Coverage Report

Created: 2025-06-13 06:58

/src/openssl31/crypto/ec/curve448/curve448.c
Line
Count
Source (jump to first uncovered line)
1
/*
2
 * Copyright 2017-2022 The OpenSSL Project Authors. All Rights Reserved.
3
 * Copyright 2015-2016 Cryptography Research, Inc.
4
 *
5
 * Licensed under the Apache License 2.0 (the "License").  You may not use
6
 * this file except in compliance with the License.  You can obtain a copy
7
 * in the file LICENSE in the source distribution or at
8
 * https://www.openssl.org/source/license.html
9
 *
10
 * Originally written by Mike Hamburg
11
 */
12
#include <openssl/crypto.h>
13
#include "word.h"
14
#include "field.h"
15
16
#include "point_448.h"
17
#include "ed448.h"
18
#include "crypto/ecx.h"
19
#include "curve448_local.h"
20
21
182
#define COFACTOR 4
22
23
55
#define C448_WNAF_FIXED_TABLE_BITS 5
24
55
#define C448_WNAF_VAR_TABLE_BITS 3
25
26
6.55k
#define EDWARDS_D       (-39081)
27
28
static const curve448_scalar_t precomputed_scalarmul_adjustment = {
29
    {
30
        {
31
            SC_LIMB(0xc873d6d54a7bb0cfULL), SC_LIMB(0xe933d8d723a70aadULL),
32
            SC_LIMB(0xbb124b65129c96fdULL), SC_LIMB(0x00000008335dc163ULL)
33
        }
34
    }
35
};
36
37
613
#define TWISTED_D (EDWARDS_D - 1)
38
39
91.8k
#define WBITS C448_WORD_BITS   /* NB this may be different from ARCH_WORD_BITS */
40
41
/* Inverse. */
42
static void gf_invert(gf y, const gf x, int assert_nonzero)
43
116
{
44
116
    mask_t ret;
45
116
    gf t1, t2;
46
47
116
    gf_sqr(t1, x);              /* o^2 */
48
116
    ret = gf_isr(t2, t1);       /* +-1/sqrt(o^2) = +-1/o */
49
116
    (void)ret;
50
116
    if (assert_nonzero)
51
25
        assert(ret);
52
116
    gf_sqr(t1, t2);
53
116
    gf_mul(t2, t1, x);          /* not direct to y in case of alias. */
54
116
    gf_copy(y, t2);
55
116
}
56
57
/** identity = (0,1) */
58
const curve448_point_t ossl_curve448_point_identity =
59
    { {{{{0}}}, {{{1}}}, {{{1}}}, {{{0}}}} };
60
61
static void point_double_internal(curve448_point_t p, const curve448_point_t q,
62
                                  int before_double)
63
26.0k
{
64
26.0k
    gf a, b, c, d;
65
66
26.0k
    gf_sqr(c, q->x);
67
26.0k
    gf_sqr(a, q->y);
68
26.0k
    gf_add_nr(d, c, a);         /* 2+e */
69
26.0k
    gf_add_nr(p->t, q->y, q->x); /* 2+e */
70
26.0k
    gf_sqr(b, p->t);
71
26.0k
    gf_subx_nr(b, b, d, 3);     /* 4+e */
72
26.0k
    gf_sub_nr(p->t, a, c);      /* 3+e */
73
26.0k
    gf_sqr(p->x, q->z);
74
26.0k
    gf_add_nr(p->z, p->x, p->x); /* 2+e */
75
26.0k
    gf_subx_nr(a, p->z, p->t, 4); /* 6+e */
76
26.0k
    if (GF_HEADROOM == 5)
77
0
        gf_weak_reduce(a);      /* or 1+e */
78
26.0k
    gf_mul(p->x, a, b);
79
26.0k
    gf_mul(p->z, p->t, a);
80
26.0k
    gf_mul(p->y, p->t, d);
81
26.0k
    if (!before_double)
82
7.20k
        gf_mul(p->t, b, d);
83
26.0k
}
84
85
void ossl_curve448_point_double(curve448_point_t p, const curve448_point_t q)
86
55
{
87
55
    point_double_internal(p, q, 0);
88
55
}
89
90
/* Operations on [p]niels */
91
static ossl_inline void cond_neg_niels(niels_t n, mask_t neg)
92
9.27k
{
93
9.27k
    gf_cond_swap(n->a, n->b, neg);
94
9.27k
    gf_cond_neg(n->c, neg);
95
9.27k
}
96
97
static void pt_to_pniels(pniels_t b, const curve448_point_t a)
98
495
{
99
495
    gf_sub(b->n->a, a->y, a->x);
100
495
    gf_add(b->n->b, a->x, a->y);
101
495
    gf_mulw(b->n->c, a->t, 2 * TWISTED_D);
102
495
    gf_add(b->z, a->z, a->z);
103
495
}
104
105
static void pniels_to_pt(curve448_point_t e, const pniels_t d)
106
50
{
107
50
    gf eu;
108
109
50
    gf_add(eu, d->n->b, d->n->a);
110
50
    gf_sub(e->y, d->n->b, d->n->a);
111
50
    gf_mul(e->t, e->y, eu);
112
50
    gf_mul(e->x, d->z, e->y);
113
50
    gf_mul(e->y, d->z, eu);
114
50
    gf_sqr(e->z, d->z);
115
50
}
116
117
static void niels_to_pt(curve448_point_t e, const niels_t n)
118
108
{
119
108
    gf_add(e->y, n->b, n->a);
120
108
    gf_sub(e->x, n->b, n->a);
121
108
    gf_mul(e->t, e->y, e->x);
122
108
    gf_copy(e->z, ONE);
123
108
}
124
125
static void add_niels_to_pt(curve448_point_t d, const niels_t e,
126
                            int before_double)
127
12.5k
{
128
12.5k
    gf a, b, c;
129
130
12.5k
    gf_sub_nr(b, d->y, d->x);   /* 3+e */
131
12.5k
    gf_mul(a, e->a, b);
132
12.5k
    gf_add_nr(b, d->x, d->y);   /* 2+e */
133
12.5k
    gf_mul(d->y, e->b, b);
134
12.5k
    gf_mul(d->x, e->c, d->t);
135
12.5k
    gf_add_nr(c, a, d->y);      /* 2+e */
136
12.5k
    gf_sub_nr(b, d->y, a);      /* 3+e */
137
12.5k
    gf_sub_nr(d->y, d->z, d->x); /* 3+e */
138
12.5k
    gf_add_nr(a, d->x, d->z);   /* 2+e */
139
12.5k
    gf_mul(d->z, a, d->y);
140
12.5k
    gf_mul(d->x, d->y, b);
141
12.5k
    gf_mul(d->y, a, c);
142
12.5k
    if (!before_double)
143
7.95k
        gf_mul(d->t, b, c);
144
12.5k
}
145
146
static void sub_niels_from_pt(curve448_point_t d, const niels_t e,
147
                              int before_double)
148
2.68k
{
149
2.68k
    gf a, b, c;
150
151
2.68k
    gf_sub_nr(b, d->y, d->x);   /* 3+e */
152
2.68k
    gf_mul(a, e->b, b);
153
2.68k
    gf_add_nr(b, d->x, d->y);   /* 2+e */
154
2.68k
    gf_mul(d->y, e->a, b);
155
2.68k
    gf_mul(d->x, e->c, d->t);
156
2.68k
    gf_add_nr(c, a, d->y);      /* 2+e */
157
2.68k
    gf_sub_nr(b, d->y, a);      /* 3+e */
158
2.68k
    gf_add_nr(d->y, d->z, d->x); /* 2+e */
159
2.68k
    gf_sub_nr(a, d->z, d->x);   /* 3+e */
160
2.68k
    gf_mul(d->z, a, d->y);
161
2.68k
    gf_mul(d->x, d->y, b);
162
2.68k
    gf_mul(d->y, a, c);
163
2.68k
    if (!before_double)
164
151
        gf_mul(d->t, b, c);
165
2.68k
}
166
167
static void add_pniels_to_pt(curve448_point_t p, const pniels_t pn,
168
                             int before_double)
169
2.31k
{
170
2.31k
    gf L0;
171
172
2.31k
    gf_mul(L0, p->z, pn->z);
173
2.31k
    gf_copy(p->z, L0);
174
2.31k
    add_niels_to_pt(p, pn->n, before_double);
175
2.31k
}
176
177
static void sub_pniels_from_pt(curve448_point_t p, const pniels_t pn,
178
                               int before_double)
179
2.04k
{
180
2.04k
    gf L0;
181
182
2.04k
    gf_mul(L0, p->z, pn->z);
183
2.04k
    gf_copy(p->z, L0);
184
2.04k
    sub_niels_from_pt(p, pn->n, before_double);
185
2.04k
}
186
187
c448_bool_t
188
ossl_curve448_point_eq(const curve448_point_t p,
189
                       const curve448_point_t q)
190
55
{
191
55
    mask_t succ;
192
55
    gf a, b;
193
194
    /* equality mod 2-torsion compares x/y */
195
55
    gf_mul(a, p->y, q->x);
196
55
    gf_mul(b, q->y, p->x);
197
55
    succ = gf_eq(a, b);
198
199
55
    return mask_to_bool(succ);
200
55
}
201
202
c448_bool_t
203
ossl_curve448_point_valid(const curve448_point_t p)
204
118
{
205
118
    mask_t out;
206
118
    gf a, b, c;
207
208
118
    gf_mul(a, p->x, p->y);
209
118
    gf_mul(b, p->z, p->t);
210
118
    out = gf_eq(a, b);
211
118
    gf_sqr(a, p->x);
212
118
    gf_sqr(b, p->y);
213
118
    gf_sub(a, b, a);
214
118
    gf_sqr(b, p->t);
215
118
    gf_mulw(c, b, TWISTED_D);
216
118
    gf_sqr(b, p->z);
217
118
    gf_add(b, b, c);
218
118
    out &= gf_eq(a, b);
219
118
    out &= ~gf_eq(p->z, ZERO);
220
118
    return mask_to_bool(out);
221
118
}
222
223
static ossl_inline void constant_time_lookup_niels(niels_s * RESTRICT ni,
224
                                                   const niels_t * table,
225
                                                   int nelts, int idx)
226
9.27k
{
227
9.27k
    constant_time_lookup(ni, table, sizeof(niels_s), nelts, idx);
228
9.27k
}
229
230
void
231
ossl_curve448_precomputed_scalarmul(curve448_point_t out,
232
                                    const curve448_precomputed_s * table,
233
                                    const curve448_scalar_t scalar)
234
103
{
235
103
    unsigned int i, j, k;
236
103
    const unsigned int n = COMBS_N, t = COMBS_T, s = COMBS_S;
237
103
    niels_t ni;
238
103
    curve448_scalar_t scalar1x;
239
240
103
    ossl_curve448_scalar_add(scalar1x, scalar, precomputed_scalarmul_adjustment);
241
103
    ossl_curve448_scalar_halve(scalar1x, scalar1x);
242
243
1.95k
    for (i = s; i > 0; i--) {
244
1.85k
        if (i != s)
245
1.75k
            point_double_internal(out, out, 0);
246
247
11.1k
        for (j = 0; j < n; j++) {
248
9.27k
            int tab = 0;
249
9.27k
            mask_t invert;
250
251
55.6k
            for (k = 0; k < t; k++) {
252
46.3k
                unsigned int bit = (i - 1) + s * (k + j * t);
253
254
46.3k
                if (bit < C448_SCALAR_BITS)
255
45.9k
                    tab |=
256
45.9k
                        (scalar1x->limb[bit / WBITS] >> (bit % WBITS) & 1) << k;
257
46.3k
            }
258
259
9.27k
            invert = (tab >> (t - 1)) - 1;
260
9.27k
            tab ^= invert;
261
9.27k
            tab &= (1 << (t - 1)) - 1;
262
263
9.27k
            constant_time_lookup_niels(ni, &table->table[j << (t - 1)],
264
9.27k
                                       1 << (t - 1), tab);
265
266
9.27k
            cond_neg_niels(ni, invert);
267
9.27k
            if ((i != s) || j != 0)
268
9.16k
                add_niels_to_pt(out, ni, j == n - 1 && i != 1);
269
103
            else
270
103
                niels_to_pt(out, ni);
271
9.27k
        }
272
1.85k
    }
273
274
103
    OPENSSL_cleanse(ni, sizeof(ni));
275
103
    OPENSSL_cleanse(scalar1x, sizeof(scalar1x));
276
103
}
277
278
void
279
ossl_curve448_point_mul_by_ratio_and_encode_like_eddsa(
280
                                    uint8_t enc[EDDSA_448_PUBLIC_BYTES],
281
                                    const curve448_point_t p)
282
25
{
283
25
    gf x, y, z, t;
284
25
    curve448_point_t q;
285
286
    /* The point is now on the twisted curve.  Move it to untwisted. */
287
25
    curve448_point_copy(q, p);
288
289
25
    {
290
        /* 4-isogeny: 2xy/(y^+x^2), (y^2-x^2)/(2z^2-y^2+x^2) */
291
25
        gf u;
292
293
25
        gf_sqr(x, q->x);
294
25
        gf_sqr(t, q->y);
295
25
        gf_add(u, x, t);
296
25
        gf_add(z, q->y, q->x);
297
25
        gf_sqr(y, z);
298
25
        gf_sub(y, y, u);
299
25
        gf_sub(z, t, x);
300
25
        gf_sqr(x, q->z);
301
25
        gf_add(t, x, x);
302
25
        gf_sub(t, t, z);
303
25
        gf_mul(x, t, y);
304
25
        gf_mul(y, z, u);
305
25
        gf_mul(z, u, t);
306
25
        OPENSSL_cleanse(u, sizeof(u));
307
25
    }
308
309
    /* Affinize */
310
25
    gf_invert(z, z, 1);
311
25
    gf_mul(t, x, z);
312
25
    gf_mul(x, y, z);
313
314
    /* Encode */
315
25
    enc[EDDSA_448_PRIVATE_BYTES - 1] = 0;
316
25
    gf_serialize(enc, x, 1);
317
25
    enc[EDDSA_448_PRIVATE_BYTES - 1] |= 0x80 & gf_lobit(t);
318
319
25
    OPENSSL_cleanse(x, sizeof(x));
320
25
    OPENSSL_cleanse(y, sizeof(y));
321
25
    OPENSSL_cleanse(z, sizeof(z));
322
25
    OPENSSL_cleanse(t, sizeof(t));
323
25
    ossl_curve448_point_destroy(q);
324
25
}
325
326
c448_error_t
327
ossl_curve448_point_decode_like_eddsa_and_mul_by_ratio(
328
                                curve448_point_t p,
329
                                const uint8_t enc[EDDSA_448_PUBLIC_BYTES])
330
118
{
331
118
    uint8_t enc2[EDDSA_448_PUBLIC_BYTES];
332
118
    mask_t low;
333
118
    mask_t succ;
334
335
118
    memcpy(enc2, enc, sizeof(enc2));
336
337
118
    low = ~word_is_zero(enc2[EDDSA_448_PRIVATE_BYTES - 1] & 0x80);
338
118
    enc2[EDDSA_448_PRIVATE_BYTES - 1] &= ~0x80;
339
340
118
    succ = gf_deserialize(p->y, enc2, 1, 0);
341
118
    succ &= word_is_zero(enc2[EDDSA_448_PRIVATE_BYTES - 1]);
342
343
118
    gf_sqr(p->x, p->y);
344
118
    gf_sub(p->z, ONE, p->x);    /* num = 1-y^2 */
345
118
    gf_mulw(p->t, p->x, EDWARDS_D); /* dy^2 */
346
118
    gf_sub(p->t, ONE, p->t);    /* denom = 1-dy^2 or 1-d + dy^2 */
347
348
118
    gf_mul(p->x, p->z, p->t);
349
118
    succ &= gf_isr(p->t, p->x); /* 1/sqrt(num * denom) */
350
351
118
    gf_mul(p->x, p->t, p->z);   /* sqrt(num / denom) */
352
118
    gf_cond_neg(p->x, gf_lobit(p->x) ^ low);
353
118
    gf_copy(p->z, ONE);
354
355
118
    {
356
118
        gf a, b, c, d;
357
358
        /* 4-isogeny 2xy/(y^2-ax^2), (y^2+ax^2)/(2-y^2-ax^2) */
359
118
        gf_sqr(c, p->x);
360
118
        gf_sqr(a, p->y);
361
118
        gf_add(d, c, a);
362
118
        gf_add(p->t, p->y, p->x);
363
118
        gf_sqr(b, p->t);
364
118
        gf_sub(b, b, d);
365
118
        gf_sub(p->t, a, c);
366
118
        gf_sqr(p->x, p->z);
367
118
        gf_add(p->z, p->x, p->x);
368
118
        gf_sub(a, p->z, d);
369
118
        gf_mul(p->x, a, b);
370
118
        gf_mul(p->z, p->t, a);
371
118
        gf_mul(p->y, p->t, d);
372
118
        gf_mul(p->t, b, d);
373
118
        OPENSSL_cleanse(a, sizeof(a));
374
118
        OPENSSL_cleanse(b, sizeof(b));
375
118
        OPENSSL_cleanse(c, sizeof(c));
376
118
        OPENSSL_cleanse(d, sizeof(d));
377
118
    }
378
379
118
    OPENSSL_cleanse(enc2, sizeof(enc2));
380
118
    assert(ossl_curve448_point_valid(p) || ~succ);
381
382
118
    return c448_succeed_if(mask_to_bool(succ));
383
118
}
384
385
c448_error_t
386
ossl_x448_int(uint8_t out[X_PUBLIC_BYTES],
387
              const uint8_t base[X_PUBLIC_BYTES],
388
              const uint8_t scalar[X_PRIVATE_BYTES])
389
13
{
390
13
    gf x1, x2, z2, x3, z3, t1, t2;
391
13
    int t;
392
13
    mask_t swap = 0;
393
13
    mask_t nz;
394
395
13
    (void)gf_deserialize(x1, base, 1, 0);
396
13
    gf_copy(x2, ONE);
397
13
    gf_copy(z2, ZERO);
398
13
    gf_copy(x3, x1);
399
13
    gf_copy(z3, ONE);
400
401
5.83k
    for (t = X_PRIVATE_BITS - 1; t >= 0; t--) {
402
5.82k
        uint8_t sb = scalar[t / 8];
403
5.82k
        mask_t k_t;
404
405
        /* Scalar conditioning */
406
5.82k
        if (t / 8 == 0)
407
104
            sb &= -(uint8_t)COFACTOR;
408
5.72k
        else if (t == X_PRIVATE_BITS - 1)
409
13
            sb = -1;
410
411
5.82k
        k_t = (sb >> (t % 8)) & 1;
412
5.82k
        k_t = 0 - k_t;             /* set to all 0s or all 1s */
413
414
5.82k
        swap ^= k_t;
415
5.82k
        gf_cond_swap(x2, x3, swap);
416
5.82k
        gf_cond_swap(z2, z3, swap);
417
5.82k
        swap = k_t;
418
419
        /*
420
         * The "_nr" below skips coefficient reduction. In the following
421
         * comments, "2+e" is saying that the coefficients are at most 2+epsilon
422
         * times the reduction limit.
423
         */
424
5.82k
        gf_add_nr(t1, x2, z2);  /* A = x2 + z2 */ /* 2+e */
425
5.82k
        gf_sub_nr(t2, x2, z2);  /* B = x2 - z2 */ /* 3+e */
426
5.82k
        gf_sub_nr(z2, x3, z3);  /* D = x3 - z3 */ /* 3+e */
427
5.82k
        gf_mul(x2, t1, z2);     /* DA */
428
5.82k
        gf_add_nr(z2, z3, x3);  /* C = x3 + z3 */ /* 2+e */
429
5.82k
        gf_mul(x3, t2, z2);     /* CB */
430
5.82k
        gf_sub_nr(z3, x2, x3);  /* DA-CB */ /* 3+e */
431
5.82k
        gf_sqr(z2, z3);         /* (DA-CB)^2 */
432
5.82k
        gf_mul(z3, x1, z2);     /* z3 = x1(DA-CB)^2 */
433
5.82k
        gf_add_nr(z2, x2, x3);  /* (DA+CB) */ /* 2+e */
434
5.82k
        gf_sqr(x3, z2);         /* x3 = (DA+CB)^2 */
435
436
5.82k
        gf_sqr(z2, t1);         /* AA = A^2 */
437
5.82k
        gf_sqr(t1, t2);         /* BB = B^2 */
438
5.82k
        gf_mul(x2, z2, t1);     /* x2 = AA*BB */
439
5.82k
        gf_sub_nr(t2, z2, t1);  /* E = AA-BB */ /* 3+e */
440
441
5.82k
        gf_mulw(t1, t2, -EDWARDS_D); /* E*-d = a24*E */
442
5.82k
        gf_add_nr(t1, t1, z2);  /* AA + a24*E */ /* 2+e */
443
5.82k
        gf_mul(z2, t2, t1);     /* z2 = E(AA+a24*E) */
444
5.82k
    }
445
446
    /* Finish */
447
13
    gf_cond_swap(x2, x3, swap);
448
13
    gf_cond_swap(z2, z3, swap);
449
13
    gf_invert(z2, z2, 0);
450
13
    gf_mul(x1, x2, z2);
451
13
    gf_serialize(out, x1, 1);
452
13
    nz = ~gf_eq(x1, ZERO);
453
454
13
    OPENSSL_cleanse(x1, sizeof(x1));
455
13
    OPENSSL_cleanse(x2, sizeof(x2));
456
13
    OPENSSL_cleanse(z2, sizeof(z2));
457
13
    OPENSSL_cleanse(x3, sizeof(x3));
458
13
    OPENSSL_cleanse(z3, sizeof(z3));
459
13
    OPENSSL_cleanse(t1, sizeof(t1));
460
13
    OPENSSL_cleanse(t2, sizeof(t2));
461
462
13
    return c448_succeed_if(mask_to_bool(nz));
463
13
}
464
465
void
466
ossl_curve448_point_mul_by_ratio_and_encode_like_x448(uint8_t
467
                                                      out[X_PUBLIC_BYTES],
468
                                                      const curve448_point_t p)
469
78
{
470
78
    curve448_point_t q;
471
472
78
    curve448_point_copy(q, p);
473
78
    gf_invert(q->t, q->x, 0);   /* 1/x */
474
78
    gf_mul(q->z, q->t, q->y);   /* y/x */
475
78
    gf_sqr(q->y, q->z);         /* (y/x)^2 */
476
78
    gf_serialize(out, q->y, 1);
477
78
    ossl_curve448_point_destroy(q);
478
78
}
479
480
void ossl_x448_derive_public_key(uint8_t out[X_PUBLIC_BYTES],
481
                                 const uint8_t scalar[X_PRIVATE_BYTES])
482
78
{
483
    /* Scalar conditioning */
484
78
    uint8_t scalar2[X_PRIVATE_BYTES];
485
78
    curve448_scalar_t the_scalar;
486
78
    curve448_point_t p;
487
78
    unsigned int i;
488
489
78
    memcpy(scalar2, scalar, sizeof(scalar2));
490
78
    scalar2[0] &= -(uint8_t)COFACTOR;
491
492
78
    scalar2[X_PRIVATE_BYTES - 1] &= ~((0u - 1u) << ((X_PRIVATE_BITS + 7) % 8));
493
78
    scalar2[X_PRIVATE_BYTES - 1] |= 1 << ((X_PRIVATE_BITS + 7) % 8);
494
495
78
    ossl_curve448_scalar_decode_long(the_scalar, scalar2, sizeof(scalar2));
496
497
    /* Compensate for the encoding ratio */
498
156
    for (i = 1; i < X448_ENCODE_RATIO; i <<= 1)
499
78
        ossl_curve448_scalar_halve(the_scalar, the_scalar);
500
501
78
    ossl_curve448_precomputed_scalarmul(p, ossl_curve448_precomputed_base,
502
78
                                        the_scalar);
503
78
    ossl_curve448_point_mul_by_ratio_and_encode_like_x448(out, p);
504
78
    ossl_curve448_point_destroy(p);
505
78
}
506
507
/* Control for variable-time scalar multiply algorithms. */
508
struct smvt_control {
509
    int power, addend;
510
};
511
512
#if defined(__GNUC__) && (__GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ > 3))
513
0
# define NUMTRAILINGZEROS       __builtin_ctz
514
#else
515
# define NUMTRAILINGZEROS       numtrailingzeros
516
static uint32_t numtrailingzeros(uint32_t i)
517
{
518
    uint32_t tmp;
519
    uint32_t num = 31;
520
521
    if (i == 0)
522
        return 32;
523
524
    tmp = i << 16;
525
    if (tmp != 0) {
526
        i = tmp;
527
        num -= 16;
528
    }
529
    tmp = i << 8;
530
    if (tmp != 0) {
531
        i = tmp;
532
        num -= 8;
533
    }
534
    tmp = i << 4;
535
    if (tmp != 0) {
536
        i = tmp;
537
        num -= 4;
538
    }
539
    tmp = i << 2;
540
    if (tmp != 0) {
541
        i = tmp;
542
        num -= 2;
543
    }
544
    tmp = i << 1;
545
    if (tmp != 0)
546
        num--;
547
548
    return num;
549
}
550
#endif
551
552
static int recode_wnaf(struct smvt_control *control,
553
                       /* [nbits/(table_bits + 1) + 3] */
554
                       const curve448_scalar_t scalar,
555
                       unsigned int table_bits)
556
0
{
557
0
    unsigned int table_size = C448_SCALAR_BITS / (table_bits + 1) + 3;
558
0
    int position = table_size - 1; /* at the end */
559
0
    uint64_t current = scalar->limb[0] & 0xFFFF;
560
0
    uint32_t mask = (1 << (table_bits + 1)) - 1;
561
0
    unsigned int w;
562
0
    const unsigned int B_OVER_16 = sizeof(scalar->limb[0]) / 2;
563
0
    unsigned int n, i;
564
565
    /* place the end marker */
566
0
    control[position].power = -1;
567
0
    control[position].addend = 0;
568
0
    position--;
569
570
    /*
571
     * PERF: Could negate scalar if it's large.  But then would need more cases
572
     * in the actual code that uses it, all for an expected reduction of like
573
     * 1/5 op. Probably not worth it.
574
     */
575
576
0
    for (w = 1; w < (C448_SCALAR_BITS - 1) / 16 + 3; w++) {
577
0
        if (w < (C448_SCALAR_BITS - 1) / 16 + 1) {
578
            /* Refill the 16 high bits of current */
579
0
            current += (uint32_t)((scalar->limb[w / B_OVER_16]
580
0
                       >> (16 * (w % B_OVER_16))) << 16);
581
0
        }
582
583
0
        while (current & 0xFFFF) {
584
0
            uint32_t pos = NUMTRAILINGZEROS((uint32_t)current);
585
0
            uint32_t odd = (uint32_t)current >> pos;
586
0
            int32_t delta = odd & mask;
587
588
0
            assert(position >= 0);
589
0
            assert(pos < 32);       /* can't fail since current & 0xFFFF != 0 */
590
0
            if (odd & (1 << (table_bits + 1)))
591
0
                delta -= (1 << (table_bits + 1));
592
0
            current -= delta * (1 << pos);
593
0
            control[position].power = pos + 16 * (w - 1);
594
0
            control[position].addend = delta;
595
0
            position--;
596
0
        }
597
0
        current >>= 16;
598
0
    }
599
0
    assert(current == 0);
600
601
0
    position++;
602
0
    n = table_size - position;
603
0
    for (i = 0; i < n; i++)
604
0
        control[i] = control[i + position];
605
606
0
    return n - 1;
607
0
}
608
609
static void prepare_wnaf_table(pniels_t * output,
610
                               const curve448_point_t working,
611
                               unsigned int tbits)
612
55
{
613
55
    curve448_point_t tmp;
614
55
    int i;
615
55
    pniels_t twop;
616
617
55
    pt_to_pniels(output[0], working);
618
619
55
    if (tbits == 0)
620
0
        return;
621
622
55
    ossl_curve448_point_double(tmp, working);
623
55
    pt_to_pniels(twop, tmp);
624
625
55
    add_pniels_to_pt(tmp, output[0], 0);
626
55
    pt_to_pniels(output[1], tmp);
627
628
385
    for (i = 2; i < 1 << tbits; i++) {
629
330
        add_pniels_to_pt(tmp, twop, 0);
630
330
        pt_to_pniels(output[i], tmp);
631
330
    }
632
633
55
    ossl_curve448_point_destroy(tmp);
634
55
    OPENSSL_cleanse(twop, sizeof(twop));
635
55
}
636
637
void
638
ossl_curve448_base_double_scalarmul_non_secret(curve448_point_t combo,
639
                                               const curve448_scalar_t scalar1,
640
                                               const curve448_point_t base2,
641
                                               const curve448_scalar_t scalar2)
642
55
{
643
55
    const int table_bits_var = C448_WNAF_VAR_TABLE_BITS;
644
55
    const int table_bits_pre = C448_WNAF_FIXED_TABLE_BITS;
645
55
    struct smvt_control control_var[C448_SCALAR_BITS /
646
55
                                    (C448_WNAF_VAR_TABLE_BITS + 1) + 3];
647
55
    struct smvt_control control_pre[C448_SCALAR_BITS /
648
55
                                    (C448_WNAF_FIXED_TABLE_BITS + 1) + 3];
649
55
    int ncb_pre = recode_wnaf(control_pre, scalar1, table_bits_pre);
650
55
    int ncb_var = recode_wnaf(control_var, scalar2, table_bits_var);
651
55
    pniels_t precmp_var[1 << C448_WNAF_VAR_TABLE_BITS];
652
55
    int contp = 0, contv = 0, i;
653
654
55
    prepare_wnaf_table(precmp_var, base2, table_bits_var);
655
55
    i = control_var[0].power;
656
657
55
    if (i < 0) {
658
0
        curve448_point_copy(combo, ossl_curve448_point_identity);
659
0
        return;
660
0
    }
661
55
    if (i > control_pre[0].power) {
662
47
        pniels_to_pt(combo, precmp_var[control_var[0].addend >> 1]);
663
47
        contv++;
664
47
    } else if (i == control_pre[0].power && i >= 0) {
665
3
        pniels_to_pt(combo, precmp_var[control_var[0].addend >> 1]);
666
3
        add_niels_to_pt(combo,
667
3
                        ossl_curve448_wnaf_base[control_pre[0].addend >> 1],
668
3
                        i);
669
3
        contv++;
670
3
        contp++;
671
5
    } else {
672
5
        i = control_pre[0].power;
673
5
        niels_to_pt(combo, ossl_curve448_wnaf_base[control_pre[0].addend >> 1]);
674
5
        contp++;
675
5
    }
676
677
24.3k
    for (i--; i >= 0; i--) {
678
24.2k
        int cv = (i == control_var[contv].power);
679
24.2k
        int cp = (i == control_pre[contp].power);
680
681
24.2k
        point_double_internal(combo, combo, i && !(cv || cp));
682
683
24.2k
        if (cv) {
684
3.97k
            assert(control_var[contv].addend);
685
686
3.97k
            if (control_var[contv].addend > 0)
687
1.93k
                add_pniels_to_pt(combo,
688
1.93k
                                 precmp_var[control_var[contv].addend >> 1],
689
1.93k
                                 i && !cp);
690
2.04k
            else
691
2.04k
                sub_pniels_from_pt(combo,
692
2.04k
                                   precmp_var[(-control_var[contv].addend)
693
2.04k
                                              >> 1], i && !cp);
694
3.97k
            contv++;
695
3.97k
        }
696
697
24.2k
        if (cp) {
698
1.67k
            assert(control_pre[contp].addend);
699
700
1.67k
            if (control_pre[contp].addend > 0)
701
1.04k
                add_niels_to_pt(combo,
702
1.04k
                                ossl_curve448_wnaf_base[control_pre[contp].addend
703
1.04k
                                                   >> 1], i);
704
634
            else
705
634
                sub_niels_from_pt(combo,
706
634
                                  ossl_curve448_wnaf_base[(-control_pre
707
634
                                                      [contp].addend) >> 1], i);
708
1.67k
            contp++;
709
1.67k
        }
710
24.2k
    }
711
712
    /* This function is non-secret, but whatever this is cheap. */
713
55
    OPENSSL_cleanse(control_var, sizeof(control_var));
714
55
    OPENSSL_cleanse(control_pre, sizeof(control_pre));
715
55
    OPENSSL_cleanse(precmp_var, sizeof(precmp_var));
716
717
55
    assert(contv == ncb_var);
718
55
    (void)ncb_var;
719
55
    assert(contp == ncb_pre);
720
55
    (void)ncb_pre;
721
55
}
722
723
void ossl_curve448_point_destroy(curve448_point_t point)
724
261
{
725
261
    OPENSSL_cleanse(point, sizeof(curve448_point_t));
726
261
}
727
728
int ossl_x448(uint8_t out_shared_key[56], const uint8_t private_key[56],
729
              const uint8_t peer_public_value[56])
730
13
{
731
13
    return ossl_x448_int(out_shared_key, peer_public_value, private_key)
732
13
           == C448_SUCCESS;
733
13
}
734
735
void ossl_x448_public_from_private(uint8_t out_public_value[56],
736
                                   const uint8_t private_key[56])
737
78
{
738
78
    ossl_x448_derive_public_key(out_public_value, private_key);
739
78
}