• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* Copyright (C) 1995-1997 Eric Young (eay@cryptsoft.com)
2  * All rights reserved.
3  *
4  * This package is an SSL implementation written
5  * by Eric Young (eay@cryptsoft.com).
6  * The implementation was written so as to conform with Netscapes SSL.
7  *
8  * This library is free for commercial and non-commercial use as long as
9  * the following conditions are aheared to.  The following conditions
10  * apply to all code found in this distribution, be it the RC4, RSA,
11  * lhash, DES, etc., code; not just the SSL code.  The SSL documentation
12  * included with this distribution is covered by the same copyright terms
13  * except that the holder is Tim Hudson (tjh@cryptsoft.com).
14  *
15  * Copyright remains Eric Young's, and as such any Copyright notices in
16  * the code are not to be removed.
17  * If this package is used in a product, Eric Young should be given attribution
18  * as the author of the parts of the library used.
19  * This can be in the form of a textual message at program startup or
20  * in documentation (online or textual) provided with the package.
21  *
22  * Redistribution and use in source and binary forms, with or without
23  * modification, are permitted provided that the following conditions
24  * are met:
25  * 1. Redistributions of source code must retain the copyright
26  *    notice, this list of conditions and the following disclaimer.
27  * 2. Redistributions in binary form must reproduce the above copyright
28  *    notice, this list of conditions and the following disclaimer in the
29  *    documentation and/or other materials provided with the distribution.
30  * 3. All advertising materials mentioning features or use of this software
31  *    must display the following acknowledgement:
32  *    "This product includes cryptographic software written by
33  *     Eric Young (eay@cryptsoft.com)"
34  *    The word 'cryptographic' can be left out if the rouines from the library
35  *    being used are not cryptographic related :-).
36  * 4. If you include any Windows specific code (or a derivative thereof) from
37  *    the apps directory (application code) you must include an acknowledgement:
38  *    "This product includes software written by Tim Hudson (tjh@cryptsoft.com)"
39  *
40  * THIS SOFTWARE IS PROVIDED BY ERIC YOUNG ``AS IS'' AND
41  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
42  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
43  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
44  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
45  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
46  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
47  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
48  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
49  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
50  * SUCH DAMAGE.
51  *
52  * The licence and distribution terms for any publically available version or
53  * derivative of this code cannot be changed.  i.e. this code cannot simply be
54  * copied and put under another distribution licence
55  * [including the GNU Public Licence.]
56  */
57 /* ====================================================================
58  * Copyright (c) 1998-2006 The OpenSSL Project.  All rights reserved.
59  *
60  * Redistribution and use in source and binary forms, with or without
61  * modification, are permitted provided that the following conditions
62  * are met:
63  *
64  * 1. Redistributions of source code must retain the above copyright
65  *    notice, this list of conditions and the following disclaimer.
66  *
67  * 2. Redistributions in binary form must reproduce the above copyright
68  *    notice, this list of conditions and the following disclaimer in
69  *    the documentation and/or other materials provided with the
70  *    distribution.
71  *
72  * 3. All advertising materials mentioning features or use of this
73  *    software must display the following acknowledgment:
74  *    "This product includes software developed by the OpenSSL Project
75  *    for use in the OpenSSL Toolkit. (http://www.openssl.org/)"
76  *
77  * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
78  *    endorse or promote products derived from this software without
79  *    prior written permission. For written permission, please contact
80  *    openssl-core@openssl.org.
81  *
82  * 5. Products derived from this software may not be called "OpenSSL"
83  *    nor may "OpenSSL" appear in their names without prior written
84  *    permission of the OpenSSL Project.
85  *
86  * 6. Redistributions of any form whatsoever must retain the following
87  *    acknowledgment:
88  *    "This product includes software developed by the OpenSSL Project
89  *    for use in the OpenSSL Toolkit (http://www.openssl.org/)"
90  *
91  * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
92  * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
93  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
94  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE OpenSSL PROJECT OR
95  * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
96  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
97  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
98  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
99  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
100  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
101  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
102  * OF THE POSSIBILITY OF SUCH DAMAGE.
103  * ====================================================================
104  *
105  * This product includes cryptographic software written by Eric Young
106  * (eay@cryptsoft.com).  This product includes software written by Tim
107  * Hudson (tjh@cryptsoft.com).
108  *
109  */
110 /* ====================================================================
111  * Copyright 2002 Sun Microsystems, Inc. ALL RIGHTS RESERVED.
112  *
113  * Portions of the attached software ("Contribution") are developed by
114  * SUN MICROSYSTEMS, INC., and are contributed to the OpenSSL project.
115  *
116  * The Contribution is licensed pursuant to the Eric Young open source
117  * license provided above.
118  *
119  * The binary polynomial arithmetic software is originally written by
120  * Sheueling Chang Shantz and Douglas Stebila of Sun Microsystems
121  * Laboratories. */
122 
123 #ifndef OPENSSL_HEADER_BN_INTERNAL_H
124 #define OPENSSL_HEADER_BN_INTERNAL_H
125 
126 #include <openssl/bn.h>
127 
128 #if defined(OPENSSL_X86_64) && defined(_MSC_VER)
129 OPENSSL_MSVC_PRAGMA(warning(push, 3))
130 #include <intrin.h>
OPENSSL_MSVC_PRAGMA(warning (pop))131 OPENSSL_MSVC_PRAGMA(warning(pop))
132 #pragma intrinsic(__umulh, _umul128)
133 #endif
134 
135 #include "../../internal.h"
136 
137 #if defined(__cplusplus)
138 extern "C" {
139 #endif
140 
141 #if defined(OPENSSL_64_BIT)
142 
143 #if defined(BORINGSSL_HAS_UINT128)
144 // MSVC doesn't support two-word integers on 64-bit.
145 #define BN_ULLONG uint128_t
146 #if defined(BORINGSSL_CAN_DIVIDE_UINT128)
147 #define BN_CAN_DIVIDE_ULLONG
148 #endif
149 #endif
150 
151 #define BN_BITS2 64
152 #define BN_BYTES 8
153 #define BN_BITS4 32
154 #define BN_MASK2 (0xffffffffffffffffUL)
155 #define BN_MASK2l (0xffffffffUL)
156 #define BN_MASK2h (0xffffffff00000000UL)
157 #define BN_MASK2h1 (0xffffffff80000000UL)
158 #define BN_MONT_CTX_N0_LIMBS 1
159 #define BN_DEC_CONV (10000000000000000000UL)
160 #define BN_DEC_NUM 19
161 #define TOBN(hi, lo) ((BN_ULONG)(hi) << 32 | (lo))
162 
163 #elif defined(OPENSSL_32_BIT)
164 
165 #define BN_ULLONG uint64_t
166 #define BN_CAN_DIVIDE_ULLONG
167 #define BN_BITS2 32
168 #define BN_BYTES 4
169 #define BN_BITS4 16
170 #define BN_MASK2 (0xffffffffUL)
171 #define BN_MASK2l (0xffffUL)
172 #define BN_MASK2h1 (0xffff8000UL)
173 #define BN_MASK2h (0xffff0000UL)
174 // On some 32-bit platforms, Montgomery multiplication is done using 64-bit
175 // arithmetic with SIMD instructions. On such platforms, |BN_MONT_CTX::n0|
176 // needs to be two words long. Only certain 32-bit platforms actually make use
177 // of n0[1] and shorter R value would suffice for the others. However,
178 // currently only the assembly files know which is which.
179 #define BN_MONT_CTX_N0_LIMBS 2
180 #define BN_DEC_CONV (1000000000UL)
181 #define BN_DEC_NUM 9
182 #define TOBN(hi, lo) (lo), (hi)
183 
184 #else
185 #error "Must define either OPENSSL_32_BIT or OPENSSL_64_BIT"
186 #endif
187 
188 #if !defined(OPENSSL_NO_ASM) && (defined(__GNUC__) || defined(__clang__))
189 #define BN_CAN_USE_INLINE_ASM
190 #endif
191 
192 // MOD_EXP_CTIME_ALIGN is the alignment needed for |BN_mod_exp_mont_consttime|'s
193 // tables.
194 //
195 // TODO(davidben): Historically, this alignment came from cache line
196 // assumptions, which we've since removed. Is 64-byte alignment still necessary
197 // or ideal? The true alignment requirement seems to now be 32 bytes, coming
198 // from RSAZ's use of VMOVDQA to a YMM register. Non-x86_64 has even fewer
199 // requirements.
200 #define MOD_EXP_CTIME_ALIGN 64
201 
202 // MOD_EXP_CTIME_STORAGE_LEN is the number of |BN_ULONG|s needed for the
203 // |BN_mod_exp_mont_consttime| stack-allocated storage buffer. The buffer is
204 // just the right size for the RSAZ and is about ~1KB larger than what's
205 // necessary (4480 bytes) for 1024-bit inputs.
206 #define MOD_EXP_CTIME_STORAGE_LEN \
207   (((320u * 3u) + (32u * 9u * 16u)) / sizeof(BN_ULONG))
208 
209 #define STATIC_BIGNUM(x)                                    \
210   {                                                         \
211     (BN_ULONG *)(x), sizeof(x) / sizeof(BN_ULONG),          \
212         sizeof(x) / sizeof(BN_ULONG), 0, BN_FLG_STATIC_DATA \
213   }
214 
215 #if defined(BN_ULLONG)
216 #define Lw(t) ((BN_ULONG)(t))
217 #define Hw(t) ((BN_ULONG)((t) >> BN_BITS2))
218 #endif
219 
220 // bn_minimal_width returns the minimal number of words needed to represent
221 // |bn|.
222 int bn_minimal_width(const BIGNUM *bn);
223 
224 // bn_set_minimal_width sets |bn->width| to |bn_minimal_width(bn)|. If |bn| is
225 // zero, |bn->neg| is set to zero.
226 void bn_set_minimal_width(BIGNUM *bn);
227 
228 // bn_wexpand ensures that |bn| has at least |words| works of space without
229 // altering its value. It returns one on success or zero on allocation
230 // failure.
231 int bn_wexpand(BIGNUM *bn, size_t words);
232 
233 // bn_expand acts the same as |bn_wexpand|, but takes a number of bits rather
234 // than a number of words.
235 int bn_expand(BIGNUM *bn, size_t bits);
236 
237 // bn_resize_words adjusts |bn->width| to be |words|. It returns one on success
238 // and zero on allocation error or if |bn|'s value is too large.
239 OPENSSL_EXPORT int bn_resize_words(BIGNUM *bn, size_t words);
240 
241 // bn_select_words sets |r| to |a| if |mask| is all ones or |b| if |mask| is
242 // all zeros.
243 void bn_select_words(BN_ULONG *r, BN_ULONG mask, const BN_ULONG *a,
244                      const BN_ULONG *b, size_t num);
245 
246 // bn_set_words sets |bn| to the value encoded in the |num| words in |words|,
247 // least significant word first.
248 int bn_set_words(BIGNUM *bn, const BN_ULONG *words, size_t num);
249 
250 // bn_set_static_words acts like |bn_set_words|, but doesn't copy the data. A
251 // flag is set on |bn| so that |BN_free| won't attempt to free the data.
252 //
253 // The |STATIC_BIGNUM| macro is probably a better solution for this outside of
254 // the FIPS module. Inside of the FIPS module that macro generates rel.ro data,
255 // which doesn't work with FIPS requirements.
256 void bn_set_static_words(BIGNUM *bn, const BN_ULONG *words, size_t num);
257 
258 // bn_fits_in_words returns one if |bn| may be represented in |num| words, plus
259 // a sign bit, and zero otherwise.
260 int bn_fits_in_words(const BIGNUM *bn, size_t num);
261 
262 // bn_copy_words copies the value of |bn| to |out| and returns one if the value
263 // is representable in |num| words. Otherwise, it returns zero.
264 int bn_copy_words(BN_ULONG *out, size_t num, const BIGNUM *bn);
265 
266 // bn_assert_fits_in_bytes asserts that |bn| fits in |num| bytes. This is a
267 // no-op in release builds, but triggers an assert in debug builds, and
268 // declassifies all bytes which are therefore known to be zero in constant-time
269 // validation.
270 void bn_assert_fits_in_bytes(const BIGNUM *bn, size_t num);
271 
272 // bn_mul_add_words multiples |ap| by |w|, adds the result to |rp|, and places
273 // the result in |rp|. |ap| and |rp| must both be |num| words long. It returns
274 // the carry word of the operation. |ap| and |rp| may be equal but otherwise may
275 // not alias.
276 BN_ULONG bn_mul_add_words(BN_ULONG *rp, const BN_ULONG *ap, size_t num,
277                           BN_ULONG w);
278 
279 // bn_mul_words multiples |ap| by |w| and places the result in |rp|. |ap| and
280 // |rp| must both be |num| words long. It returns the carry word of the
281 // operation. |ap| and |rp| may be equal but otherwise may not alias.
282 BN_ULONG bn_mul_words(BN_ULONG *rp, const BN_ULONG *ap, size_t num, BN_ULONG w);
283 
284 // bn_sqr_words sets |rp[2*i]| and |rp[2*i+1]| to |ap[i]|'s square, for all |i|
285 // up to |num|. |ap| is an array of |num| words and |rp| an array of |2*num|
286 // words. |ap| and |rp| may not alias.
287 //
288 // This gives the contribution of the |ap[i]*ap[i]| terms when squaring |ap|.
289 void bn_sqr_words(BN_ULONG *rp, const BN_ULONG *ap, size_t num);
290 
291 // bn_add_words adds |ap| to |bp| and places the result in |rp|, each of which
292 // are |num| words long. It returns the carry bit, which is one if the operation
293 // overflowed and zero otherwise. Any pair of |ap|, |bp|, and |rp| may be equal
294 // to each other but otherwise may not alias.
295 BN_ULONG bn_add_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
296                       size_t num);
297 
298 // bn_sub_words subtracts |bp| from |ap| and places the result in |rp|. It
299 // returns the borrow bit, which is one if the computation underflowed and zero
300 // otherwise. Any pair of |ap|, |bp|, and |rp| may be equal to each other but
301 // otherwise may not alias.
302 BN_ULONG bn_sub_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
303                       size_t num);
304 
305 // bn_mul_comba4 sets |r| to the product of |a| and |b|.
306 void bn_mul_comba4(BN_ULONG r[8], const BN_ULONG a[4], const BN_ULONG b[4]);
307 
308 // bn_mul_comba8 sets |r| to the product of |a| and |b|.
309 void bn_mul_comba8(BN_ULONG r[16], const BN_ULONG a[8], const BN_ULONG b[8]);
310 
311 // bn_sqr_comba8 sets |r| to |a|^2.
312 void bn_sqr_comba8(BN_ULONG r[16], const BN_ULONG a[8]);
313 
314 // bn_sqr_comba4 sets |r| to |a|^2.
315 void bn_sqr_comba4(BN_ULONG r[8], const BN_ULONG a[4]);
316 
317 // bn_less_than_words returns one if |a| < |b| and zero otherwise, where |a|
318 // and |b| both are |len| words long. It runs in constant time.
319 int bn_less_than_words(const BN_ULONG *a, const BN_ULONG *b, size_t len);
320 
321 // bn_in_range_words returns one if |min_inclusive| <= |a| < |max_exclusive|,
322 // where |a| and |max_exclusive| both are |len| words long. |a| and
323 // |max_exclusive| are treated as secret.
324 int bn_in_range_words(const BN_ULONG *a, BN_ULONG min_inclusive,
325                       const BN_ULONG *max_exclusive, size_t len);
326 
327 // bn_rand_range_words sets |out| to a uniformly distributed random number from
328 // |min_inclusive| to |max_exclusive|. Both |out| and |max_exclusive| are |len|
329 // words long.
330 //
331 // This function runs in time independent of the result, but |min_inclusive| and
332 // |max_exclusive| are public data. (Information about the range is unavoidably
333 // leaked by how many iterations it took to select a number.)
334 int bn_rand_range_words(BN_ULONG *out, BN_ULONG min_inclusive,
335                         const BN_ULONG *max_exclusive, size_t len,
336                         const uint8_t additional_data[32]);
337 
338 // bn_range_secret_range behaves like |BN_rand_range_ex|, but treats
339 // |max_exclusive| as secret. Because of this constraint, the distribution of
340 // values returned is more complex.
341 //
342 // Rather than repeatedly generating values until one is in range, which would
343 // leak information, it generates one value. If the value is in range, it sets
344 // |*out_is_uniform| to one. Otherwise, it sets |*out_is_uniform| to zero,
345 // fixing up the value to force it in range.
346 //
347 // The subset of calls to |bn_rand_secret_range| which set |*out_is_uniform| to
348 // one are uniformly distributed in the target range. Calls overall are not.
349 // This function is intended for use in situations where the extra values are
350 // still usable and where the number of iterations needed to reach the target
351 // number of uniform outputs may be blinded for negligible probabilities of
352 // timing leaks.
353 //
354 // Although this function treats |max_exclusive| as secret, it treats the number
355 // of bits in |max_exclusive| as public.
356 int bn_rand_secret_range(BIGNUM *r, int *out_is_uniform, BN_ULONG min_inclusive,
357                          const BIGNUM *max_exclusive);
358 
359 // BN_MONTGOMERY_MAX_WORDS is the maximum numer of words allowed in a |BIGNUM|
360 // used with Montgomery reduction. Ideally this limit would be applied to all
361 // |BIGNUM|s, in |bn_wexpand|, but the exactfloat library needs to create 8 MiB
362 // values for other operations.
363 #define BN_MONTGOMERY_MAX_WORDS (8 * 1024 / sizeof(BN_ULONG))
364 
365 #if !defined(OPENSSL_NO_ASM) &&                         \
366     (defined(OPENSSL_X86) || defined(OPENSSL_X86_64) || \
367      defined(OPENSSL_ARM) || defined(OPENSSL_AARCH64))
368 #define OPENSSL_BN_ASM_MONT
369 // bn_mul_mont writes |ap| * |bp| mod |np| to |rp|, each |num| words
370 // long. Inputs and outputs are in Montgomery form. |n0| is a pointer to the
371 // corresponding field in |BN_MONT_CTX|. It returns one if |bn_mul_mont| handles
372 // inputs of this size and zero otherwise.
373 //
374 // If at least one of |ap| or |bp| is fully reduced, |rp| will be fully reduced.
375 // If neither is fully-reduced, the output may not be either.
376 //
377 // This function allocates |num| words on the stack, so |num| should be at most
378 // |BN_MONTGOMERY_MAX_WORDS|.
379 //
380 // TODO(davidben): The x86_64 implementation expects a 32-bit input and masks
381 // off upper bits. The aarch64 implementation expects a 64-bit input and does
382 // not. |size_t| is the safer option but not strictly correct for x86_64. But
383 // the |BN_MONTGOMERY_MAX_WORDS| bound makes this moot.
384 //
385 // See also discussion in |ToWord| in abi_test.h for notes on smaller-than-word
386 // inputs.
387 int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
388                 const BN_ULONG *np, const BN_ULONG *n0, size_t num);
389 #endif
390 
391 #if !defined(OPENSSL_NO_ASM) && defined(OPENSSL_X86_64)
392 #define OPENSSL_BN_ASM_MONT5
393 
394 // bn_mul_mont_gather5 multiples loads index |power| of |table|, multiplies it
395 // by |ap| modulo |np|, and stores the result in |rp|. The values are |num|
396 // words long and represented in Montgomery form. |n0| is a pointer to the
397 // corresponding field in |BN_MONT_CTX|. |table| must be aligned to at least
398 // 16 bytes. |power| must be less than 32 and is treated as secret.
399 //
400 // WARNING: This function implements Almost Montgomery Multiplication from
401 // https://eprint.iacr.org/2011/239. The inputs do not need to be fully reduced.
402 // However, even if they are fully reduced, the output may not be.
403 void bn_mul_mont_gather5(BN_ULONG *rp, const BN_ULONG *ap,
404                          const BN_ULONG *table, const BN_ULONG *np,
405                          const BN_ULONG *n0, int num, int power);
406 
407 // bn_scatter5 stores |inp| to index |power| of |table|. |inp| and each entry of
408 // |table| are |num| words long. |power| must be less than 32 and is treated as
409 // public. |table| must be 32*|num| words long. |table| must be aligned to at
410 // least 16 bytes.
411 void bn_scatter5(const BN_ULONG *inp, size_t num, BN_ULONG *table,
412                  size_t power);
413 
414 // bn_gather5 loads index |power| of |table| and stores it in |out|. |out| and
415 // each entry of |table| are |num| words long. |power| must be less than 32 and
416 // is treated as secret. |table| must be aligned to at least 16 bytes.
417 void bn_gather5(BN_ULONG *out, size_t num, const BN_ULONG *table, size_t power);
418 
419 // bn_power5 squares |ap| five times and multiplies it by the value stored at
420 // index |power| of |table|, modulo |np|. It stores the result in |rp|. The
421 // values are |num| words long and represented in Montgomery form. |n0| is a
422 // pointer to the corresponding field in |BN_MONT_CTX|. |num| must be divisible
423 // by 8. |power| must be less than 32 and is treated as secret.
424 //
425 // WARNING: This function implements Almost Montgomery Multiplication from
426 // https://eprint.iacr.org/2011/239. The inputs do not need to be fully reduced.
427 // However, even if they are fully reduced, the output may not be.
428 void bn_power5(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *table,
429                const BN_ULONG *np, const BN_ULONG *n0, int num, int power);
430 #endif  // !OPENSSL_NO_ASM && OPENSSL_X86_64
431 
432 uint64_t bn_mont_n0(const BIGNUM *n);
433 
434 // bn_mont_ctx_set_RR_consttime initializes |mont->RR|. It returns one on
435 // success and zero on error. |mont->N| and |mont->n0| must have been
436 // initialized already. The bit width of |mont->N| is assumed public, but
437 // |mont->N| is otherwise treated as secret.
438 int bn_mont_ctx_set_RR_consttime(BN_MONT_CTX *mont, BN_CTX *ctx);
439 
440 #if defined(_MSC_VER)
441 #if defined(OPENSSL_X86_64)
442 #define BN_UMULT_LOHI(low, high, a, b) ((low) = _umul128((a), (b), &(high)))
443 #elif defined(OPENSSL_AARCH64)
444 #define BN_UMULT_LOHI(low, high, a, b) \
445   do {                                 \
446     const BN_ULONG _a = (a);           \
447     const BN_ULONG _b = (b);           \
448     (low) = _a * _b;                   \
449     (high) = __umulh(_a, _b);          \
450   } while (0)
451 #endif
452 #endif  // _MSC_VER
453 
454 #if !defined(BN_ULLONG) && !defined(BN_UMULT_LOHI)
455 #error "Either BN_ULLONG or BN_UMULT_LOHI must be defined on every platform."
456 #endif
457 
458 // bn_jacobi returns the Jacobi symbol of |a| and |b| (which is -1, 0 or 1), or
459 // -2 on error.
460 int bn_jacobi(const BIGNUM *a, const BIGNUM *b, BN_CTX *ctx);
461 
462 // bn_is_bit_set_words returns one if bit |bit| is set in |a| and zero
463 // otherwise.
464 int bn_is_bit_set_words(const BN_ULONG *a, size_t num, size_t bit);
465 
466 // bn_one_to_montgomery sets |r| to one in Montgomery form. It returns one on
467 // success and zero on error. This function treats the bit width of the modulus
468 // as public.
469 int bn_one_to_montgomery(BIGNUM *r, const BN_MONT_CTX *mont, BN_CTX *ctx);
470 
471 // bn_less_than_montgomery_R returns one if |bn| is less than the Montgomery R
472 // value for |mont| and zero otherwise.
473 int bn_less_than_montgomery_R(const BIGNUM *bn, const BN_MONT_CTX *mont);
474 
475 // bn_mod_u16_consttime returns |bn| mod |d|, ignoring |bn|'s sign bit. It runs
476 // in time independent of the value of |bn|, but it treats |d| as public.
477 OPENSSL_EXPORT uint16_t bn_mod_u16_consttime(const BIGNUM *bn, uint16_t d);
478 
479 // bn_odd_number_is_obviously_composite returns one if |bn| is divisible by one
480 // of the first several odd primes and zero otherwise.
481 int bn_odd_number_is_obviously_composite(const BIGNUM *bn);
482 
483 // A BN_MILLER_RABIN stores state common to each Miller-Rabin iteration. It is
484 // initialized within an existing |BN_CTX| scope and may not be used after
485 // that scope is released with |BN_CTX_end|. Field names match those in FIPS
486 // 186-4, section C.3.1.
487 typedef struct {
488   // w1 is w-1.
489   BIGNUM *w1;
490   // m is (w-1)/2^a.
491   BIGNUM *m;
492   // one_mont is 1 (mod w) in Montgomery form.
493   BIGNUM *one_mont;
494   // w1_mont is w-1 (mod w) in Montgomery form.
495   BIGNUM *w1_mont;
496   // w_bits is BN_num_bits(w).
497   int w_bits;
498   // a is the largest integer such that 2^a divides w-1.
499   int a;
500 } BN_MILLER_RABIN;
501 
502 // bn_miller_rabin_init initializes |miller_rabin| for testing if |mont->N| is
503 // prime. It returns one on success and zero on error.
504 OPENSSL_EXPORT int bn_miller_rabin_init(BN_MILLER_RABIN *miller_rabin,
505                                         const BN_MONT_CTX *mont, BN_CTX *ctx);
506 
507 // bn_miller_rabin_iteration performs one Miller-Rabin iteration, checking if
508 // |b| is a composite witness for |mont->N|. |miller_rabin| must have been
509 // initialized with |bn_miller_rabin_setup|. On success, it returns one and sets
510 // |*out_is_possibly_prime| to one if |mont->N| may still be prime or zero if
511 // |b| shows it is composite. On allocation or internal failure, it returns
512 // zero.
513 OPENSSL_EXPORT int bn_miller_rabin_iteration(
514     const BN_MILLER_RABIN *miller_rabin, int *out_is_possibly_prime,
515     const BIGNUM *b, const BN_MONT_CTX *mont, BN_CTX *ctx);
516 
517 // bn_rshift1_words sets |r| to |a| >> 1, where both arrays are |num| bits wide.
518 void bn_rshift1_words(BN_ULONG *r, const BN_ULONG *a, size_t num);
519 
520 // bn_rshift_words sets |r| to |a| >> |shift|, where both arrays are |num| bits
521 // wide.
522 void bn_rshift_words(BN_ULONG *r, const BN_ULONG *a, unsigned shift,
523                      size_t num);
524 
525 // bn_rshift_secret_shift behaves like |BN_rshift| but runs in time independent
526 // of both |a| and |n|.
527 OPENSSL_EXPORT int bn_rshift_secret_shift(BIGNUM *r, const BIGNUM *a,
528                                           unsigned n, BN_CTX *ctx);
529 
530 // bn_reduce_once sets |r| to |a| mod |m| where 0 <= |a| < 2*|m|. It returns
531 // zero if |a| < |m| and a mask of all ones if |a| >= |m|. Each array is |num|
532 // words long, but |a| has an additional word specified by |carry|. |carry| must
533 // be zero or one, as implied by the bounds on |a|.
534 //
535 // |r|, |a|, and |m| may not alias. Use |bn_reduce_once_in_place| if |r| and |a|
536 // must alias.
537 BN_ULONG bn_reduce_once(BN_ULONG *r, const BN_ULONG *a, BN_ULONG carry,
538                         const BN_ULONG *m, size_t num);
539 
540 // bn_reduce_once_in_place behaves like |bn_reduce_once| but acts in-place on
541 // |r|, using |tmp| as scratch space. |r|, |tmp|, and |m| may not alias.
542 BN_ULONG bn_reduce_once_in_place(BN_ULONG *r, BN_ULONG carry, const BN_ULONG *m,
543                                  BN_ULONG *tmp, size_t num);
544 
545 
546 // Constant-time non-modular arithmetic.
547 //
548 // The following functions implement non-modular arithmetic in constant-time
549 // and pessimally set |r->width| to the largest possible word size.
550 //
551 // Note this means that, e.g., repeatedly multiplying by one will cause widths
552 // to increase without bound. The corresponding public API functions minimize
553 // their outputs to avoid regressing calculator consumers.
554 
555 // bn_uadd_consttime behaves like |BN_uadd|, but it pessimally sets
556 // |r->width| = |a->width| + |b->width| + 1.
557 int bn_uadd_consttime(BIGNUM *r, const BIGNUM *a, const BIGNUM *b);
558 
559 // bn_usub_consttime behaves like |BN_usub|, but it pessimally sets
560 // |r->width| = |a->width|.
561 int bn_usub_consttime(BIGNUM *r, const BIGNUM *a, const BIGNUM *b);
562 
563 // bn_abs_sub_consttime sets |r| to the absolute value of |a| - |b|, treating
564 // both inputs as secret. It returns one on success and zero on error.
565 OPENSSL_EXPORT int bn_abs_sub_consttime(BIGNUM *r, const BIGNUM *a,
566                                         const BIGNUM *b, BN_CTX *ctx);
567 
568 // bn_mul_consttime behaves like |BN_mul|, but it rejects negative inputs and
569 // pessimally sets |r->width| to |a->width| + |b->width|, to avoid leaking
570 // information about |a| and |b|.
571 int bn_mul_consttime(BIGNUM *r, const BIGNUM *a, const BIGNUM *b, BN_CTX *ctx);
572 
573 // bn_sqrt_consttime behaves like |BN_sqrt|, but it pessimally sets |r->width|
574 // to 2*|a->width|, to avoid leaking information about |a| and |b|.
575 int bn_sqr_consttime(BIGNUM *r, const BIGNUM *a, BN_CTX *ctx);
576 
577 // bn_div_consttime behaves like |BN_div|, but it rejects negative inputs and
578 // treats both inputs, including their magnitudes, as secret. It is, as a
579 // result, much slower than |BN_div| and should only be used for rare operations
580 // where Montgomery reduction is not available. |divisor_min_bits| is a
581 // public lower bound for |BN_num_bits(divisor)|. When |divisor|'s bit width is
582 // public, this can speed up the operation.
583 //
584 // Note that |quotient->width| will be set pessimally to |numerator->width|.
585 OPENSSL_EXPORT int bn_div_consttime(BIGNUM *quotient, BIGNUM *remainder,
586                                     const BIGNUM *numerator,
587                                     const BIGNUM *divisor,
588                                     unsigned divisor_min_bits, BN_CTX *ctx);
589 
590 // bn_is_relatively_prime checks whether GCD(|x|, |y|) is one. On success, it
591 // returns one and sets |*out_relatively_prime| to one if the GCD was one and
592 // zero otherwise. On error, it returns zero.
593 OPENSSL_EXPORT int bn_is_relatively_prime(int *out_relatively_prime,
594                                           const BIGNUM *x, const BIGNUM *y,
595                                           BN_CTX *ctx);
596 
597 // bn_lcm_consttime sets |r| to LCM(|a|, |b|). It returns one and success and
598 // zero on error. |a| and |b| are both treated as secret.
599 OPENSSL_EXPORT int bn_lcm_consttime(BIGNUM *r, const BIGNUM *a, const BIGNUM *b,
600                                     BN_CTX *ctx);
601 
602 // bn_mont_ctx_init zero-initialies |mont|.
603 void bn_mont_ctx_init(BN_MONT_CTX *mont);
604 
605 // bn_mont_ctx_cleanup releases memory associated with |mont|, without freeing
606 // |mont| itself.
607 void bn_mont_ctx_cleanup(BN_MONT_CTX *mont);
608 
609 
610 // Constant-time modular arithmetic.
611 //
612 // The following functions implement basic constant-time modular arithmetic.
613 
614 // bn_mod_add_words sets |r| to |a| + |b| (mod |m|), using |tmp| as scratch
615 // space. Each array is |num| words long. |a| and |b| must be < |m|. Any pair of
616 // |r|, |a|, and |b| may alias.
617 void bn_mod_add_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
618                       const BN_ULONG *m, BN_ULONG *tmp, size_t num);
619 
620 // bn_mod_add_consttime acts like |BN_mod_add_quick| but takes a |BN_CTX|.
621 int bn_mod_add_consttime(BIGNUM *r, const BIGNUM *a, const BIGNUM *b,
622                          const BIGNUM *m, BN_CTX *ctx);
623 
624 // bn_mod_sub_words sets |r| to |a| - |b| (mod |m|), using |tmp| as scratch
625 // space. Each array is |num| words long. |a| and |b| must be < |m|. Any pair of
626 // |r|, |a|, and |b| may alias.
627 void bn_mod_sub_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b,
628                       const BN_ULONG *m, BN_ULONG *tmp, size_t num);
629 
630 // bn_mod_sub_consttime acts like |BN_mod_sub_quick| but takes a |BN_CTX|.
631 int bn_mod_sub_consttime(BIGNUM *r, const BIGNUM *a, const BIGNUM *b,
632                          const BIGNUM *m, BN_CTX *ctx);
633 
634 // bn_mod_lshift1_consttime acts like |BN_mod_lshift1_quick| but takes a
635 // |BN_CTX|.
636 int bn_mod_lshift1_consttime(BIGNUM *r, const BIGNUM *a, const BIGNUM *m,
637                              BN_CTX *ctx);
638 
639 // bn_mod_lshift_consttime acts like |BN_mod_lshift_quick| but takes a |BN_CTX|.
640 int bn_mod_lshift_consttime(BIGNUM *r, const BIGNUM *a, int n, const BIGNUM *m,
641                             BN_CTX *ctx);
642 
643 // bn_mod_inverse_consttime sets |r| to |a|^-1, mod |n|. |a| must be non-
644 // negative and less than |n|. It returns one on success and zero on error. On
645 // failure, if the failure was caused by |a| having no inverse mod |n| then
646 // |*out_no_inverse| will be set to one; otherwise it will be set to zero.
647 //
648 // This function treats both |a| and |n| as secret, provided they are both non-
649 // zero and the inverse exists. It should only be used for even moduli where
650 // none of the less general implementations are applicable.
651 OPENSSL_EXPORT int bn_mod_inverse_consttime(BIGNUM *r, int *out_no_inverse,
652                                             const BIGNUM *a, const BIGNUM *n,
653                                             BN_CTX *ctx);
654 
655 // bn_mod_inverse_prime sets |out| to the modular inverse of |a| modulo |p|,
656 // computed with Fermat's Little Theorem. It returns one on success and zero on
657 // error. If |mont_p| is NULL, one will be computed temporarily.
658 int bn_mod_inverse_prime(BIGNUM *out, const BIGNUM *a, const BIGNUM *p,
659                          BN_CTX *ctx, const BN_MONT_CTX *mont_p);
660 
661 // bn_mod_inverse_secret_prime behaves like |bn_mod_inverse_prime| but uses
662 // |BN_mod_exp_mont_consttime| instead of |BN_mod_exp_mont| in hopes of
663 // protecting the exponent.
664 int bn_mod_inverse_secret_prime(BIGNUM *out, const BIGNUM *a, const BIGNUM *p,
665                                 BN_CTX *ctx, const BN_MONT_CTX *mont_p);
666 
667 // BN_MONT_CTX_set_locked takes |lock| and checks whether |*pmont| is NULL. If
668 // so, it creates a new |BN_MONT_CTX| and sets the modulus for it to |mod|. It
669 // then stores it as |*pmont|. It returns one on success and zero on error. Note
670 // this function assumes |mod| is public.
671 //
672 // If |*pmont| is already non-NULL then it does nothing and returns one.
673 int BN_MONT_CTX_set_locked(BN_MONT_CTX **pmont, CRYPTO_MUTEX *lock,
674                            const BIGNUM *mod, BN_CTX *bn_ctx);
675 
676 
677 // Low-level operations for small numbers.
678 //
679 // The following functions implement algorithms suitable for use with scalars
680 // and field elements in elliptic curves. They rely on the number being small
681 // both to stack-allocate various temporaries and because they do not implement
682 // optimizations useful for the larger values used in RSA.
683 
684 // BN_SMALL_MAX_WORDS is the largest size input these functions handle. This
685 // limit allows temporaries to be more easily stack-allocated. This limit is set
686 // to accommodate P-521.
687 #if defined(OPENSSL_32_BIT)
688 #define BN_SMALL_MAX_WORDS 17
689 #else
690 #define BN_SMALL_MAX_WORDS 9
691 #endif
692 
693 // bn_mul_small sets |r| to |a|*|b|. |num_r| must be |num_a| + |num_b|. |r| may
694 // not alias with |a| or |b|.
695 void bn_mul_small(BN_ULONG *r, size_t num_r, const BN_ULONG *a, size_t num_a,
696                  const BN_ULONG *b, size_t num_b);
697 
698 // bn_sqr_small sets |r| to |a|^2. |num_a| must be at most |BN_SMALL_MAX_WORDS|.
699 // |num_r| must be |num_a|*2. |r| and |a| may not alias.
700 void bn_sqr_small(BN_ULONG *r, size_t num_r, const BN_ULONG *a, size_t num_a);
701 
702 // In the following functions, the modulus must be at most |BN_SMALL_MAX_WORDS|
703 // words long.
704 
705 // bn_to_montgomery_small sets |r| to |a| translated to the Montgomery domain.
706 // |r| and |a| are |num| words long, which must be |mont->N.width|. |a| must be
707 // fully reduced and may alias |r|.
708 void bn_to_montgomery_small(BN_ULONG *r, const BN_ULONG *a, size_t num,
709                             const BN_MONT_CTX *mont);
710 
711 // bn_from_montgomery_small sets |r| to |a| translated out of the Montgomery
712 // domain. |r| and |a| are |num_r| and |num_a| words long, respectively. |num_r|
713 // must be |mont->N.width|. |a| must be at most |mont->N|^2 and may alias |r|.
714 //
715 // Unlike most of these functions, only |num_r| is bounded by
716 // |BN_SMALL_MAX_WORDS|. |num_a| may exceed it, but must be at most 2 * |num_r|.
717 void bn_from_montgomery_small(BN_ULONG *r, size_t num_r, const BN_ULONG *a,
718                               size_t num_a, const BN_MONT_CTX *mont);
719 
720 // bn_mod_mul_montgomery_small sets |r| to |a| * |b| mod |mont->N|. Both inputs
721 // and outputs are in the Montgomery domain. Each array is |num| words long,
722 // which must be |mont->N.width|. Any two of |r|, |a|, and |b| may alias. |a|
723 // and |b| must be reduced on input.
724 void bn_mod_mul_montgomery_small(BN_ULONG *r, const BN_ULONG *a,
725                                  const BN_ULONG *b, size_t num,
726                                  const BN_MONT_CTX *mont);
727 
728 // bn_mod_exp_mont_small sets |r| to |a|^|p| mod |mont->N|. It returns one on
729 // success and zero on programmer or internal error. Both inputs and outputs are
730 // in the Montgomery domain. |r| and |a| are |num| words long, which must be
731 // |mont->N.width| and at most |BN_SMALL_MAX_WORDS|. |num_p|, measured in bits,
732 // must fit in |size_t|. |a| must be fully-reduced. This function runs in time
733 // independent of |a|, but |p| and |mont->N| are public values. |a| must be
734 // fully-reduced and may alias with |r|.
735 //
736 // Note this function differs from |BN_mod_exp_mont| which uses Montgomery
737 // reduction but takes input and output outside the Montgomery domain. Combine
738 // this function with |bn_from_montgomery_small| and |bn_to_montgomery_small|
739 // if necessary.
740 void bn_mod_exp_mont_small(BN_ULONG *r, const BN_ULONG *a, size_t num,
741                            const BN_ULONG *p, size_t num_p,
742                            const BN_MONT_CTX *mont);
743 
744 // bn_mod_inverse0_prime_mont_small sets |r| to |a|^-1 mod |mont->N|. If |a| is
745 // zero, |r| is set to zero. |mont->N| must be a prime. |r| and |a| are |num|
746 // words long, which must be |mont->N.width| and at most |BN_SMALL_MAX_WORDS|.
747 // |a| must be fully-reduced and may alias |r|. This function runs in time
748 // independent of |a|, but |mont->N| is a public value.
749 void bn_mod_inverse0_prime_mont_small(BN_ULONG *r, const BN_ULONG *a,
750                                       size_t num, const BN_MONT_CTX *mont);
751 
752 
753 // Word-based byte conversion functions.
754 
755 // bn_big_endian_to_words interprets |in_len| bytes from |in| as a big-endian,
756 // unsigned integer and writes the result to |out_len| words in |out|. |out_len|
757 // must be large enough to represent any |in_len|-byte value. That is, |out_len|
758 // must be at least |BN_BYTES * in_len|.
759 void bn_big_endian_to_words(BN_ULONG *out, size_t out_len, const uint8_t *in,
760                             size_t in_len);
761 
762 // bn_words_to_big_endian represents |in_len| words from |in| as a big-endian,
763 // unsigned integer in |out_len| bytes. It writes the result to |out|. |out_len|
764 // must be large enough to represent |in| without truncation.
765 //
766 // Note |out_len| may be less than |BN_BYTES * in_len| if |in| is known to have
767 // leading zeros.
768 void bn_words_to_big_endian(uint8_t *out, size_t out_len, const BN_ULONG *in,
769                             size_t in_len);
770 
771 
772 #if defined(__cplusplus)
773 }  // extern C
774 #endif
775 
776 #endif  // OPENSSL_HEADER_BN_INTERNAL_H
777