xref: /freebsd/crypto/openssl/crypto/bn/bn_local.h (revision 6b405053)
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 #  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 #  define BN_BITS4        32
94 #  define BN_MASK2        (0xffffffffffffffffL)
95 #  define BN_MASK2l       (0xffffffffL)
96 #  define BN_MASK2h       (0xffffffff00000000L)
97 #  define BN_MASK2h1      (0xffffffff80000000L)
98 #  define BN_DEC_CONV     (10000000000000000000UL)
99 #  define BN_DEC_NUM      19
100 #  define BN_DEC_FMT1     "%lu"
101 #  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 #  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;                /* Pointer to an array of 'BN_BITS2' bit
247                                  * chunks. */
248     int top;                    /* Index of last used d +1. */
249     /* The next are internal book keeping for bn_expand. */
250     int dmax;                   /* Size of the d array. */
251     int neg;                    /* one if the number is negative */
252     int flags;
253 };
254 
255 /* Used for montgomery multiplication */
256 struct bn_mont_ctx_st {
257     int ri;                     /* number of bits in R */
258     BIGNUM RR;                  /* used to convert to montgomery form,
259                                    possibly zero-padded */
260     BIGNUM N;                   /* The modulus */
261     BIGNUM Ni;                  /* R*(1/R mod N) - N*Ni = 1 (Ni is only
262                                  * stored for bignum algorithm) */
263     BN_ULONG n0[2];             /* least significant word(s) of Ni; (type
264                                  * changed with 0.9.9, was "BN_ULONG n0;"
265                                  * before) */
266     int flags;
267 };
268 
269 /*
270  * Used for reciprocal division/mod functions It cannot be shared between
271  * threads
272  */
273 struct bn_recp_ctx_st {
274     BIGNUM N;                   /* the divisor */
275     BIGNUM Nr;                  /* the reciprocal */
276     int num_bits;
277     int shift;
278     int flags;
279 };
280 
281 /* Used for slow "generation" functions. */
282 struct bn_gencb_st {
283     unsigned int ver;           /* To handle binary (in)compatibility */
284     void *arg;                  /* callback-specific data */
285     union {
286         /* if (ver==1) - handles old style callbacks */
287         void (*cb_1) (int, int, void *);
288         /* if (ver==2) - new callback style */
289         int (*cb_2) (int, int, BN_GENCB *);
290     } cb;
291 };
292 
293 /*-
294  * BN_window_bits_for_exponent_size -- macro for sliding window mod_exp functions
295  *
296  *
297  * For window size 'w' (w >= 2) and a random 'b' bits exponent,
298  * the number of multiplications is a constant plus on average
299  *
300  *    2^(w-1) + (b-w)/(w+1);
301  *
302  * here  2^(w-1)  is for precomputing the table (we actually need
303  * entries only for windows that have the lowest bit set), and
304  * (b-w)/(w+1)  is an approximation for the expected number of
305  * w-bit windows, not counting the first one.
306  *
307  * Thus we should use
308  *
309  *    w >= 6  if        b > 671
310  *     w = 5  if  671 > b > 239
311  *     w = 4  if  239 > b >  79
312  *     w = 3  if   79 > b >  23
313  *    w <= 2  if   23 > b
314  *
315  * (with draws in between).  Very small exponents are often selected
316  * with low Hamming weight, so we use  w = 1  for b <= 23.
317  */
318 # define BN_window_bits_for_exponent_size(b) \
319                 ((b) > 671 ? 6 : \
320                  (b) > 239 ? 5 : \
321                  (b) >  79 ? 4 : \
322                  (b) >  23 ? 3 : 1)
323 
324 /*
325  * BN_mod_exp_mont_consttime is based on the assumption that the L1 data cache
326  * line width of the target processor is at least the following value.
327  */
328 # define MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH      ( 64 )
329 # define MOD_EXP_CTIME_MIN_CACHE_LINE_MASK       (MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH - 1)
330 
331 /*
332  * Window sizes optimized for fixed window size modular exponentiation
333  * algorithm (BN_mod_exp_mont_consttime). To achieve the security goals of
334  * BN_mode_exp_mont_consttime, the maximum size of the window must not exceed
335  * log_2(MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH). Window size thresholds are
336  * defined for cache line sizes of 32 and 64, cache line sizes where
337  * log_2(32)=5 and log_2(64)=6 respectively. A window size of 7 should only be
338  * used on processors that have a 128 byte or greater cache line size.
339  */
340 # if MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 64
341 
342 #  define BN_window_bits_for_ctime_exponent_size(b) \
343                 ((b) > 937 ? 6 : \
344                  (b) > 306 ? 5 : \
345                  (b) >  89 ? 4 : \
346                  (b) >  22 ? 3 : 1)
347 #  define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE    (6)
348 
349 # elif MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 32
350 
351 #  define BN_window_bits_for_ctime_exponent_size(b) \
352                 ((b) > 306 ? 5 : \
353                  (b) >  89 ? 4 : \
354                  (b) >  22 ? 3 : 1)
355 #  define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE    (5)
356 
357 # endif
358 
359 /* Pentium pro 16,16,16,32,64 */
360 /* Alpha       16,16,16,16.64 */
361 # define BN_MULL_SIZE_NORMAL                     (16)/* 32 */
362 # define BN_MUL_RECURSIVE_SIZE_NORMAL            (16)/* 32 less than */
363 # define BN_SQR_RECURSIVE_SIZE_NORMAL            (16)/* 32 */
364 # define BN_MUL_LOW_RECURSIVE_SIZE_NORMAL        (32)/* 32 */
365 # define BN_MONT_CTX_SET_SIZE_WORD               (64)/* 32 */
366 
367 # if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) && !defined(PEDANTIC)
368 /*
369  * BN_UMULT_HIGH section.
370  * If the compiler doesn't support 2*N integer type, then you have to
371  * replace every N*N multiplication with 4 (N/2)*(N/2) accompanied by some
372  * shifts and additions which unavoidably results in severe performance
373  * penalties. Of course provided that the hardware is capable of producing
374  * 2*N result... That's when you normally start considering assembler
375  * implementation. However! It should be pointed out that some CPUs (e.g.,
376  * PowerPC, Alpha, and IA-64) provide *separate* instruction calculating
377  * the upper half of the product placing the result into a general
378  * purpose register. Now *if* the compiler supports inline assembler,
379  * then it's not impossible to implement the "bignum" routines (and have
380  * the compiler optimize 'em) exhibiting "native" performance in C. That's
381  * what BN_UMULT_HIGH macro is about:-) Note that more recent compilers do
382  * support 2*64 integer type, which is also used here.
383  */
384 #  if defined(__SIZEOF_INT128__) && __SIZEOF_INT128__==16 && \
385       (defined(SIXTY_FOUR_BIT) || defined(SIXTY_FOUR_BIT_LONG))
386 #   define BN_UMULT_HIGH(a,b)          (((uint128_t)(a)*(b))>>64)
387 #   define BN_UMULT_LOHI(low,high,a,b) ({       \
388         uint128_t ret=(uint128_t)(a)*(b);   \
389         (high)=ret>>64; (low)=ret;      })
390 #  elif defined(__alpha) && (defined(SIXTY_FOUR_BIT_LONG) || defined(SIXTY_FOUR_BIT))
391 #   if defined(__DECC)
392 #    include <c_asm.h>
393 #    define BN_UMULT_HIGH(a,b)   (BN_ULONG)asm("umulh %a0,%a1,%v0",(a),(b))
394 #   elif defined(__GNUC__) && __GNUC__>=2
395 #    define BN_UMULT_HIGH(a,b)   ({     \
396         register BN_ULONG ret;          \
397         asm ("umulh     %1,%2,%0"       \
398              : "=r"(ret)                \
399              : "r"(a), "r"(b));         \
400         ret;                      })
401 #   endif                       /* compiler */
402 #  elif defined(_ARCH_PPC64) && defined(SIXTY_FOUR_BIT_LONG)
403 #   if defined(__GNUC__) && __GNUC__>=2
404 #    define BN_UMULT_HIGH(a,b)   ({     \
405         register BN_ULONG ret;          \
406         asm ("mulhdu    %0,%1,%2"       \
407              : "=r"(ret)                \
408              : "r"(a), "r"(b));         \
409         ret;                      })
410 #   endif                       /* compiler */
411 #  elif (defined(__x86_64) || defined(__x86_64__)) && \
412        (defined(SIXTY_FOUR_BIT_LONG) || defined(SIXTY_FOUR_BIT))
413 #   if defined(__GNUC__) && __GNUC__>=2
414 #    define BN_UMULT_HIGH(a,b)   ({     \
415         register BN_ULONG ret,discard;  \
416         asm ("mulq      %3"             \
417              : "=a"(discard),"=d"(ret)  \
418              : "a"(a), "g"(b)           \
419              : "cc");                   \
420         ret;                      })
421 #    define BN_UMULT_LOHI(low,high,a,b) \
422         asm ("mulq      %3"             \
423                 : "=a"(low),"=d"(high)  \
424                 : "a"(a),"g"(b)         \
425                 : "cc");
426 #   endif
427 #  elif (defined(_M_AMD64) || defined(_M_X64)) && defined(SIXTY_FOUR_BIT)
428 #   if defined(_MSC_VER) && _MSC_VER>=1400
429 unsigned __int64 __umulh(unsigned __int64 a, unsigned __int64 b);
430 unsigned __int64 _umul128(unsigned __int64 a, unsigned __int64 b,
431                           unsigned __int64 *h);
432 #    pragma intrinsic(__umulh,_umul128)
433 #    define BN_UMULT_HIGH(a,b)           __umulh((a),(b))
434 #    define BN_UMULT_LOHI(low,high,a,b)  ((low)=_umul128((a),(b),&(high)))
435 #   endif
436 #  elif defined(__mips) && (defined(SIXTY_FOUR_BIT) || defined(SIXTY_FOUR_BIT_LONG))
437 #   if defined(__GNUC__) && __GNUC__>=2
438 #    define BN_UMULT_HIGH(a,b) ({       \
439         register BN_ULONG ret;          \
440         asm ("dmultu    %1,%2"          \
441              : "=h"(ret)                \
442              : "r"(a), "r"(b) : "l");   \
443         ret;                    })
444 #    define BN_UMULT_LOHI(low,high,a,b) \
445         asm ("dmultu    %2,%3"          \
446              : "=l"(low),"=h"(high)     \
447              : "r"(a), "r"(b));
448 #   endif
449 #  elif defined(__aarch64__) && defined(SIXTY_FOUR_BIT_LONG)
450 #   if defined(__GNUC__) && __GNUC__>=2
451 #    define BN_UMULT_HIGH(a,b)   ({     \
452         register BN_ULONG ret;          \
453         asm ("umulh     %0,%1,%2"       \
454              : "=r"(ret)                \
455              : "r"(a), "r"(b));         \
456         ret;                      })
457 #   endif
458 #  endif                        /* cpu */
459 # endif                         /* OPENSSL_NO_ASM */
460 
461 # ifdef BN_RAND_DEBUG
462 #  define bn_clear_top2max(a) \
463         { \
464         int      ind = (a)->dmax - (a)->top; \
465         BN_ULONG *ftl = &(a)->d[(a)->top-1]; \
466         for (; ind != 0; ind--) \
467                 *(++ftl) = 0x0; \
468         }
469 # else
470 #  define bn_clear_top2max(a)
471 # endif
472 
473 # ifdef BN_LLONG
474 /*******************************************************************
475  * Using the long long type, has to be twice as wide as BN_ULONG...
476  */
477 #  define Lw(t)    (((BN_ULONG)(t))&BN_MASK2)
478 #  define Hw(t)    (((BN_ULONG)((t)>>BN_BITS2))&BN_MASK2)
479 
480 #  define mul_add(r,a,w,c) { \
481         BN_ULLONG t; \
482         t=(BN_ULLONG)w * (a) + (r) + (c); \
483         (r)= Lw(t); \
484         (c)= Hw(t); \
485         }
486 
487 #  define mul(r,a,w,c) { \
488         BN_ULLONG t; \
489         t=(BN_ULLONG)w * (a) + (c); \
490         (r)= Lw(t); \
491         (c)= Hw(t); \
492         }
493 
494 #  define sqr(r0,r1,a) { \
495         BN_ULLONG t; \
496         t=(BN_ULLONG)(a)*(a); \
497         (r0)=Lw(t); \
498         (r1)=Hw(t); \
499         }
500 
501 # elif defined(BN_UMULT_LOHI)
502 #  define mul_add(r,a,w,c) {              \
503         BN_ULONG high,low,ret,tmp=(a);  \
504         ret =  (r);                     \
505         BN_UMULT_LOHI(low,high,w,tmp);  \
506         ret += (c);                     \
507         (c) =  (ret<(c));               \
508         (c) += high;                    \
509         ret += low;                     \
510         (c) += (ret<low);               \
511         (r) =  ret;                     \
512         }
513 
514 #  define mul(r,a,w,c)    {               \
515         BN_ULONG high,low,ret,ta=(a);   \
516         BN_UMULT_LOHI(low,high,w,ta);   \
517         ret =  low + (c);               \
518         (c) =  high;                    \
519         (c) += (ret<low);               \
520         (r) =  ret;                     \
521         }
522 
523 #  define sqr(r0,r1,a)    {               \
524         BN_ULONG tmp=(a);               \
525         BN_UMULT_LOHI(r0,r1,tmp,tmp);   \
526         }
527 
528 # elif defined(BN_UMULT_HIGH)
529 #  define mul_add(r,a,w,c) {              \
530         BN_ULONG high,low,ret,tmp=(a);  \
531         ret =  (r);                     \
532         high=  BN_UMULT_HIGH(w,tmp);    \
533         ret += (c);                     \
534         low =  (w) * tmp;               \
535         (c) =  (ret<(c));               \
536         (c) += high;                    \
537         ret += low;                     \
538         (c) += (ret<low);               \
539         (r) =  ret;                     \
540         }
541 
542 #  define mul(r,a,w,c)    {               \
543         BN_ULONG high,low,ret,ta=(a);   \
544         low =  (w) * ta;                \
545         high=  BN_UMULT_HIGH(w,ta);     \
546         ret =  low + (c);               \
547         (c) =  high;                    \
548         (c) += (ret<low);               \
549         (r) =  ret;                     \
550         }
551 
552 #  define sqr(r0,r1,a)    {               \
553         BN_ULONG tmp=(a);               \
554         (r0) = tmp * tmp;               \
555         (r1) = BN_UMULT_HIGH(tmp,tmp);  \
556         }
557 
558 # else
559 /*************************************************************
560  * No long long type
561  */
562 
563 #  define LBITS(a)        ((a)&BN_MASK2l)
564 #  define HBITS(a)        (((a)>>BN_BITS4)&BN_MASK2l)
565 #  define L2HBITS(a)      (((a)<<BN_BITS4)&BN_MASK2)
566 
567 #  define LLBITS(a)       ((a)&BN_MASKl)
568 #  define LHBITS(a)       (((a)>>BN_BITS2)&BN_MASKl)
569 #  define LL2HBITS(a)     ((BN_ULLONG)((a)&BN_MASKl)<<BN_BITS2)
570 
571 #  define mul64(l,h,bl,bh) \
572         { \
573         BN_ULONG m,m1,lt,ht; \
574  \
575         lt=l; \
576         ht=h; \
577         m =(bh)*(lt); \
578         lt=(bl)*(lt); \
579         m1=(bl)*(ht); \
580         ht =(bh)*(ht); \
581         m=(m+m1)&BN_MASK2; ht += L2HBITS((BN_ULONG)(m < m1)); \
582         ht+=HBITS(m); \
583         m1=L2HBITS(m); \
584         lt=(lt+m1)&BN_MASK2; ht += (lt < m1); \
585         (l)=lt; \
586         (h)=ht; \
587         }
588 
589 #  define sqr64(lo,ho,in) \
590         { \
591         BN_ULONG l,h,m; \
592  \
593         h=(in); \
594         l=LBITS(h); \
595         h=HBITS(h); \
596         m =(l)*(h); \
597         l*=l; \
598         h*=h; \
599         h+=(m&BN_MASK2h1)>>(BN_BITS4-1); \
600         m =(m&BN_MASK2l)<<(BN_BITS4+1); \
601         l=(l+m)&BN_MASK2; h += (l < m); \
602         (lo)=l; \
603         (ho)=h; \
604         }
605 
606 #  define mul_add(r,a,bl,bh,c) { \
607         BN_ULONG l,h; \
608  \
609         h= (a); \
610         l=LBITS(h); \
611         h=HBITS(h); \
612         mul64(l,h,(bl),(bh)); \
613  \
614         /* non-multiply part */ \
615         l=(l+(c))&BN_MASK2; h += (l < (c)); \
616         (c)=(r); \
617         l=(l+(c))&BN_MASK2; h += (l < (c)); \
618         (c)=h&BN_MASK2; \
619         (r)=l; \
620         }
621 
622 #  define mul(r,a,bl,bh,c) { \
623         BN_ULONG l,h; \
624  \
625         h= (a); \
626         l=LBITS(h); \
627         h=HBITS(h); \
628         mul64(l,h,(bl),(bh)); \
629  \
630         /* non-multiply part */ \
631         l+=(c); h += ((l&BN_MASK2) < (c)); \
632         (c)=h&BN_MASK2; \
633         (r)=l&BN_MASK2; \
634         }
635 # endif                         /* !BN_LLONG */
636 
637 void BN_RECP_CTX_init(BN_RECP_CTX *recp);
638 void BN_MONT_CTX_init(BN_MONT_CTX *ctx);
639 
640 void bn_init(BIGNUM *a);
641 void bn_mul_normal(BN_ULONG *r, BN_ULONG *a, int na, BN_ULONG *b, int nb);
642 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b);
643 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b);
644 void bn_sqr_normal(BN_ULONG *r, const BN_ULONG *a, int n, BN_ULONG *tmp);
645 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a);
646 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a);
647 int bn_cmp_words(const BN_ULONG *a, const BN_ULONG *b, int n);
648 int bn_cmp_part_words(const BN_ULONG *a, const BN_ULONG *b, int cl, int dl);
649 void bn_mul_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2,
650                       int dna, int dnb, BN_ULONG *t);
651 void bn_mul_part_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b,
652                            int n, int tna, int tnb, BN_ULONG *t);
653 void bn_sqr_recursive(BN_ULONG *r, const BN_ULONG *a, int n2, BN_ULONG *t);
654 void bn_mul_low_normal(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n);
655 void bn_mul_low_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2,
656                           BN_ULONG *t);
657 BN_ULONG bn_sub_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
658                            int cl, int dl);
659 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
660                 const BN_ULONG *np, const BN_ULONG *n0, int num);
661 void bn_correct_top_consttime(BIGNUM *a);
662 BIGNUM *int_bn_mod_inverse(BIGNUM *in,
663                            const BIGNUM *a, const BIGNUM *n, BN_CTX *ctx,
664                            int *noinv);
665 
bn_expand(BIGNUM * a,int bits)666 static ossl_inline BIGNUM *bn_expand(BIGNUM *a, int bits)
667 {
668     if (bits > (INT_MAX - BN_BITS2 + 1))
669         return NULL;
670 
671     if (((bits+BN_BITS2-1)/BN_BITS2) <= (a)->dmax)
672         return a;
673 
674     return bn_expand2((a),(bits+BN_BITS2-1)/BN_BITS2);
675 }
676 
677 int ossl_bn_check_prime(const BIGNUM *w, int checks, BN_CTX *ctx,
678                         int do_trial_division, BN_GENCB *cb);
679 
680 #endif
681