|
---|
388 | asm ("umulh %1,%2,%0" \
|
---|
389 | : "=r"(ret) \
|
---|
390 | : "r"(a), "r"(b)); \
|
---|
391 | ret; })
|
---|
392 | # endif /* compiler */
|
---|
393 | # elif defined(_ARCH_PPC) && defined(__64BIT__) && defined(SIXTY_FOUR_BIT_LONG)
|
---|
394 | # if defined(__GNUC__) && __GNUC__>=2
|
---|
395 | # define BN_UMULT_HIGH(a,b) ({ \
|
---|
396 | register BN_ULONG ret; \
|
---|
397 | asm ("mulhdu %0,%1,%2" \
|
---|
398 | : "=r"(ret) \
|
---|
399 | : "r"(a), "r"(b)); \
|
---|
400 | ret; })
|
---|
401 | # endif /* compiler */
|
---|
402 | # elif (defined(__x86_64) || defined(__x86_64__)) && \
|
---|
403 | (defined(SIXTY_FOUR_BIT_LONG) || defined(SIXTY_FOUR_BIT))
|
---|
404 | # if defined(__GNUC__) && __GNUC__>=2
|
---|
405 | # define BN_UMULT_HIGH(a,b) ({ \
|
---|
406 | register BN_ULONG ret,discard; \
|
---|
407 | asm ("mulq %3" \
|
---|
408 | : "=a"(discard),"=d"(ret) \
|
---|
409 | : "a"(a), "g"(b) \
|
---|
410 | : "cc"); \
|
---|
411 | ret; })
|
---|
412 | # define BN_UMULT_LOHI(low,high,a,b) \
|
---|
413 | asm ("mulq %3" \
|
---|
414 | : "=a"(low),"=d"(high) \
|
---|
415 | : "a"(a),"g"(b) \
|
---|
416 | : "cc");
|
---|
417 | # endif
|
---|
418 | # elif (defined(_M_AMD64) || defined(_M_X64)) && defined(SIXTY_FOUR_BIT)
|
---|
419 | # if defined(_MSC_VER) && _MSC_VER>=1400
|
---|
420 | unsigned __int64 __umulh(unsigned __int64 a, unsigned __int64 b);
|
---|
421 | unsigned __int64 _umul128(unsigned __int64 a, unsigned __int64 b,
|
---|
422 | unsigned __int64 *h);
|
---|
423 | # pragma intrinsic(__umulh,_umul128)
|
---|
424 | # define BN_UMULT_HIGH(a,b) __umulh((a),(b))
|
---|
425 | # define BN_UMULT_LOHI(low,high,a,b) ((low)=_umul128((a),(b),&(high)))
|
---|
426 | # endif
|
---|
427 | # elif defined(__mips) && (defined(SIXTY_FOUR_BIT) || defined(SIXTY_FOUR_BIT_LONG))
|
---|
428 | # if defined(__GNUC__) && __GNUC__>=2
|
---|
429 | # if defined(__SIZEOF_INT128__) && __SIZEOF_INT128__==16
|
---|
430 | /* "h" constraint is not an option on R6 and was removed in 4.4 */
|
---|
431 | # define BN_UMULT_HIGH(a,b) (((__uint128_t)(a)*(b))>>64)
|
---|
432 | # define BN_UMULT_LOHI(low,high,a,b) ({ \
|
---|
433 | __uint128_t ret=(__uint128_t)(a)*(b); \
|
---|
434 | (high)=ret>>64; (low)=ret; })
|
---|
435 | # else
|
---|
436 | # define BN_UMULT_HIGH(a,b) ({ \
|
---|
437 | register BN_ULONG ret; \
|
---|
438 | asm ("dmultu %1,%2" \
|
---|
439 | : "=h"(ret) \
|
---|
440 | : "r"(a), "r"(b) : "l"); \
|
---|
441 | ret; })
|
---|
442 | # define BN_UMULT_LOHI(low,high,a,b)\
|
---|
443 | asm ("dmultu %2,%3" \
|
---|
444 | : "=l"(low),"=h"(high) \
|
---|
445 | : "r"(a), "r"(b));
|
---|
446 | # endif
|
---|
447 | # endif
|
---|
448 | # elif defined(__aarch64__) && defined(SIXTY_FOUR_BIT_LONG)
|
---|
449 | # if defined(__GNUC__) && __GNUC__>=2
|
---|
450 | # define BN_UMULT_HIGH(a,b) ({ \
|
---|
451 | register BN_ULONG ret; \
|
---|
452 | asm ("umulh %0,%1,%2" \
|
---|
453 | : "=r"(ret) \
|
---|
454 | : "r"(a), "r"(b)); \
|
---|
455 | ret; })
|
---|
456 | # endif
|
---|
457 | # endif /* cpu */
|
---|
458 | # endif /* OPENSSL_NO_ASM */
|
---|
459 |
|
---|
460 | /*************************************************************
|
---|
461 | * Using the long long type
|
---|
462 | */
|
---|
463 | # define Lw(t) (((BN_ULONG)(t))&BN_MASK2)
|
---|
464 | # define Hw(t) (((BN_ULONG)((t)>>BN_BITS2))&BN_MASK2)
|
---|
465 |
|
---|
466 | # ifdef BN_DEBUG_RAND
|
---|
467 | # define bn_clear_top2max(a) \
|
---|
468 | { \
|
---|
469 | int ind = (a)->dmax - (a)->top; \
|
---|
470 | BN_ULONG *ftl = &(a)->d[(a)->top-1]; \
|
---|
471 | for (; ind != 0; ind--) \
|
---|
472 | *(++ftl) = 0x0; \
|
---|
473 | }
|
---|
474 | # else
|
---|
475 | # define bn_clear_top2max(a)
|
---|
476 | # endif
|
---|
477 |
|
---|
478 | # ifdef BN_LLONG
|
---|
479 | # define mul_add(r,a,w,c) { \
|
---|
480 | BN_ULLONG t; \
|
---|
481 | t=(BN_ULLONG)w * (a) + (r) + (c); \
|
---|
482 | (r)= Lw(t); \
|
---|
483 | (c)= Hw(t); \
|
---|
484 | }
|
---|
485 |
|
---|
486 | # define mul(r,a,w,c) { \
|
---|
487 | BN_ULLONG t; \
|
---|
488 | t=(BN_ULLONG)w * (a) + (c); \
|
---|
489 | (r)= Lw(t); \
|
---|
490 | (c)= Hw(t); \
|
---|
491 | }
|
---|
492 |
|
---|
493 | # define sqr(r0,r1,a) { \
|
---|
494 | BN_ULLONG t; \
|
---|
495 | t=(BN_ULLONG)(a)*(a); \
|
---|
496 | (r0)=Lw(t); \
|
---|
497 | (r1)=Hw(t); \
|
---|
498 | }
|
---|
499 |
|
---|
500 | # elif defined(BN_UMULT_LOHI)
|
---|
501 | # define mul_add(r,a,w,c) { \
|
---|
502 | BN_ULONG high,low,ret,tmp=(a); \
|
---|
503 | ret = (r); \
|
---|
504 | BN_UMULT_LOHI(low,high,w,tmp); \
|
---|
505 | ret += (c); \
|
---|
506 | (c) = (ret<(c))?1:0; \
|
---|
507 | (c) += high; \
|
---|
508 | ret += low; \
|
---|
509 | (c) += (ret<low)?1:0; \
|
---|
510 | (r) = ret; \
|
---|
511 | }
|
---|
512 |
|
---|
513 | # define mul(r,a,w,c) { \
|
---|
514 | BN_ULONG high,low,ret,ta=(a); \
|
---|
515 | BN_UMULT_LOHI(low,high,w,ta); \
|
---|
516 | ret = low + (c); \
|
---|
517 | (c) = high; \
|
---|
518 | (c) += (ret<low)?1:0; \
|
---|
519 | (r) = ret; \
|
---|
520 | }
|
---|
521 |
|
---|
522 | # define sqr(r0,r1,a) { \
|
---|
523 | BN_ULONG tmp=(a); \
|
---|
524 | BN_UMULT_LOHI(r0,r1,tmp,tmp); \
|
---|
525 | }
|
---|
526 |
|
---|
527 | # elif defined(BN_UMULT_HIGH)
|
---|
528 | # define mul_add(r,a,w,c) { \
|
---|
529 | BN_ULONG high,low,ret,tmp=(a); \
|
---|
530 | ret = (r); \
|
---|
531 | high= BN_UMULT_HIGH(w,tmp); \
|
---|
532 | ret += (c); \
|
---|
533 | low = (w) * tmp; \
|
---|
534 | (c) = (ret<(c))?1:0; \
|
---|
535 | (c) += high; \
|
---|
536 | ret += low; \
|
---|
537 | (c) += (ret<low)?1:0; \
|
---|
538 | (r) = ret; \
|
---|
539 | }
|
---|
540 |
|
---|
541 | # define mul(r,a,w,c) { \
|
---|
542 | BN_ULONG high,low,ret,ta=(a); \
|
---|
543 | low = (w) * ta; \
|
---|
544 | high= BN_UMULT_HIGH(w,ta); \
|
---|
545 | ret = low + (c); \
|
---|
546 | (c) = high; \
|
---|
547 | (c) += (ret<low)?1:0; \
|
---|
548 | (r) = ret; \
|
---|
549 | }
|
---|
550 |
|
---|
551 | # define sqr(r0,r1,a) { \
|
---|
552 | BN_ULONG tmp=(a); \
|
---|
553 | (r0) = tmp * tmp; \
|
---|
554 | (r1) = BN_UMULT_HIGH(tmp,tmp); \
|
---|
555 | }
|
---|
556 |
|
---|
557 | # else
|
---|
558 | /*************************************************************
|
---|
559 | * No long long type
|
---|
560 | */
|
---|
561 |
|
---|
562 | # define LBITS(a) ((a)&BN_MASK2l)
|
---|
563 | # define HBITS(a) (((a)>>BN_BITS4)&BN_MASK2l)
|
---|
564 | # define L2HBITS(a) (((a)<<BN_BITS4)&BN_MASK2)
|
---|
565 |
|
---|
566 | # define LLBITS(a) ((a)&BN_MASKl)
|
---|
567 | # define LHBITS(a) (((a)>>BN_BITS2)&BN_MASKl)
|
---|
568 | # define LL2HBITS(a) ((BN_ULLONG)((a)&BN_MASKl)<<BN_BITS2)
|
---|
569 |
|
---|
570 | # define mul64(l,h,bl,bh) \
|
---|
571 | { \
|
---|
572 | BN_ULONG m,m1,lt,ht; \
|
---|
573 | \
|
---|
574 | lt=l; \
|
---|
575 | ht=h; \
|
---|
576 | m =(bh)*(lt); \
|
---|
577 | lt=(bl)*(lt); \
|
---|
578 | m1=(bl)*(ht); \
|
---|
579 | ht =(bh)*(ht); \
|
---|
580 | m=(m+m1)&BN_MASK2; if (m < m1) ht+=L2HBITS((BN_ULONG)1); \
|
---|
581 | ht+=HBITS(m); \
|
---|
582 | m1=L2HBITS(m); \
|
---|
583 | lt=(lt+m1)&BN_MASK2; if (lt < m1) ht++; \
|
---|
584 | (l)=lt; \
|
---|
585 | (h)=ht; \
|
---|
586 | }
|
---|
587 |
|
---|
588 | # define sqr64(lo,ho,in) \
|
---|
589 | { \
|
---|
590 | BN_ULONG l,h,m; \
|
---|
591 | \
|
---|
592 | h=(in); \
|
---|
593 | l=LBITS(h); \
|
---|
594 | h=HBITS(h); \
|
---|
595 | m =(l)*(h); \
|
---|
596 | l*=l; \
|
---|
597 | h*=h; \
|
---|
598 | h+=(m&BN_MASK2h1)>>(BN_BITS4-1); \
|
---|
599 | m =(m&BN_MASK2l)<<(BN_BITS4+1); \
|
---|
600 | l=(l+m)&BN_MASK2; if (l < m) h++; \
|
---|
601 | (lo)=l; \
|
---|
602 | (ho)=h; \
|
---|
603 | }
|
---|
604 |
|
---|
605 | # define mul_add(r,a,bl,bh,c) { \
|
---|
606 | BN_ULONG l,h; \
|
---|
607 | \
|
---|
608 | h= (a); \
|
---|
609 | l=LBITS(h); \
|
---|
610 | h=HBITS(h); \
|
---|
611 | mul64(l,h,(bl),(bh)); \
|
---|
612 | \
|
---|
613 | /* non-multiply part */ \
|
---|
614 | l=(l+(c))&BN_MASK2; if (l < (c)) h++; \
|
---|
615 | (c)=(r); \
|
---|
616 | l=(l+(c))&BN_MASK2; if (l < (c)) h++; \
|
---|
617 | (c)=h&BN_MASK2; \
|
---|
618 | (r)=l; \
|
---|
619 | }
|
---|
620 |
|
---|
621 | # define mul(r,a,bl,bh,c) { \
|
---|
622 | BN_ULONG l,h; \
|
---|
623 | \
|
---|
624 | h= (a); \
|
---|
625 | l=LBITS(h); \
|
---|
626 | h=HBITS(h); \
|
---|
627 | mul64(l,h,(bl),(bh)); \
|
---|
628 | \
|
---|
629 | /* non-multiply part */ \
|
---|
630 | l+=(c); if ((l&BN_MASK2) < (c)) h++; \
|
---|
631 | (c)=h&BN_MASK2; \
|
---|
632 | (r)=l&BN_MASK2; \
|
---|
633 | }
|
---|
634 | # endif /* !BN_LLONG */
|
---|
635 |
|
---|
636 | void BN_RECP_CTX_init(BN_RECP_CTX *recp);
|
---|
637 | void BN_MONT_CTX_init(BN_MONT_CTX *ctx);
|
---|
638 |
|
---|
639 | void bn_init(BIGNUM *a);
|
---|
640 | void bn_mul_normal(BN_ULONG *r, BN_ULONG *a, int na, BN_ULONG *b, int nb);
|
---|
641 | void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b);
|
---|
642 | void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b);
|
---|
643 | void bn_sqr_normal(BN_ULONG *r, const BN_ULONG *a, int n, BN_ULONG *tmp);
|
---|
644 | void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a);
|
---|
645 | void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a);
|
---|
646 | int bn_cmp_words(const BN_ULONG *a, const BN_ULONG *b, int n);
|
---|
647 | int bn_cmp_part_words(const BN_ULONG *a, const BN_ULONG *b, int cl, int dl);
|
---|
648 | void bn_mul_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2,
|
---|
649 | int dna, int dnb, BN_ULONG *t);
|
---|
650 | void bn_mul_part_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b,
|
---|
651 | int n, int tna, int tnb, BN_ULONG *t);
|
---|
652 | void bn_sqr_recursive(BN_ULONG *r, const BN_ULONG *a, int n2, BN_ULONG *t);
|
---|
653 | void bn_mul_low_normal(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n);
|
---|
654 | void bn_mul_low_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2,
|
---|
655 | BN_ULONG *t);
|
---|
656 | void bn_mul_high(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, BN_ULONG *l, int n2,
|
---|
657 | BN_ULONG *t);
|
---|
658 | BN_ULONG bn_add_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
|
---|
659 | int cl, int dl);
|
---|
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 |
|
---|
665 | BIGNUM *int_bn_mod_inverse(BIGNUM *in,
|
---|
666 | const BIGNUM *a, const BIGNUM *n, BN_CTX *ctx,
|
---|
667 | int *noinv);
|
---|
668 |
|
---|
669 | int bn_probable_prime_dh(BIGNUM *rnd, int bits,
|
---|
670 | const BIGNUM *add, const BIGNUM *rem, BN_CTX *ctx);
|
---|
671 | int bn_probable_prime_dh_retry(BIGNUM *rnd, int bits, BN_CTX *ctx);
|
---|
672 | int bn_probable_prime_dh_coprime(BIGNUM *rnd, int bits, BN_CTX *ctx);
|
---|
673 |
|
---|
674 | static ossl_inline BIGNUM *bn_expand(BIGNUM *a, int bits)
|
---|
675 | {
|
---|
676 | if (bits > (INT_MAX - BN_BITS2 + 1))
|
---|
677 | return NULL;
|
---|
678 |
|
---|
679 | if (((bits+BN_BITS2-1)/BN_BITS2) <= (a)->dmax)
|
---|
680 | return a;
|
---|
681 |
|
---|
682 | return bn_expand2((a),(bits+BN_BITS2-1)/BN_BITS2);
|
---|
683 | }
|
---|
684 |
|
---|
685 | #ifdef __cplusplus
|
---|
686 | }
|
---|
687 | #endif
|
---|
688 |
|
---|
689 | #endif
|
---|