xref: /dragonfly/crypto/libressl/crypto/bn/bn_lcl.h (revision 6f5ec8b5)
1 /* $OpenBSD: bn_lcl.h,v 1.35 2022/07/15 06:10:00 tb Exp $ */
2 /* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com)
3  * All rights reserved.
4  *
5  * This package is an SSL implementation written
6  * by Eric Young (eay@cryptsoft.com).
7  * The implementation was written so as to conform with Netscapes SSL.
8  *
9  * This library is free for commercial and non-commercial use as long as
10  * the following conditions are aheared to.  The following conditions
11  * apply to all code found in this distribution, be it the RC4, RSA,
12  * lhash, DES, etc., code; not just the SSL code.  The SSL documentation
13  * included with this distribution is covered by the same copyright terms
14  * except that the holder is Tim Hudson (tjh@cryptsoft.com).
15  *
16  * Copyright remains Eric Young's, and as such any Copyright notices in
17  * the code are not to be removed.
18  * If this package is used in a product, Eric Young should be given attribution
19  * as the author of the parts of the library used.
20  * This can be in the form of a textual message at program startup or
21  * in documentation (online or textual) provided with the package.
22  *
23  * Redistribution and use in source and binary forms, with or without
24  * modification, are permitted provided that the following conditions
25  * are met:
26  * 1. Redistributions of source code must retain the copyright
27  *    notice, this list of conditions and the following disclaimer.
28  * 2. Redistributions in binary form must reproduce the above copyright
29  *    notice, this list of conditions and the following disclaimer in the
30  *    documentation and/or other materials provided with the distribution.
31  * 3. All advertising materials mentioning features or use of this software
32  *    must display the following acknowledgement:
33  *    "This product includes cryptographic software written by
34  *     Eric Young (eay@cryptsoft.com)"
35  *    The word 'cryptographic' can be left out if the rouines from the library
36  *    being used are not cryptographic related :-).
37  * 4. If you include any Windows specific code (or a derivative thereof) from
38  *    the apps directory (application code) you must include an acknowledgement:
39  *    "This product includes software written by Tim Hudson (tjh@cryptsoft.com)"
40  *
41  * THIS SOFTWARE IS PROVIDED BY ERIC YOUNG ``AS IS'' AND
42  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
43  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
44  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
45  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
46  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
47  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
48  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
49  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
50  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
51  * SUCH DAMAGE.
52  *
53  * The licence and distribution terms for any publically available version or
54  * derivative of this code cannot be changed.  i.e. this code cannot simply be
55  * copied and put under another distribution licence
56  * [including the GNU Public Licence.]
57  */
58 /* ====================================================================
59  * Copyright (c) 1998-2000 The OpenSSL Project.  All rights reserved.
60  *
61  * Redistribution and use in source and binary forms, with or without
62  * modification, are permitted provided that the following conditions
63  * are met:
64  *
65  * 1. Redistributions of source code must retain the above copyright
66  *    notice, this list of conditions and the following disclaimer.
67  *
68  * 2. Redistributions in binary form must reproduce the above copyright
69  *    notice, this list of conditions and the following disclaimer in
70  *    the documentation and/or other materials provided with the
71  *    distribution.
72  *
73  * 3. All advertising materials mentioning features or use of this
74  *    software must display the following acknowledgment:
75  *    "This product includes software developed by the OpenSSL Project
76  *    for use in the OpenSSL Toolkit. (http://www.openssl.org/)"
77  *
78  * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
79  *    endorse or promote products derived from this software without
80  *    prior written permission. For written permission, please contact
81  *    openssl-core@openssl.org.
82  *
83  * 5. Products derived from this software may not be called "OpenSSL"
84  *    nor may "OpenSSL" appear in their names without prior written
85  *    permission of the OpenSSL Project.
86  *
87  * 6. Redistributions of any form whatsoever must retain the following
88  *    acknowledgment:
89  *    "This product includes software developed by the OpenSSL Project
90  *    for use in the OpenSSL Toolkit (http://www.openssl.org/)"
91  *
92  * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
93  * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
94  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
95  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE OpenSSL PROJECT OR
96  * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
97  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
98  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
99  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
100  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
101  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
102  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
103  * OF THE POSSIBILITY OF SUCH DAMAGE.
104  * ====================================================================
105  *
106  * This product includes cryptographic software written by Eric Young
107  * (eay@cryptsoft.com).  This product includes software written by Tim
108  * Hudson (tjh@cryptsoft.com).
109  *
110  */
111 
112 #ifndef HEADER_BN_LCL_H
113 #define HEADER_BN_LCL_H
114 
115 #include <openssl/opensslconf.h>
116 
117 #include <openssl/bn.h>
118 
119 __BEGIN_HIDDEN_DECLS
120 
121 struct bignum_st {
122 	BN_ULONG *d;	/* Pointer to an array of 'BN_BITS2' bit chunks. */
123 	int top;	/* Index of last used d +1. */
124 	/* The next are internal book keeping for bn_expand. */
125 	int dmax;	/* Size of the d array. */
126 	int neg;	/* one if the number is negative */
127 	int flags;
128 };
129 
130 /* Used for montgomery multiplication */
131 struct bn_mont_ctx_st {
132 	int ri;        /* number of bits in R */
133 	BIGNUM RR;     /* used to convert to montgomery form */
134 	BIGNUM N;      /* The modulus */
135 	BIGNUM Ni;     /* R*(1/R mod N) - N*Ni = 1
136 	                * (Ni is only stored for bignum algorithm) */
137 	BN_ULONG n0[2];/* least significant word(s) of Ni;
138 	                  (type changed with 0.9.9, was "BN_ULONG n0;" before) */
139 	int flags;
140 };
141 
142 /* Used for reciprocal division/mod functions
143  * It cannot be shared between threads
144  */
145 struct bn_recp_ctx_st {
146 	BIGNUM N;	/* the divisor */
147 	BIGNUM Nr;	/* the reciprocal */
148 	int num_bits;
149 	int shift;
150 	int flags;
151 };
152 
153 /* Used for slow "generation" functions. */
154 struct bn_gencb_st {
155 	unsigned int ver;	/* To handle binary (in)compatibility */
156 	void *arg;		/* callback-specific data */
157 	union {
158 		/* if(ver==1) - handles old style callbacks */
159 		void (*cb_1)(int, int, void *);
160 		/* if(ver==2) - new callback style */
161 		int (*cb_2)(int, int, BN_GENCB *);
162 	} cb;
163 };
164 
165 /*
166  * BN_window_bits_for_exponent_size -- macro for sliding window mod_exp functions
167  *
168  *
169  * For window size 'w' (w >= 2) and a random 'b' bits exponent,
170  * the number of multiplications is a constant plus on average
171  *
172  *    2^(w-1) + (b-w)/(w+1);
173  *
174  * here  2^(w-1)  is for precomputing the table (we actually need
175  * entries only for windows that have the lowest bit set), and
176  * (b-w)/(w+1)  is an approximation for the expected number of
177  * w-bit windows, not counting the first one.
178  *
179  * Thus we should use
180  *
181  *    w >= 6  if        b > 671
182  *     w = 5  if  671 > b > 239
183  *     w = 4  if  239 > b >  79
184  *     w = 3  if   79 > b >  23
185  *    w <= 2  if   23 > b
186  *
187  * (with draws in between).  Very small exponents are often selected
188  * with low Hamming weight, so we use  w = 1  for b <= 23.
189  */
190 #define BN_window_bits_for_exponent_size(b) \
191 		((b) > 671 ? 6 : \
192 		 (b) > 239 ? 5 : \
193 		 (b) >  79 ? 4 : \
194 		 (b) >  23 ? 3 : 1)
195 
196 
197 /* BN_mod_exp_mont_consttime is based on the assumption that the
198  * L1 data cache line width of the target processor is at least
199  * the following value.
200  */
201 #define MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH	( 64 )
202 #define MOD_EXP_CTIME_MIN_CACHE_LINE_MASK	(MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH - 1)
203 
204 /* Window sizes optimized for fixed window size modular exponentiation
205  * algorithm (BN_mod_exp_mont_consttime).
206  *
207  * To achieve the security goals of BN_mode_exp_mont_consttime, the
208  * maximum size of the window must not exceed
209  * log_2(MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH).
210  *
211  * Window size thresholds are defined for cache line sizes of 32 and 64,
212  * cache line sizes where log_2(32)=5 and log_2(64)=6 respectively. A
213  * window size of 7 should only be used on processors that have a 128
214  * byte or greater cache line size.
215  */
216 #if MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 64
217 
218 #  define BN_window_bits_for_ctime_exponent_size(b) \
219 		((b) > 937 ? 6 : \
220 		 (b) > 306 ? 5 : \
221 		 (b) >  89 ? 4 : \
222 		 (b) >  22 ? 3 : 1)
223 #  define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE	(6)
224 
225 #elif MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH == 32
226 
227 #  define BN_window_bits_for_ctime_exponent_size(b) \
228 		((b) > 306 ? 5 : \
229 		 (b) >  89 ? 4 : \
230 		 (b) >  22 ? 3 : 1)
231 #  define BN_MAX_WINDOW_BITS_FOR_CTIME_EXPONENT_SIZE	(5)
232 
233 #endif
234 
235 
236 /* Pentium pro 16,16,16,32,64 */
237 /* Alpha       16,16,16,16.64 */
238 #define BN_MULL_SIZE_NORMAL			(16) /* 32 */
239 #define BN_MUL_RECURSIVE_SIZE_NORMAL		(16) /* 32 less than */
240 #define BN_SQR_RECURSIVE_SIZE_NORMAL		(16) /* 32 */
241 #define BN_MUL_LOW_RECURSIVE_SIZE_NORMAL	(32) /* 32 */
242 #define BN_MONT_CTX_SET_SIZE_WORD		(64) /* 32 */
243 
244 #if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
245 /*
246  * BN_UMULT_HIGH section.
247  *
248  * No, I'm not trying to overwhelm you when stating that the
249  * product of N-bit numbers is 2*N bits wide:-) No, I don't expect
250  * you to be impressed when I say that if the compiler doesn't
251  * support 2*N integer type, then you have to replace every N*N
252  * multiplication with 4 (N/2)*(N/2) accompanied by some shifts
253  * and additions which unavoidably results in severe performance
254  * penalties. Of course provided that the hardware is capable of
255  * producing 2*N result... That's when you normally start
256  * considering assembler implementation. However! It should be
257  * pointed out that some CPUs (most notably Alpha, PowerPC and
258  * upcoming IA-64 family:-) provide *separate* instruction
259  * calculating the upper half of the product placing the result
260  * into a general purpose register. Now *if* the compiler supports
261  * inline assembler, then it's not impossible to implement the
262  * "bignum" routines (and have the compiler optimize 'em)
263  * exhibiting "native" performance in C. That's what BN_UMULT_HIGH
264  * macro is about:-)
265  *
266  *					<appro@fy.chalmers.se>
267  */
268 # if defined(__alpha)
269 #  if defined(__GNUC__) && __GNUC__>=2
270 #   define BN_UMULT_HIGH(a,b)	({	\
271 	BN_ULONG ret;		\
272 	asm ("umulh	%1,%2,%0"	\
273 	     : "=r"(ret)		\
274 	     : "r"(a), "r"(b));		\
275 	ret;			})
276 #  endif	/* compiler */
277 # elif defined(_ARCH_PPC) && defined(_LP64)
278 #  if defined(__GNUC__) && __GNUC__>=2
279 #   define BN_UMULT_HIGH(a,b)	({	\
280 	BN_ULONG ret;		\
281 	asm ("mulhdu	%0,%1,%2"	\
282 	     : "=r"(ret)		\
283 	     : "r"(a), "r"(b));		\
284 	ret;			})
285 #  endif	/* compiler */
286 # elif (defined(__x86_64) || defined(__x86_64__)) && defined(_LP64)
287 #  if defined(__GNUC__) && __GNUC__>=2
288 #   define BN_UMULT_HIGH(a,b)	({	\
289 	BN_ULONG ret,discard;	\
290 	asm ("mulq	%3"		\
291 	     : "=a"(discard),"=d"(ret)	\
292 	     : "a"(a), "g"(b)		\
293 	     : "cc");			\
294 	ret;			})
295 #   define BN_UMULT_LOHI(low,high,a,b)	\
296 	asm ("mulq	%3"		\
297 		: "=a"(low),"=d"(high)	\
298 		: "a"(a),"g"(b)		\
299 		: "cc");
300 #  endif
301 # elif defined(__mips) && defined(_LP64)
302 #  if defined(__GNUC__) && __GNUC__>=2
303 #   if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 4) /* "h" constraint is no more since 4.4 */
304 #     define BN_UMULT_HIGH(a,b)		 (((__uint128_t)(a)*(b))>>64)
305 #     define BN_UMULT_LOHI(low,high,a,b) ({	\
306 	__uint128_t ret=(__uint128_t)(a)*(b);	\
307 	(high)=ret>>64; (low)=ret;	 })
308 #   else
309 #     define BN_UMULT_HIGH(a,b)	({	\
310 	BN_ULONG ret;		\
311 	asm ("dmultu	%1,%2"		\
312 	     : "=h"(ret)		\
313 	     : "r"(a), "r"(b) : "l");	\
314 	ret;			})
315 #     define BN_UMULT_LOHI(low,high,a,b)\
316 	asm ("dmultu	%2,%3"		\
317 	     : "=l"(low),"=h"(high)	\
318 	     : "r"(a), "r"(b));
319 #    endif
320 #  endif
321 # endif		/* cpu */
322 #endif		/* OPENSSL_NO_ASM */
323 
324 /*************************************************************
325  * Using the long long type
326  */
327 #define Lw(t)    (((BN_ULONG)(t))&BN_MASK2)
328 #define Hw(t)    (((BN_ULONG)((t)>>BN_BITS2))&BN_MASK2)
329 
330 #ifdef BN_DEBUG_RAND
331 #define bn_clear_top2max(a) \
332 	{ \
333 	int      ind = (a)->dmax - (a)->top; \
334 	BN_ULONG *ftl = &(a)->d[(a)->top-1]; \
335 	for (; ind != 0; ind--) \
336 		*(++ftl) = 0x0; \
337 	}
338 #else
339 #define bn_clear_top2max(a)
340 #endif
341 
342 #ifdef BN_LLONG
343 #define mul_add(r,a,w,c) { \
344 	BN_ULLONG t; \
345 	t=(BN_ULLONG)w * (a) + (r) + (c); \
346 	(r)= Lw(t); \
347 	(c)= Hw(t); \
348 	}
349 
350 #define mul(r,a,w,c) { \
351 	BN_ULLONG t; \
352 	t=(BN_ULLONG)w * (a) + (c); \
353 	(r)= Lw(t); \
354 	(c)= Hw(t); \
355 	}
356 
357 #define sqr(r0,r1,a) { \
358 	BN_ULLONG t; \
359 	t=(BN_ULLONG)(a)*(a); \
360 	(r0)=Lw(t); \
361 	(r1)=Hw(t); \
362 	}
363 
364 #elif defined(BN_UMULT_LOHI)
365 #define mul_add(r,a,w,c) {		\
366 	BN_ULONG high,low,ret,tmp=(a);	\
367 	ret =  (r);			\
368 	BN_UMULT_LOHI(low,high,w,tmp);	\
369 	ret += (c);			\
370 	(c) =  (ret<(c))?1:0;		\
371 	(c) += high;			\
372 	ret += low;			\
373 	(c) += (ret<low)?1:0;		\
374 	(r) =  ret;			\
375 	}
376 
377 #define mul(r,a,w,c)	{		\
378 	BN_ULONG high,low,ret,ta=(a);	\
379 	BN_UMULT_LOHI(low,high,w,ta);	\
380 	ret =  low + (c);		\
381 	(c) =  high;			\
382 	(c) += (ret<low)?1:0;		\
383 	(r) =  ret;			\
384 	}
385 
386 #define sqr(r0,r1,a)	{		\
387 	BN_ULONG tmp=(a);		\
388 	BN_UMULT_LOHI(r0,r1,tmp,tmp);	\
389 	}
390 
391 #elif defined(BN_UMULT_HIGH)
392 #define mul_add(r,a,w,c) {		\
393 	BN_ULONG high,low,ret,tmp=(a);	\
394 	ret =  (r);			\
395 	high=  BN_UMULT_HIGH(w,tmp);	\
396 	ret += (c);			\
397 	low =  (w) * tmp;		\
398 	(c) =  (ret<(c))?1:0;		\
399 	(c) += high;			\
400 	ret += low;			\
401 	(c) += (ret<low)?1:0;		\
402 	(r) =  ret;			\
403 	}
404 
405 #define mul(r,a,w,c)	{		\
406 	BN_ULONG high,low,ret,ta=(a);	\
407 	low =  (w) * ta;		\
408 	high=  BN_UMULT_HIGH(w,ta);	\
409 	ret =  low + (c);		\
410 	(c) =  high;			\
411 	(c) += (ret<low)?1:0;		\
412 	(r) =  ret;			\
413 	}
414 
415 #define sqr(r0,r1,a)	{		\
416 	BN_ULONG tmp=(a);		\
417 	(r0) = tmp * tmp;		\
418 	(r1) = BN_UMULT_HIGH(tmp,tmp);	\
419 	}
420 
421 #else
422 /*************************************************************
423  * No long long type
424  */
425 
426 #define LBITS(a)	((a)&BN_MASK2l)
427 #define HBITS(a)	(((a)>>BN_BITS4)&BN_MASK2l)
428 #define	L2HBITS(a)	(((a)<<BN_BITS4)&BN_MASK2)
429 
430 #define mul64(l,h,bl,bh) \
431 	{ \
432 	BN_ULONG m,m1,lt,ht; \
433  \
434 	lt=l; \
435 	ht=h; \
436 	m =(bh)*(lt); \
437 	lt=(bl)*(lt); \
438 	m1=(bl)*(ht); \
439 	ht =(bh)*(ht); \
440 	m=(m+m1)&BN_MASK2; if (m < m1) ht+=L2HBITS((BN_ULONG)1); \
441 	ht+=HBITS(m); \
442 	m1=L2HBITS(m); \
443 	lt=(lt+m1)&BN_MASK2; if (lt < m1) ht++; \
444 	(l)=lt; \
445 	(h)=ht; \
446 	}
447 
448 #define sqr64(lo,ho,in) \
449 	{ \
450 	BN_ULONG l,h,m; \
451  \
452 	h=(in); \
453 	l=LBITS(h); \
454 	h=HBITS(h); \
455 	m =(l)*(h); \
456 	l*=l; \
457 	h*=h; \
458 	h+=(m&BN_MASK2h1)>>(BN_BITS4-1); \
459 	m =(m&BN_MASK2l)<<(BN_BITS4+1); \
460 	l=(l+m)&BN_MASK2; if (l < m) h++; \
461 	(lo)=l; \
462 	(ho)=h; \
463 	}
464 
465 #define mul_add(r,a,bl,bh,c) { \
466 	BN_ULONG l,h; \
467  \
468 	h= (a); \
469 	l=LBITS(h); \
470 	h=HBITS(h); \
471 	mul64(l,h,(bl),(bh)); \
472  \
473 	/* non-multiply part */ \
474 	l=(l+(c))&BN_MASK2; if (l < (c)) h++; \
475 	(c)=(r); \
476 	l=(l+(c))&BN_MASK2; if (l < (c)) h++; \
477 	(c)=h&BN_MASK2; \
478 	(r)=l; \
479 	}
480 
481 #define mul(r,a,bl,bh,c) { \
482 	BN_ULONG l,h; \
483  \
484 	h= (a); \
485 	l=LBITS(h); \
486 	h=HBITS(h); \
487 	mul64(l,h,(bl),(bh)); \
488  \
489 	/* non-multiply part */ \
490 	l+=(c); if ((l&BN_MASK2) < (c)) h++; \
491 	(c)=h&BN_MASK2; \
492 	(r)=l&BN_MASK2; \
493 	}
494 #endif /* !BN_LLONG */
495 
496 /* The least significant word of a BIGNUM. */
497 #define BN_lsw(n) (((n)->top == 0) ? (BN_ULONG) 0 : (n)->d[0])
498 
499 void bn_mul_normal(BN_ULONG *r, BN_ULONG *a, int na, BN_ULONG *b, int nb);
500 void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b);
501 void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b);
502 void bn_sqr_normal(BN_ULONG *r, const BN_ULONG *a, int n, BN_ULONG *tmp);
503 void bn_sqr_comba8(BN_ULONG *r, const BN_ULONG *a);
504 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a);
505 int bn_cmp_words(const BN_ULONG *a, const BN_ULONG *b, int n);
506 int bn_cmp_part_words(const BN_ULONG *a, const BN_ULONG *b,
507     int cl, int dl);
508 void bn_mul_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2,
509     int dna, int dnb, BN_ULONG *t);
510 void bn_mul_part_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b,
511     int n, int tna, int tnb, BN_ULONG *t);
512 void bn_sqr_recursive(BN_ULONG *r, const BN_ULONG *a, int n2, BN_ULONG *t);
513 void bn_mul_low_normal(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n);
514 void bn_mul_low_recursive(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n2,
515     BN_ULONG *t);
516 void bn_mul_high(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, BN_ULONG *l, int n2,
517     BN_ULONG *t);
518 BN_ULONG bn_add_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
519     int cl, int dl);
520 BN_ULONG bn_sub_part_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
521     int cl, int dl);
522 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, const BN_ULONG *np, const BN_ULONG *n0, int num);
523 
524 #define bn_wexpand(a,words) (((words) <= (a)->dmax)?(a):bn_expand2((a),(words)))
525 BIGNUM *bn_expand2(BIGNUM *a, int words);
526 BIGNUM *bn_expand(BIGNUM *a, int bits);
527 
528 BIGNUM *bn_dup_expand(const BIGNUM *a, int words); /* unused */
529 
530 /* Bignum consistency macros
531  * There is one "API" macro, bn_fix_top(), for stripping leading zeroes from
532  * bignum data after direct manipulations on the data. There is also an
533  * "internal" macro, bn_check_top(), for verifying that there are no leading
534  * zeroes. Unfortunately, some auditing is required due to the fact that
535  * bn_fix_top() has become an overabused duct-tape because bignum data is
536  * occasionally passed around in an inconsistent state. So the following
537  * changes have been made to sort this out;
538  * - bn_fix_top()s implementation has been moved to bn_correct_top()
539  * - if BN_DEBUG isn't defined, bn_fix_top() maps to bn_correct_top(), and
540  *   bn_check_top() is as before.
541  * - if BN_DEBUG *is* defined;
542  *   - bn_check_top() tries to pollute unused words even if the bignum 'top' is
543  *     consistent. (ed: only if BN_DEBUG_RAND is defined)
544  *   - bn_fix_top() maps to bn_check_top() rather than "fixing" anything.
545  * The idea is to have debug builds flag up inconsistent bignums when they
546  * occur. If that occurs in a bn_fix_top(), we examine the code in question; if
547  * the use of bn_fix_top() was appropriate (ie. it follows directly after code
548  * that manipulates the bignum) it is converted to bn_correct_top(), and if it
549  * was not appropriate, we convert it permanently to bn_check_top() and track
550  * down the cause of the bug. Eventually, no internal code should be using the
551  * bn_fix_top() macro. External applications and libraries should try this with
552  * their own code too, both in terms of building against the openssl headers
553  * with BN_DEBUG defined *and* linking with a version of OpenSSL built with it
554  * defined. This not only improves external code, it provides more test
555  * coverage for openssl's own code.
556  */
557 
558 #ifdef BN_DEBUG
559 
560 /* We only need assert() when debugging */
561 #include <assert.h>
562 
563 #ifdef BN_DEBUG_RAND
564 #define bn_pollute(a) \
565 	do { \
566 		const BIGNUM *_bnum1 = (a); \
567 		if(_bnum1->top < _bnum1->dmax) { \
568 			unsigned char _tmp_char; \
569 			/* We cast away const without the compiler knowing, any \
570 			 * *genuinely* constant variables that aren't mutable \
571 			 * wouldn't be constructed with top!=dmax. */ \
572 			BN_ULONG *_not_const; \
573 			memcpy(&_not_const, &_bnum1->d, sizeof(BN_ULONG*)); \
574 			arc4random_buf(&_tmp_char, 1); \
575 			memset((unsigned char *)(_not_const + _bnum1->top), _tmp_char, \
576 				(_bnum1->dmax - _bnum1->top) * sizeof(BN_ULONG)); \
577 		} \
578 	} while(0)
579 #else
580 #define bn_pollute(a)
581 #endif
582 
583 #define bn_check_top(a) \
584 	do { \
585 		const BIGNUM *_bnum2 = (a); \
586 		if (_bnum2 != NULL) { \
587 			assert((_bnum2->top == 0) || \
588 				(_bnum2->d[_bnum2->top - 1] != 0)); \
589 			bn_pollute(_bnum2); \
590 		} \
591 	} while(0)
592 
593 #define bn_fix_top(a)		bn_check_top(a)
594 
595 #define bn_check_size(bn, bits) bn_wcheck_size(bn, ((bits+BN_BITS2-1))/BN_BITS2)
596 #define bn_wcheck_size(bn, words) \
597 	do { \
598 		const BIGNUM *_bnum2 = (bn); \
599 		assert(words <= (_bnum2)->dmax && words >= (_bnum2)->top); \
600 	} while(0)
601 
602 #else /* !BN_DEBUG */
603 
604 #define bn_pollute(a)
605 #define bn_check_top(a)
606 #define bn_fix_top(a)		bn_correct_top(a)
607 #define bn_check_size(bn, bits)
608 #define bn_wcheck_size(bn, words)
609 
610 #endif
611 
612 #define bn_correct_top(a) \
613         { \
614         BN_ULONG *ftl; \
615 	int tmp_top = (a)->top; \
616 	if (tmp_top > 0) \
617 		{ \
618 		for (ftl= &((a)->d[tmp_top-1]); tmp_top > 0; tmp_top--) \
619 			if (*(ftl--)) break; \
620 		(a)->top = tmp_top; \
621 		} \
622 	bn_pollute(a); \
623 	}
624 
625 BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w);
626 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, int num, BN_ULONG w);
627 void     bn_sqr_words(BN_ULONG *rp, const BN_ULONG *ap, int num);
628 BN_ULONG bn_div_words(BN_ULONG h, BN_ULONG l, BN_ULONG d);
629 BN_ULONG bn_add_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, int num);
630 BN_ULONG bn_sub_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, int num);
631 
632 int BN_bntest_rand(BIGNUM *rnd, int bits, int top, int bottom);
633 int bn_rand_interval(BIGNUM *rnd, const BIGNUM *lower_inc, const BIGNUM *upper_exc);
634 
635 /* Explicitly const time / non-const time versions for internal use */
636 int BN_mod_exp_ct(BIGNUM *r, const BIGNUM *a, const BIGNUM *p,
637     const BIGNUM *m, BN_CTX *ctx);
638 int BN_mod_exp_nonct(BIGNUM *r, const BIGNUM *a, const BIGNUM *p,
639     const BIGNUM *m, BN_CTX *ctx);
640 int BN_mod_exp_mont_ct(BIGNUM *r, const BIGNUM *a, const BIGNUM *p,
641     const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *m_ctx);
642 int BN_mod_exp_mont_nonct(BIGNUM *r, const BIGNUM *a, const BIGNUM *p,
643     const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *m_ctx);
644 int BN_div_nonct(BIGNUM *dv, BIGNUM *rem, const BIGNUM *m, const BIGNUM *d,
645     BN_CTX *ctx);
646 int BN_div_ct(BIGNUM *dv, BIGNUM *rem, const BIGNUM *m, const BIGNUM *d,
647     BN_CTX *ctx);
648 #define BN_mod_ct(rem,m,d,ctx) BN_div_ct(NULL,(rem),(m),(d),(ctx))
649 #define BN_mod_nonct(rem,m,d,ctx) BN_div_nonct(NULL,(rem),(m),(d),(ctx))
650 BIGNUM *BN_mod_inverse_ct(BIGNUM *ret, const BIGNUM *a, const BIGNUM *n,
651     BN_CTX *ctx);
652 BIGNUM *BN_mod_inverse_nonct(BIGNUM *ret, const BIGNUM *a, const BIGNUM *n,
653     BN_CTX *ctx);
654 int	BN_gcd_ct(BIGNUM *r, const BIGNUM *a, const BIGNUM *b, BN_CTX *ctx);
655 int	BN_gcd_nonct(BIGNUM *r, const BIGNUM *a, const BIGNUM *b, BN_CTX *ctx);
656 
657 int	BN_swap_ct(BN_ULONG swap, BIGNUM *a, BIGNUM *b, size_t nwords);
658 
659 int bn_isqrt(BIGNUM *out_sqrt, int *out_perfect, const BIGNUM *n, BN_CTX *ctx);
660 int bn_is_perfect_square(int *out_perfect, const BIGNUM *n, BN_CTX *ctx);
661 
662 int bn_is_prime_bpsw(int *is_prime, const BIGNUM *n, BN_CTX *in_ctx);
663 
664 __END_HIDDEN_DECLS
665 #endif
666