1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
|
/*
* Copyright 1995-2022 The OpenSSL Project Authors. All Rights Reserved.
*
* Licensed under the Apache License 2.0 (the "License"). You may not use
* this file except in compliance with the License. You can obtain a copy
* in the file LICENSE in the source distribution or at
* https://www.openssl.org/source/license.html
*/
#include <assert.h>
#include <openssl/bn.h>
#include "internal/cryptlib.h"
#include "bn_local.h"
/* The old slow way */
#if 0
int BN_div(BIGNUM *dv, BIGNUM *rem, const BIGNUM *m, const BIGNUM *d,
BN_CTX *ctx)
{
int i, nm, nd;
int ret = 0;
BIGNUM *D;
bn_check_top(m);
bn_check_top(d);
if (BN_is_zero(d)) {
ERR_raise(ERR_LIB_BN, BN_R_DIV_BY_ZERO);
return 0;
}
if (BN_ucmp(m, d) < 0) {
if (rem != NULL) {
if (BN_copy(rem, m) == NULL)
return 0;
}
if (dv != NULL)
BN_zero(dv);
return 1;
}
BN_CTX_start(ctx);
D = BN_CTX_get(ctx);
if (dv == NULL)
dv = BN_CTX_get(ctx);
if (rem == NULL)
rem = BN_CTX_get(ctx);
if (D == NULL || dv == NULL || rem == NULL)
goto end;
nd = BN_num_bits(d);
nm = BN_num_bits(m);
if (BN_copy(D, d) == NULL)
goto end;
if (BN_copy(rem, m) == NULL)
goto end;
/*
* The next 2 are needed so we can do a dv->d[0]|=1 later since
* BN_lshift1 will only work once there is a value :-)
*/
BN_zero(dv);
if (bn_wexpand(dv, 1) == NULL)
goto end;
dv->top = 1;
if (!BN_lshift(D, D, nm - nd))
goto end;
for (i = nm - nd; i >= 0; i--) {
if (!BN_lshift1(dv, dv))
goto end;
if (BN_ucmp(rem, D) >= 0) {
dv->d[0] |= 1;
if (!BN_usub(rem, rem, D))
goto end;
}
/* CAN IMPROVE (and have now :=) */
if (!BN_rshift1(D, D))
goto end;
}
rem->neg = BN_is_zero(rem) ? 0 : m->neg;
dv->neg = m->neg ^ d->neg;
ret = 1;
end:
BN_CTX_end(ctx);
return ret;
}
#else
# if defined(BN_DIV3W)
BN_ULONG bn_div_3_words(const BN_ULONG *m, BN_ULONG d1, BN_ULONG d0);
# elif 0
/*
* This is #if-ed away, because it's a reference for assembly implementations,
* where it can and should be made constant-time. But if you want to test it,
* just replace 0 with 1.
*/
# if BN_BITS2 == 64 && defined(__SIZEOF_INT128__) && __SIZEOF_INT128__==16
# undef BN_ULLONG
# define BN_ULLONG uint128_t
# define BN_LLONG
# endif
# ifdef BN_LLONG
# define BN_DIV3W
/*
* Interface is somewhat quirky, |m| is pointer to most significant limb,
* and less significant limb is referred at |m[-1]|. This means that caller
* is responsible for ensuring that |m[-1]| is valid. Second condition that
* has to be met is that |d0|'s most significant bit has to be set. Or in
* other words divisor has to be "bit-aligned to the left." bn_div_fixed_top
* does all this. The subroutine considers four limbs, two of which are
* "overlapping," hence the name...
*/
static BN_ULONG bn_div_3_words(const BN_ULONG *m, BN_ULONG d1, BN_ULONG d0)
{
BN_ULLONG R = ((BN_ULLONG)m[0] << BN_BITS2) | m[-1];
BN_ULLONG D = ((BN_ULLONG)d0 << BN_BITS2) | d1;
BN_ULONG Q = 0, mask;
int i;
for (i = 0; i < BN_BITS2; i++) {
Q <<= 1;
if (R >= D) {
Q |= 1;
R -= D;
}
D >>= 1;
}
mask = 0 - (Q >> (BN_BITS2 - 1)); /* does it overflow? */
Q <<= 1;
Q |= (R >= D);
return (Q | mask) & BN_MASK2;
}
# endif
# endif
static int bn_left_align(BIGNUM *num)
{
BN_ULONG *d = num->d, n, m, rmask;
int top = num->top;
int rshift = BN_num_bits_word(d[top - 1]), lshift, i;
lshift = BN_BITS2 - rshift;
rshift %= BN_BITS2; /* say no to undefined behaviour */
rmask = (BN_ULONG)0 - rshift; /* rmask = 0 - (rshift != 0) */
rmask |= rmask >> 8;
for (i = 0, m = 0; i < top; i++) {
n = d[i];
d[i] = ((n << lshift) | m) & BN_MASK2;
m = (n >> rshift) & rmask;
}
return lshift;
}
# if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) \
&& !defined(PEDANTIC) && !defined(BN_DIV3W)
# if defined(__GNUC__) && __GNUC__>=2
# if defined(__i386) || defined (__i386__)
/*-
* There were two reasons for implementing this template:
* - GNU C generates a call to a function (__udivdi3 to be exact)
* in reply to ((((BN_ULLONG)n0)<<BN_BITS2)|n1)/d0 (I fail to
* understand why...);
* - divl doesn't only calculate quotient, but also leaves
* remainder in %edx which we can definitely use here:-)
*/
# undef bn_div_words
# define bn_div_words(n0,n1,d0) \
({ asm volatile ( \
"divl %4" \
: "=a"(q), "=d"(rem) \
: "a"(n1), "d"(n0), "r"(d0) \
: "cc"); \
q; \
})
# define REMAINDER_IS_ALREADY_CALCULATED
# elif defined(__x86_64) && defined(SIXTY_FOUR_BIT_LONG)
/*
* Same story here, but it's 128-bit by 64-bit division. Wow!
*/
# undef bn_div_words
# define bn_div_words(n0,n1,d0) \
({ asm volatile ( \
"divq %4" \
: "=a"(q), "=d"(rem) \
: "a"(n1), "d"(n0), "r"(d0) \
: "cc"); \
q; \
})
# define REMAINDER_IS_ALREADY_CALCULATED
# endif /* __<cpu> */
# endif /* __GNUC__ */
# endif /* OPENSSL_NO_ASM */
/*-
* BN_div computes dv := num / divisor, rounding towards
* zero, and sets up rm such that dv*divisor + rm = num holds.
* Thus:
* dv->neg == num->neg ^ divisor->neg (unless the result is zero)
* rm->neg == num->neg (unless the remainder is zero)
* If 'dv' or 'rm' is NULL, the respective value is not returned.
*/
int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
BN_CTX *ctx)
{
int ret;
if (BN_is_zero(divisor)) {
ERR_raise(ERR_LIB_BN, BN_R_DIV_BY_ZERO);
return 0;
}
/*
* Invalid zero-padding would have particularly bad consequences so don't
* just rely on bn_check_top() here (bn_check_top() works only for
* BN_DEBUG builds)
*/
if (divisor->d[divisor->top - 1] == 0) {
ERR_raise(ERR_LIB_BN, BN_R_NOT_INITIALIZED);
return 0;
}
ret = bn_div_fixed_top(dv, rm, num, divisor, ctx);
if (ret) {
if (dv != NULL)
bn_correct_top(dv);
if (rm != NULL)
bn_correct_top(rm);
}
return ret;
}
/*
* It's argued that *length* of *significant* part of divisor is public.
* Even if it's private modulus that is. Again, *length* is assumed
* public, but not *value*. Former is likely to be pre-defined by
* algorithm with bit granularity, though below subroutine is invariant
* of limb length. Thanks to this assumption we can require that |divisor|
* may not be zero-padded, yet claim this subroutine "constant-time"(*).
* This is because zero-padded dividend, |num|, is tolerated, so that
* caller can pass dividend of public length(*), but with smaller amount
* of significant limbs. This naturally means that quotient, |dv|, would
* contain correspongly less significant limbs as well, and will be zero-
* padded accordingly. Returned remainder, |rm|, will have same bit length
* as divisor, also zero-padded if needed. These actually leave sign bits
* in ambiguous state. In sense that we try to avoid negative zeros, while
* zero-padded zeros would retain sign.
*
* (*) "Constant-time-ness" has two pre-conditions:
*
* - availability of constant-time bn_div_3_words;
* - dividend is at least as "wide" as divisor, limb-wise, zero-padded
* if so required, which shouldn't be a privacy problem, because
* divisor's length is considered public;
*/
int bn_div_fixed_top(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num,
const BIGNUM *divisor, BN_CTX *ctx)
{
int norm_shift, i, j, loop;
BIGNUM *tmp, *snum, *sdiv, *res;
BN_ULONG *resp, *wnum, *wnumtop;
BN_ULONG d0, d1;
int num_n, div_n, num_neg;
assert(divisor->top > 0 && divisor->d[divisor->top - 1] != 0);
bn_check_top(num);
bn_check_top(divisor);
bn_check_top(dv);
bn_check_top(rm);
BN_CTX_start(ctx);
res = (dv == NULL) ? BN_CTX_get(ctx) : dv;
tmp = BN_CTX_get(ctx);
snum = BN_CTX_get(ctx);
sdiv = BN_CTX_get(ctx);
if (sdiv == NULL)
goto err;
/* First we normalise the numbers */
if (!BN_copy(sdiv, divisor))
goto err;
norm_shift = bn_left_align(sdiv);
sdiv->neg = 0;
/*
* Note that bn_lshift_fixed_top's output is always one limb longer
* than input, even when norm_shift is zero. This means that amount of
* inner loop iterations is invariant of dividend value, and that one
* doesn't need to compare dividend and divisor if they were originally
* of the same bit length.
*/
if (!(bn_lshift_fixed_top(snum, num, norm_shift)))
goto err;
div_n = sdiv->top;
num_n = snum->top;
if (num_n <= div_n) {
/* caller didn't pad dividend -> no constant-time guarantee... */
if (bn_wexpand(snum, div_n + 1) == NULL)
goto err;
memset(&(snum->d[num_n]), 0, (div_n - num_n + 1) * sizeof(BN_ULONG));
snum->top = num_n = div_n + 1;
}
loop = num_n - div_n;
/*
* Lets setup a 'window' into snum This is the part that corresponds to
* the current 'area' being divided
*/
wnum = &(snum->d[loop]);
wnumtop = &(snum->d[num_n - 1<
|