1 /*
2  * Copyright 2017-2021 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 #define COFACTOR 4
22 
23 #define C448_WNAF_FIXED_TABLE_BITS 5
24 #define C448_WNAF_VAR_TABLE_BITS 3
25 
26 #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 #define TWISTED_D (EDWARDS_D - 1)
38 
39 #define WBITS C448_WORD_BITS   /* NB this may be different from ARCH_WORD_BITS */
40 
41 /* Inverse. */
gf_invert(gf y,const gf x,int assert_nonzero)42 static void gf_invert(gf y, const gf x, int assert_nonzero)
43 {
44     mask_t ret;
45     gf t1, t2;
46 
47     gf_sqr(t1, x);              /* o^2 */
48     ret = gf_isr(t2, t1);       /* +-1/sqrt(o^2) = +-1/o */
49     (void)ret;
50     if (assert_nonzero)
51         assert(ret);
52     gf_sqr(t1, t2);
53     gf_mul(t2, t1, x);          /* not direct to y in case of alias. */
54     gf_copy(y, t2);
55 }
56 
57 /** identity = (0,1) */
58 const curve448_point_t ossl_curve448_point_identity =
59     { {{{{0}}}, {{{1}}}, {{{1}}}, {{{0}}}} };
60 
point_double_internal(curve448_point_t p,const curve448_point_t q,int before_double)61 static void point_double_internal(curve448_point_t p, const curve448_point_t q,
62                                   int before_double)
63 {
64     gf a, b, c, d;
65 
66     gf_sqr(c, q->x);
67     gf_sqr(a, q->y);
68     gf_add_nr(d, c, a);         /* 2+e */
69     gf_add_nr(p->t, q->y, q->x); /* 2+e */
70     gf_sqr(b, p->t);
71     gf_subx_nr(b, b, d, 3);     /* 4+e */
72     gf_sub_nr(p->t, a, c);      /* 3+e */
73     gf_sqr(p->x, q->z);
74     gf_add_nr(p->z, p->x, p->x); /* 2+e */
75     gf_subx_nr(a, p->z, p->t, 4); /* 6+e */
76     if (GF_HEADROOM == 5)
77         gf_weak_reduce(a);      /* or 1+e */
78     gf_mul(p->x, a, b);
79     gf_mul(p->z, p->t, a);
80     gf_mul(p->y, p->t, d);
81     if (!before_double)
82         gf_mul(p->t, b, d);
83 }
84 
ossl_curve448_point_double(curve448_point_t p,const curve448_point_t q)85 void ossl_curve448_point_double(curve448_point_t p, const curve448_point_t q)
86 {
87     point_double_internal(p, q, 0);
88 }
89 
90 /* Operations on [p]niels */
cond_neg_niels(niels_t n,mask_t neg)91 static ossl_inline void cond_neg_niels(niels_t n, mask_t neg)
92 {
93     gf_cond_swap(n->a, n->b, neg);
94     gf_cond_neg(n->c, neg);
95 }
96 
pt_to_pniels(pniels_t b,const curve448_point_t a)97 static void pt_to_pniels(pniels_t b, const curve448_point_t a)
98 {
99     gf_sub(b->n->a, a->y, a->x);
100     gf_add(b->n->b, a->x, a->y);
101     gf_mulw(b->n->c, a->t, 2 * TWISTED_D);
102     gf_add(b->z, a->z, a->z);
103 }
104 
pniels_to_pt(curve448_point_t e,const pniels_t d)105 static void pniels_to_pt(curve448_point_t e, const pniels_t d)
106 {
107     gf eu;
108 
109     gf_add(eu, d->n->b, d->n->a);
110     gf_sub(e->y, d->n->b, d->n->a);
111     gf_mul(e->t, e->y, eu);
112     gf_mul(e->x, d->z, e->y);
113     gf_mul(e->y, d->z, eu);
114     gf_sqr(e->z, d->z);
115 }
116 
niels_to_pt(curve448_point_t e,const niels_t n)117 static void niels_to_pt(curve448_point_t e, const niels_t n)
118 {
119     gf_add(e->y, n->b, n->a);
120     gf_sub(e->x, n->b, n->a);
121     gf_mul(e->t, e->y, e->x);
122     gf_copy(e->z, ONE);
123 }
124 
add_niels_to_pt(curve448_point_t d,const niels_t e,int before_double)125 static void add_niels_to_pt(curve448_point_t d, const niels_t e,
126                             int before_double)
127 {
128     gf a, b, c;
129 
130     gf_sub_nr(b, d->y, d->x);   /* 3+e */
131     gf_mul(a, e->a, b);
132     gf_add_nr(b, d->x, d->y);   /* 2+e */
133     gf_mul(d->y, e->b, b);
134     gf_mul(d->x, e->c, d->t);
135     gf_add_nr(c, a, d->y);      /* 2+e */
136     gf_sub_nr(b, d->y, a);      /* 3+e */
137     gf_sub_nr(d->y, d->z, d->x); /* 3+e */
138     gf_add_nr(a, d->x, d->z);   /* 2+e */
139     gf_mul(d->z, a, d->y);
140     gf_mul(d->x, d->y, b);
141     gf_mul(d->y, a, c);
142     if (!before_double)
143         gf_mul(d->t, b, c);
144 }
145 
sub_niels_from_pt(curve448_point_t d,const niels_t e,int before_double)146 static void sub_niels_from_pt(curve448_point_t d, const niels_t e,
147                               int before_double)
148 {
149     gf a, b, c;
150 
151     gf_sub_nr(b, d->y, d->x);   /* 3+e */
152     gf_mul(a, e->b, b);
153     gf_add_nr(b, d->x, d->y);   /* 2+e */
154     gf_mul(d->y, e->a, b);
155     gf_mul(d->x, e->c, d->t);
156     gf_add_nr(c, a, d->y);      /* 2+e */
157     gf_sub_nr(b, d->y, a);      /* 3+e */
158     gf_add_nr(d->y, d->z, d->x); /* 2+e */
159     gf_sub_nr(a, d->z, d->x);   /* 3+e */
160     gf_mul(d->z, a, d->y);
161     gf_mul(d->x, d->y, b);
162     gf_mul(d->y, a, c);
163     if (!before_double)
164         gf_mul(d->t, b, c);
165 }
166 
add_pniels_to_pt(curve448_point_t p,const pniels_t pn,int before_double)167 static void add_pniels_to_pt(curve448_point_t p, const pniels_t pn,
168                              int before_double)
169 {
170     gf L0;
171 
172     gf_mul(L0, p->z, pn->z);
173     gf_copy(p->z, L0);
174     add_niels_to_pt(p, pn->n, before_double);
175 }
176 
sub_pniels_from_pt(curve448_point_t p,const pniels_t pn,int before_double)177 static void sub_pniels_from_pt(curve448_point_t p, const pniels_t pn,
178                                int before_double)
179 {
180     gf L0;
181 
182     gf_mul(L0, p->z, pn->z);
183     gf_copy(p->z, L0);
184     sub_niels_from_pt(p, pn->n, before_double);
185 }
186 
187 c448_bool_t
ossl_curve448_point_eq(const curve448_point_t p,const curve448_point_t q)188 ossl_curve448_point_eq(const curve448_point_t p,
189                        const curve448_point_t q)
190 {
191     mask_t succ;
192     gf a, b;
193 
194     /* equality mod 2-torsion compares x/y */
195     gf_mul(a, p->y, q->x);
196     gf_mul(b, q->y, p->x);
197     succ = gf_eq(a, b);
198 
199     return mask_to_bool(succ);
200 }
201 
202 c448_bool_t
ossl_curve448_point_valid(const curve448_point_t p)203 ossl_curve448_point_valid(const curve448_point_t p)
204 {
205     mask_t out;
206     gf a, b, c;
207 
208     gf_mul(a, p->x, p->y);
209     gf_mul(b, p->z, p->t);
210     out = gf_eq(a, b);
211     gf_sqr(a, p->x);
212     gf_sqr(b, p->y);
213     gf_sub(a, b, a);
214     gf_sqr(b, p->t);
215     gf_mulw(c, b, TWISTED_D);
216     gf_sqr(b, p->z);
217     gf_add(b, b, c);
218     out &= gf_eq(a, b);
219     out &= ~gf_eq(p->z, ZERO);
220     return mask_to_bool(out);
221 }
222 
constant_time_lookup_niels(niels_s * RESTRICT ni,const niels_t * table,int nelts,int idx)223 static ossl_inline void constant_time_lookup_niels(niels_s * RESTRICT ni,
224                                                    const niels_t * table,
225                                                    int nelts, int idx)
226 {
227     constant_time_lookup(ni, table, sizeof(niels_s), nelts, idx);
228 }
229 
230 void
ossl_curve448_precomputed_scalarmul(curve448_point_t out,const curve448_precomputed_s * table,const curve448_scalar_t scalar)231 ossl_curve448_precomputed_scalarmul(curve448_point_t out,
232                                     const curve448_precomputed_s * table,
233                                     const curve448_scalar_t scalar)
234 {
235     unsigned int i, j, k;
236     const unsigned int n = COMBS_N, t = COMBS_T, s = COMBS_S;
237     niels_t ni;
238     curve448_scalar_t scalar1x;
239 
240     ossl_curve448_scalar_add(scalar1x, scalar, precomputed_scalarmul_adjustment);
241     ossl_curve448_scalar_halve(scalar1x, scalar1x);
242 
243     for (i = s; i > 0; i--) {
244         if (i != s)
245             point_double_internal(out, out, 0);
246 
247         for (j = 0; j < n; j++) {
248             int tab = 0;
249             mask_t invert;
250 
251             for (k = 0; k < t; k++) {
252                 unsigned int bit = (i - 1) + s * (k + j * t);
253 
254                 if (bit < C448_SCALAR_BITS)
255                     tab |=
256                         (scalar1x->limb[bit / WBITS] >> (bit % WBITS) & 1) << k;
257             }
258 
259             invert = (tab >> (t - 1)) - 1;
260             tab ^= invert;
261             tab &= (1 << (t - 1)) - 1;
262 
263             constant_time_lookup_niels(ni, &table->table[j << (t - 1)],
264                                        1 << (t - 1), tab);
265 
266             cond_neg_niels(ni, invert);
267             if ((i != s) || j != 0)
268                 add_niels_to_pt(out, ni, j == n - 1 && i != 1);
269             else
270                 niels_to_pt(out, ni);
271         }
272     }
273 
274     OPENSSL_cleanse(ni, sizeof(ni));
275     OPENSSL_cleanse(scalar1x, sizeof(scalar1x));
276 }
277 
278 void
ossl_curve448_point_mul_by_ratio_and_encode_like_eddsa(uint8_t enc[EDDSA_448_PUBLIC_BYTES],const curve448_point_t p)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 {
283     gf x, y, z, t;
284     curve448_point_t q;
285 
286     /* The point is now on the twisted curve.  Move it to untwisted. */
287     curve448_point_copy(q, p);
288 
289     {
290         /* 4-isogeny: 2xy/(y^+x^2), (y^2-x^2)/(2z^2-y^2+x^2) */
291         gf u;
292 
293         gf_sqr(x, q->x);
294         gf_sqr(t, q->y);
295         gf_add(u, x, t);
296         gf_add(z, q->y, q->x);
297         gf_sqr(y, z);
298         gf_sub(y, y, u);
299         gf_sub(z, t, x);
300         gf_sqr(x, q->z);
301         gf_add(t, x, x);
302         gf_sub(t, t, z);
303         gf_mul(x, t, y);
304         gf_mul(y, z, u);
305         gf_mul(z, u, t);
306         OPENSSL_cleanse(u, sizeof(u));
307     }
308 
309     /* Affinize */
310     gf_invert(z, z, 1);
311     gf_mul(t, x, z);
312     gf_mul(x, y, z);
313 
314     /* Encode */
315     enc[EDDSA_448_PRIVATE_BYTES - 1] = 0;
316     gf_serialize(enc, x, 1);
317     enc[EDDSA_448_PRIVATE_BYTES - 1] |= 0x80 & gf_lobit(t);
318 
319     OPENSSL_cleanse(x, sizeof(x));
320     OPENSSL_cleanse(y, sizeof(y));
321     OPENSSL_cleanse(z, sizeof(z));
322     OPENSSL_cleanse(t, sizeof(t));
323     ossl_curve448_point_destroy(q);
324 }
325 
326 c448_error_t
ossl_curve448_point_decode_like_eddsa_and_mul_by_ratio(curve448_point_t p,const uint8_t enc[EDDSA_448_PUBLIC_BYTES])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 {
331     uint8_t enc2[EDDSA_448_PUBLIC_BYTES];
332     mask_t low;
333     mask_t succ;
334 
335     memcpy(enc2, enc, sizeof(enc2));
336 
337     low = ~word_is_zero(enc2[EDDSA_448_PRIVATE_BYTES - 1] & 0x80);
338     enc2[EDDSA_448_PRIVATE_BYTES - 1] &= ~0x80;
339 
340     succ = gf_deserialize(p->y, enc2, 1, 0);
341     succ &= word_is_zero(enc2[EDDSA_448_PRIVATE_BYTES - 1]);
342 
343     gf_sqr(p->x, p->y);
344     gf_sub(p->z, ONE, p->x);    /* num = 1-y^2 */
345     gf_mulw(p->t, p->x, EDWARDS_D); /* dy^2 */
346     gf_sub(p->t, ONE, p->t);    /* denom = 1-dy^2 or 1-d + dy^2 */
347 
348     gf_mul(p->x, p->z, p->t);
349     succ &= gf_isr(p->t, p->x); /* 1/sqrt(num * denom) */
350 
351     gf_mul(p->x, p->t, p->z);   /* sqrt(num / denom) */
352     gf_cond_neg(p->x, gf_lobit(p->x) ^ low);
353     gf_copy(p->z, ONE);
354 
355     {
356         gf a, b, c, d;
357 
358         /* 4-isogeny 2xy/(y^2-ax^2), (y^2+ax^2)/(2-y^2-ax^2) */
359         gf_sqr(c, p->x);
360         gf_sqr(a, p->y);
361         gf_add(d, c, a);
362         gf_add(p->t, p->y, p->x);
363         gf_sqr(b, p->t);
364         gf_sub(b, b, d);
365         gf_sub(p->t, a, c);
366         gf_sqr(p->x, p->z);
367         gf_add(p->z, p->x, p->x);
368         gf_sub(a, p->z, d);
369         gf_mul(p->x, a, b);
370         gf_mul(p->z, p->t, a);
371         gf_mul(p->y, p->t, d);
372         gf_mul(p->t, b, d);
373         OPENSSL_cleanse(a, sizeof(a));
374         OPENSSL_cleanse(b, sizeof(b));
375         OPENSSL_cleanse(c, sizeof(c));
376         OPENSSL_cleanse(d, sizeof(d));
377     }
378 
379     OPENSSL_cleanse(enc2, sizeof(enc2));
380     assert(ossl_curve448_point_valid(p) || ~succ);
381 
382     return c448_succeed_if(mask_to_bool(succ));
383 }
384 
385 c448_error_t
ossl_x448_int(uint8_t out[X_PUBLIC_BYTES],const uint8_t base[X_PUBLIC_BYTES],const uint8_t scalar[X_PRIVATE_BYTES])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 {
390     gf x1, x2, z2, x3, z3, t1, t2;
391     int t;
392     mask_t swap = 0;
393     mask_t nz;
394 
395     (void)gf_deserialize(x1, base, 1, 0);
396     gf_copy(x2, ONE);
397     gf_copy(z2, ZERO);
398     gf_copy(x3, x1);
399     gf_copy(z3, ONE);
400 
401     for (t = X_PRIVATE_BITS - 1; t >= 0; t--) {
402         uint8_t sb = scalar[t / 8];
403         mask_t k_t;
404 
405         /* Scalar conditioning */
406         if (t / 8 == 0)
407             sb &= -(uint8_t)COFACTOR;
408         else if (t == X_PRIVATE_BITS - 1)
409             sb = -1;
410 
411         k_t = (sb >> (t % 8)) & 1;
412         k_t = 0 - k_t;             /* set to all 0s or all 1s */
413 
414         swap ^= k_t;
415         gf_cond_swap(x2, x3, swap);
416         gf_cond_swap(z2, z3, swap);
417         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         gf_add_nr(t1, x2, z2);  /* A = x2 + z2 */ /* 2+e */
425         gf_sub_nr(t2, x2, z2);  /* B = x2 - z2 */ /* 3+e */
426         gf_sub_nr(z2, x3, z3);  /* D = x3 - z3 */ /* 3+e */
427         gf_mul(x2, t1, z2);     /* DA */
428         gf_add_nr(z2, z3, x3);  /* C = x3 + z3 */ /* 2+e */
429         gf_mul(x3, t2, z2);     /* CB */
430         gf_sub_nr(z3, x2, x3);  /* DA-CB */ /* 3+e */
431         gf_sqr(z2, z3);         /* (DA-CB)^2 */
432         gf_mul(z3, x1, z2);     /* z3 = x1(DA-CB)^2 */
433         gf_add_nr(z2, x2, x3);  /* (DA+CB) */ /* 2+e */
434         gf_sqr(x3, z2);         /* x3 = (DA+CB)^2 */
435 
436         gf_sqr(z2, t1);         /* AA = A^2 */
437         gf_sqr(t1, t2);         /* BB = B^2 */
438         gf_mul(x2, z2, t1);     /* x2 = AA*BB */
439         gf_sub_nr(t2, z2, t1);  /* E = AA-BB */ /* 3+e */
440 
441         gf_mulw(t1, t2, -EDWARDS_D); /* E*-d = a24*E */
442         gf_add_nr(t1, t1, z2);  /* AA + a24*E */ /* 2+e */
443         gf_mul(z2, t2, t1);     /* z2 = E(AA+a24*E) */
444     }
445 
446     /* Finish */
447     gf_cond_swap(x2, x3, swap);
448     gf_cond_swap(z2, z3, swap);
449     gf_invert(z2, z2, 0);
450     gf_mul(x1, x2, z2);
451     gf_serialize(out, x1, 1);
452     nz = ~gf_eq(x1, ZERO);
453 
454     OPENSSL_cleanse(x1, sizeof(x1));
455     OPENSSL_cleanse(x2, sizeof(x2));
456     OPENSSL_cleanse(z2, sizeof(z2));
457     OPENSSL_cleanse(x3, sizeof(x3));
458     OPENSSL_cleanse(z3, sizeof(z3));
459     OPENSSL_cleanse(t1, sizeof(t1));
460     OPENSSL_cleanse(t2, sizeof(t2));
461 
462     return c448_succeed_if(mask_to_bool(nz));
463 }
464 
465 void
ossl_curve448_point_mul_by_ratio_and_encode_like_x448(uint8_t out[X_PUBLIC_BYTES],const curve448_point_t p)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 {
470     curve448_point_t q;
471 
472     curve448_point_copy(q, p);
473     gf_invert(q->t, q->x, 0);   /* 1/x */
474     gf_mul(q->z, q->t, q->y);   /* y/x */
475     gf_sqr(q->y, q->z);         /* (y/x)^2 */
476     gf_serialize(out, q->y, 1);
477     ossl_curve448_point_destroy(q);
478 }
479 
ossl_x448_derive_public_key(uint8_t out[X_PUBLIC_BYTES],const uint8_t scalar[X_PRIVATE_BYTES])480 void ossl_x448_derive_public_key(uint8_t out[X_PUBLIC_BYTES],
481                                  const uint8_t scalar[X_PRIVATE_BYTES])
482 {
483     /* Scalar conditioning */
484     uint8_t scalar2[X_PRIVATE_BYTES];
485     curve448_scalar_t the_scalar;
486     curve448_point_t p;
487     unsigned int i;
488 
489     memcpy(scalar2, scalar, sizeof(scalar2));
490     scalar2[0] &= -(uint8_t)COFACTOR;
491 
492     scalar2[X_PRIVATE_BYTES - 1] &= ~((0u - 1u) << ((X_PRIVATE_BITS + 7) % 8));
493     scalar2[X_PRIVATE_BYTES - 1] |= 1 << ((X_PRIVATE_BITS + 7) % 8);
494 
495     ossl_curve448_scalar_decode_long(the_scalar, scalar2, sizeof(scalar2));
496 
497     /* Compensate for the encoding ratio */
498     for (i = 1; i < X448_ENCODE_RATIO; i <<= 1)
499         ossl_curve448_scalar_halve(the_scalar, the_scalar);
500 
501     ossl_curve448_precomputed_scalarmul(p, ossl_curve448_precomputed_base,
502                                         the_scalar);
503     ossl_curve448_point_mul_by_ratio_and_encode_like_x448(out, p);
504     ossl_curve448_point_destroy(p);
505 }
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 # define NUMTRAILINGZEROS       __builtin_ctz
514 #else
515 # define NUMTRAILINGZEROS       numtrailingzeros
numtrailingzeros(uint32_t i)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 
recode_wnaf(struct smvt_control * control,const curve448_scalar_t scalar,unsigned int table_bits)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 {
557     unsigned int table_size = C448_SCALAR_BITS / (table_bits + 1) + 3;
558     int position = table_size - 1; /* at the end */
559     uint64_t current = scalar->limb[0] & 0xFFFF;
560     uint32_t mask = (1 << (table_bits + 1)) - 1;
561     unsigned int w;
562     const unsigned int B_OVER_16 = sizeof(scalar->limb[0]) / 2;
563     unsigned int n, i;
564 
565     /* place the end marker */
566     control[position].power = -1;
567     control[position].addend = 0;
568     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     for (w = 1; w < (C448_SCALAR_BITS - 1) / 16 + 3; w++) {
577         if (w < (C448_SCALAR_BITS - 1) / 16 + 1) {
578             /* Refill the 16 high bits of current */
579             current += (uint32_t)((scalar->limb[w / B_OVER_16]
580                        >> (16 * (w % B_OVER_16))) << 16);
581         }
582 
583         while (current & 0xFFFF) {
584             uint32_t pos = NUMTRAILINGZEROS((uint32_t)current);
585             uint32_t odd = (uint32_t)current >> pos;
586             int32_t delta = odd & mask;
587 
588             assert(position >= 0);
589             if (odd & (1 << (table_bits + 1)))
590                 delta -= (1 << (table_bits + 1));
591             current -= delta * (1 << pos);
592             control[position].power = pos + 16 * (w - 1);
593             control[position].addend = delta;
594             position--;
595         }
596         current >>= 16;
597     }
598     assert(current == 0);
599 
600     position++;
601     n = table_size - position;
602     for (i = 0; i < n; i++)
603         control[i] = control[i + position];
604 
605     return n - 1;
606 }
607 
prepare_wnaf_table(pniels_t * output,const curve448_point_t working,unsigned int tbits)608 static void prepare_wnaf_table(pniels_t * output,
609                                const curve448_point_t working,
610                                unsigned int tbits)
611 {
612     curve448_point_t tmp;
613     int i;
614     pniels_t twop;
615 
616     pt_to_pniels(output[0], working);
617 
618     if (tbits == 0)
619         return;
620 
621     ossl_curve448_point_double(tmp, working);
622     pt_to_pniels(twop, tmp);
623 
624     add_pniels_to_pt(tmp, output[0], 0);
625     pt_to_pniels(output[1], tmp);
626 
627     for (i = 2; i < 1 << tbits; i++) {
628         add_pniels_to_pt(tmp, twop, 0);
629         pt_to_pniels(output[i], tmp);
630     }
631 
632     ossl_curve448_point_destroy(tmp);
633     OPENSSL_cleanse(twop, sizeof(twop));
634 }
635 
636 void
ossl_curve448_base_double_scalarmul_non_secret(curve448_point_t combo,const curve448_scalar_t scalar1,const curve448_point_t base2,const curve448_scalar_t scalar2)637 ossl_curve448_base_double_scalarmul_non_secret(curve448_point_t combo,
638                                                const curve448_scalar_t scalar1,
639                                                const curve448_point_t base2,
640                                                const curve448_scalar_t scalar2)
641 {
642     const int table_bits_var = C448_WNAF_VAR_TABLE_BITS;
643     const int table_bits_pre = C448_WNAF_FIXED_TABLE_BITS;
644     struct smvt_control control_var[C448_SCALAR_BITS /
645                                     (C448_WNAF_VAR_TABLE_BITS + 1) + 3];
646     struct smvt_control control_pre[C448_SCALAR_BITS /
647                                     (C448_WNAF_FIXED_TABLE_BITS + 1) + 3];
648     int ncb_pre = recode_wnaf(control_pre, scalar1, table_bits_pre);
649     int ncb_var = recode_wnaf(control_var, scalar2, table_bits_var);
650     pniels_t precmp_var[1 << C448_WNAF_VAR_TABLE_BITS];
651     int contp = 0, contv = 0, i;
652 
653     prepare_wnaf_table(precmp_var, base2, table_bits_var);
654     i = control_var[0].power;
655 
656     if (i < 0) {
657         curve448_point_copy(combo, ossl_curve448_point_identity);
658         return;
659     }
660     if (i > control_pre[0].power) {
661         pniels_to_pt(combo, precmp_var[control_var[0].addend >> 1]);
662         contv++;
663     } else if (i == control_pre[0].power && i >= 0) {
664         pniels_to_pt(combo, precmp_var[control_var[0].addend >> 1]);
665         add_niels_to_pt(combo,
666                         ossl_curve448_wnaf_base[control_pre[0].addend >> 1],
667                         i);
668         contv++;
669         contp++;
670     } else {
671         i = control_pre[0].power;
672         niels_to_pt(combo, ossl_curve448_wnaf_base[control_pre[0].addend >> 1]);
673         contp++;
674     }
675 
676     for (i--; i >= 0; i--) {
677         int cv = (i == control_var[contv].power);
678         int cp = (i == control_pre[contp].power);
679 
680         point_double_internal(combo, combo, i && !(cv || cp));
681 
682         if (cv) {
683             assert(control_var[contv].addend);
684 
685             if (control_var[contv].addend > 0)
686                 add_pniels_to_pt(combo,
687                                  precmp_var[control_var[contv].addend >> 1],
688                                  i && !cp);
689             else
690                 sub_pniels_from_pt(combo,
691                                    precmp_var[(-control_var[contv].addend)
692                                               >> 1], i && !cp);
693             contv++;
694         }
695 
696         if (cp) {
697             assert(control_pre[contp].addend);
698 
699             if (control_pre[contp].addend > 0)
700                 add_niels_to_pt(combo,
701                                 ossl_curve448_wnaf_base[control_pre[contp].addend
702                                                    >> 1], i);
703             else
704                 sub_niels_from_pt(combo,
705                                   ossl_curve448_wnaf_base[(-control_pre
706                                                       [contp].addend) >> 1], i);
707             contp++;
708         }
709     }
710 
711     /* This function is non-secret, but whatever this is cheap. */
712     OPENSSL_cleanse(control_var, sizeof(control_var));
713     OPENSSL_cleanse(control_pre, sizeof(control_pre));
714     OPENSSL_cleanse(precmp_var, sizeof(precmp_var));
715 
716     assert(contv == ncb_var);
717     (void)ncb_var;
718     assert(contp == ncb_pre);
719     (void)ncb_pre;
720 }
721 
ossl_curve448_point_destroy(curve448_point_t point)722 void ossl_curve448_point_destroy(curve448_point_t point)
723 {
724     OPENSSL_cleanse(point, sizeof(curve448_point_t));
725 }
726 
ossl_x448(uint8_t out_shared_key[56],const uint8_t private_key[56],const uint8_t peer_public_value[56])727 int ossl_x448(uint8_t out_shared_key[56], const uint8_t private_key[56],
728               const uint8_t peer_public_value[56])
729 {
730     return ossl_x448_int(out_shared_key, peer_public_value, private_key)
731            == C448_SUCCESS;
732 }
733 
ossl_x448_public_from_private(uint8_t out_public_value[56],const uint8_t private_key[56])734 void ossl_x448_public_from_private(uint8_t out_public_value[56],
735                                    const uint8_t private_key[56])
736 {
737     ossl_x448_derive_public_key(out_public_value, private_key);
738 }
739