2 * @file ed448goldilocks/decaf.c
6 * Copyright (c) 2015-2016 Cryptography Research, Inc. \n
7 * Released under the MIT License. See LICENSE.txt for license information.
9 * @brief Decaf high-level functions.
11 * @warning This file was automatically generated in Python.
12 * Please do not edit it.
14 #include <openssl/crypto.h>
18 #include "point_448.h"
20 #include "curve448_lcl.h"
24 /* Comb config: number of combs, n, t, s. */
28 #define DECAF_WINDOW_BITS 5
29 #define DECAF_WNAF_FIXED_TABLE_BITS 5
30 #define DECAF_WNAF_VAR_TABLE_BITS 3
32 static const int EDWARDS_D = -39081;
33 static const curve448_scalar_t precomputed_scalarmul_adjustment = {{{
34 SC_LIMB(0xc873d6d54a7bb0cf), SC_LIMB(0xe933d8d723a70aad), SC_LIMB(0xbb124b65129c96fd), SC_LIMB(0x00000008335dc163)
37 const uint8_t decaf_x448_base_point[DECAF_X448_PUBLIC_BYTES] = { 0x05 };
39 #define RISTRETTO_FACTOR DECAF_448_RISTRETTO_FACTOR
40 const gf RISTRETTO_FACTOR = {{{
41 0x42ef0f45572736, 0x7bf6aa20ce5296, 0xf4fd6eded26033, 0x968c14ba839a66, 0xb8d54b64a2d780, 0x6aa0a1f1a7b8a5, 0x683bf68d722fa2, 0x22d962fbeb24f7
45 #define TWISTED_D ((EDWARDS_D)-1)
47 #define EFF_D (-(TWISTED_D))
50 /* End of template stuff */
52 #define WBITS DECAF_WORD_BITS /* NB this may be different from ARCH_WORD_BITS */
54 /* Projective Niels coordinates */
55 typedef struct { gf a, b, c; } niels_s, niels_t[1];
56 typedef struct { niels_t n; gf z; } VECTOR_ALIGNED pniels_s, pniels_t[1];
58 /* Precomputed base */
59 struct curve448_precomputed_s { niels_t table [COMBS_N<<(COMBS_T-1)]; };
61 extern const gf curve448_precomputed_base_as_fe[];
62 const curve448_precomputed_s *curve448_precomputed_base =
63 (const curve448_precomputed_s *) &curve448_precomputed_base_as_fe;
67 gf_invert(gf y, const gf x, int assert_nonzero) {
70 mask_t ret = gf_isr(t2, t1); // +-1/sqrt(o^2) = +-1/o
72 if (assert_nonzero) assert(ret);
74 gf_mul(t2, t1, x); // not direct to y in case of alias.
78 /** identity = (0,1) */
79 const curve448_point_t curve448_point_identity = {{{{{0}}},{{{1}}},{{{1}}},{{{0}}}}};
82 point_double_internal (
84 const curve448_point_t q,
90 gf_add_nr ( d, c, a ); /* 2+e */
91 gf_add_nr ( p->t, q->y, q->x ); /* 2+e */
93 gf_subx_nr ( b, b, d, 3 ); /* 4+e */
94 gf_sub_nr ( p->t, a, c ); /* 3+e */
95 gf_sqr ( p->x, q->z );
96 gf_add_nr ( p->z, p->x, p->x ); /* 2+e */
97 gf_subx_nr ( a, p->z, p->t, 4 ); /* 6+e */
98 if (GF_HEADROOM == 5) gf_weak_reduce(a); /* or 1+e */
99 gf_mul ( p->x, a, b );
100 gf_mul ( p->z, p->t, a );
101 gf_mul ( p->y, p->t, d );
102 if (!before_double) gf_mul ( p->t, b, d );
105 void curve448_point_double(curve448_point_t p, const curve448_point_t q) {
106 point_double_internal(p,q,0);
109 /* Operations on [p]niels */
110 static ossl_inline void
115 gf_cond_swap(n->a, n->b, neg);
116 gf_cond_neg(n->c, neg);
119 static void pt_to_pniels (
121 const curve448_point_t a
123 gf_sub ( b->n->a, a->y, a->x );
124 gf_add ( b->n->b, a->x, a->y );
125 gf_mulw ( b->n->c, a->t, 2*TWISTED_D );
126 gf_add ( b->z, a->z, a->z );
129 static void pniels_to_pt (
134 gf_add ( eu, d->n->b, d->n->a );
135 gf_sub ( e->y, d->n->b, d->n->a );
136 gf_mul ( e->t, e->y, eu);
137 gf_mul ( e->x, d->z, e->y );
138 gf_mul ( e->y, d->z, eu );
139 gf_sqr ( e->z, d->z );
147 gf_add ( e->y, n->b, n->a );
148 gf_sub ( e->x, n->b, n->a );
149 gf_mul ( e->t, e->y, e->x );
150 gf_copy ( e->z, ONE );
160 gf_sub_nr ( b, d->y, d->x ); /* 3+e */
161 gf_mul ( a, e->a, b );
162 gf_add_nr ( b, d->x, d->y ); /* 2+e */
163 gf_mul ( d->y, e->b, b );
164 gf_mul ( d->x, e->c, d->t );
165 gf_add_nr ( c, a, d->y ); /* 2+e */
166 gf_sub_nr ( b, d->y, a ); /* 3+e */
167 gf_sub_nr ( d->y, d->z, d->x ); /* 3+e */
168 gf_add_nr ( a, d->x, d->z ); /* 2+e */
169 gf_mul ( d->z, a, d->y );
170 gf_mul ( d->x, d->y, b );
171 gf_mul ( d->y, a, c );
172 if (!before_double) gf_mul ( d->t, b, c );
182 gf_sub_nr ( b, d->y, d->x ); /* 3+e */
183 gf_mul ( a, e->b, b );
184 gf_add_nr ( b, d->x, d->y ); /* 2+e */
185 gf_mul ( d->y, e->a, b );
186 gf_mul ( d->x, e->c, d->t );
187 gf_add_nr ( c, a, d->y ); /* 2+e */
188 gf_sub_nr ( b, d->y, a ); /* 3+e */
189 gf_add_nr ( d->y, d->z, d->x ); /* 2+e */
190 gf_sub_nr ( a, d->z, d->x ); /* 3+e */
191 gf_mul ( d->z, a, d->y );
192 gf_mul ( d->x, d->y, b );
193 gf_mul ( d->y, a, c );
194 if (!before_double) gf_mul ( d->t, b, c );
204 gf_mul ( L0, p->z, pn->z );
205 gf_copy ( p->z, L0 );
206 add_niels_to_pt( p, pn->n, before_double );
216 gf_mul ( L0, p->z, pn->z );
217 gf_copy ( p->z, L0 );
218 sub_niels_from_pt( p, pn->n, before_double );
221 decaf_bool_t curve448_point_eq ( const curve448_point_t p, const curve448_point_t q ) {
222 /* equality mod 2-torsion compares x/y */
224 gf_mul ( a, p->y, q->x );
225 gf_mul ( b, q->y, p->x );
226 mask_t succ = gf_eq(a,b);
228 return mask_to_bool(succ);
231 decaf_bool_t curve448_point_valid (
232 const curve448_point_t p
237 mask_t out = gf_eq(a,b);
242 gf_mulw(c,b,TWISTED_D);
246 out &= ~gf_eq(p->z,ZERO);
247 return mask_to_bool(out);
250 static ossl_inline void
251 constant_time_lookup_niels (
252 niels_s *__restrict__ ni,
253 const niels_t *table,
257 constant_time_lookup(ni, table, sizeof(niels_s), nelts, idx);
260 void curve448_precomputed_scalarmul (
261 curve448_point_t out,
262 const curve448_precomputed_s *table,
263 const curve448_scalar_t scalar
267 const unsigned int n = COMBS_N, t = COMBS_T, s = COMBS_S;
269 curve448_scalar_t scalar1x;
270 curve448_scalar_add(scalar1x, scalar, precomputed_scalarmul_adjustment);
271 curve448_scalar_halve(scalar1x,scalar1x);
275 for (i=s-1; i>=0; i--) {
276 if (i != (int)s-1) point_double_internal(out,out,0);
278 for (j=0; j<n; j++) {
281 for (k=0; k<t; k++) {
282 unsigned int bit = i + s*(k + j*t);
283 if (bit < DECAF_448_SCALAR_BITS) {
284 tab |= (scalar1x->limb[bit/WBITS] >> (bit%WBITS) & 1) << k;
288 mask_t invert = (tab>>(t-1))-1;
290 tab &= (1<<(t-1)) - 1;
292 constant_time_lookup_niels(ni, &table->table[j<<(t-1)], 1<<(t-1), tab);
294 cond_neg_niels(ni, invert);
295 if ((i!=(int)s-1)||j) {
296 add_niels_to_pt(out, ni, j==n-1 && i);
298 niels_to_pt(out, ni);
303 OPENSSL_cleanse(ni,sizeof(ni));
304 OPENSSL_cleanse(scalar1x,sizeof(scalar1x));
307 void curve448_point_mul_by_ratio_and_encode_like_eddsa (
308 uint8_t enc[DECAF_EDDSA_448_PUBLIC_BYTES],
309 const curve448_point_t p
312 /* The point is now on the twisted curve. Move it to untwisted. */
315 curve448_point_copy(q,p);
318 /* 4-isogeny: 2xy/(y^+x^2), (y^2-x^2)/(2z^2-y^2+x^2) */
323 gf_add( z, q->y, q->x );
333 OPENSSL_cleanse(u,sizeof(u));
342 enc[DECAF_EDDSA_448_PRIVATE_BYTES-1] = 0;
343 gf_serialize(enc, x, 1);
344 enc[DECAF_EDDSA_448_PRIVATE_BYTES-1] |= 0x80 & gf_lobit(t);
346 OPENSSL_cleanse(x,sizeof(x));
347 OPENSSL_cleanse(y,sizeof(y));
348 OPENSSL_cleanse(z,sizeof(z));
349 OPENSSL_cleanse(t,sizeof(t));
350 curve448_point_destroy(q);
354 decaf_error_t curve448_point_decode_like_eddsa_and_mul_by_ratio (
356 const uint8_t enc[DECAF_EDDSA_448_PUBLIC_BYTES]
358 uint8_t enc2[DECAF_EDDSA_448_PUBLIC_BYTES];
359 memcpy(enc2,enc,sizeof(enc2));
361 mask_t low = ~word_is_zero(enc2[DECAF_EDDSA_448_PRIVATE_BYTES-1] & 0x80);
362 enc2[DECAF_EDDSA_448_PRIVATE_BYTES-1] &= ~0x80;
364 mask_t succ = gf_deserialize(p->y, enc2, 1, 0);
366 succ &= word_is_zero(enc2[DECAF_EDDSA_448_PRIVATE_BYTES-1]);
370 gf_sub(p->z,ONE,p->x); /* num = 1-y^2 */
371 gf_mulw(p->t,p->x,EDWARDS_D); /* dy^2 */
372 gf_sub(p->t,ONE,p->t); /* denom = 1-dy^2 or 1-d + dy^2 */
374 gf_mul(p->x,p->z,p->t);
375 succ &= gf_isr(p->t,p->x); /* 1/sqrt(num * denom) */
377 gf_mul(p->x,p->t,p->z); /* sqrt(num / denom) */
378 gf_cond_neg(p->x,gf_lobit(p->x)^low);
382 /* 4-isogeny 2xy/(y^2-ax^2), (y^2+ax^2)/(2-y^2-ax^2) */
387 gf_add ( p->t, p->y, p->x );
390 gf_sub ( p->t, a, c );
391 gf_sqr ( p->x, p->z );
392 gf_add ( p->z, p->x, p->x );
393 gf_sub ( a, p->z, d );
394 gf_mul ( p->x, a, b );
395 gf_mul ( p->z, p->t, a );
396 gf_mul ( p->y, p->t, d );
397 gf_mul ( p->t, b, d );
398 OPENSSL_cleanse(a,sizeof(a));
399 OPENSSL_cleanse(b,sizeof(b));
400 OPENSSL_cleanse(c,sizeof(c));
401 OPENSSL_cleanse(d,sizeof(d));
404 OPENSSL_cleanse(enc2,sizeof(enc2));
405 assert(curve448_point_valid(p) || ~succ);
407 return decaf_succeed_if(mask_to_bool(succ));
410 decaf_error_t decaf_x448 (
411 uint8_t out[X_PUBLIC_BYTES],
412 const uint8_t base[X_PUBLIC_BYTES],
413 const uint8_t scalar[X_PRIVATE_BYTES]
415 gf x1, x2, z2, x3, z3, t1, t2;
416 ignore_result(gf_deserialize(x1,base,1,0));
425 for (t = X_PRIVATE_BITS-1; t>=0; t--) {
426 uint8_t sb = scalar[t/8];
428 /* Scalar conditioning */
429 if (t/8==0) sb &= -(uint8_t)COFACTOR;
430 else if (t == X_PRIVATE_BITS-1) sb = -1;
432 mask_t k_t = (sb>>(t%8)) & 1;
433 k_t = -k_t; /* set to all 0s or all 1s */
436 gf_cond_swap(x2,x3,swap);
437 gf_cond_swap(z2,z3,swap);
440 gf_add_nr(t1,x2,z2); /* A = x2 + z2 */ /* 2+e */
441 gf_sub_nr(t2,x2,z2); /* B = x2 - z2 */ /* 3+e */
442 gf_sub_nr(z2,x3,z3); /* D = x3 - z3 */ /* 3+e */
443 gf_mul(x2,t1,z2); /* DA */
444 gf_add_nr(z2,z3,x3); /* C = x3 + z3 */ /* 2+e */
445 gf_mul(x3,t2,z2); /* CB */
446 gf_sub_nr(z3,x2,x3); /* DA-CB */ /* 3+e */
447 gf_sqr(z2,z3); /* (DA-CB)^2 */
448 gf_mul(z3,x1,z2); /* z3 = x1(DA-CB)^2 */
449 gf_add_nr(z2,x2,x3); /* (DA+CB) */ /* 2+e */
450 gf_sqr(x3,z2); /* x3 = (DA+CB)^2 */
452 gf_sqr(z2,t1); /* AA = A^2 */
453 gf_sqr(t1,t2); /* BB = B^2 */
454 gf_mul(x2,z2,t1); /* x2 = AA*BB */
455 gf_sub_nr(t2,z2,t1); /* E = AA-BB */ /* 3+e */
457 gf_mulw(t1,t2,-EDWARDS_D); /* E*-d = a24*E */
458 gf_add_nr(t1,t1,z2); /* AA + a24*E */ /* 2+e */
459 gf_mul(z2,t2,t1); /* z2 = E(AA+a24*E) */
463 gf_cond_swap(x2,x3,swap);
464 gf_cond_swap(z2,z3,swap);
467 gf_serialize(out,x1,1);
468 mask_t nz = ~gf_eq(x1,ZERO);
470 OPENSSL_cleanse(x1,sizeof(x1));
471 OPENSSL_cleanse(x2,sizeof(x2));
472 OPENSSL_cleanse(z2,sizeof(z2));
473 OPENSSL_cleanse(x3,sizeof(x3));
474 OPENSSL_cleanse(z3,sizeof(z3));
475 OPENSSL_cleanse(t1,sizeof(t1));
476 OPENSSL_cleanse(t2,sizeof(t2));
478 return decaf_succeed_if(mask_to_bool(nz));
481 /* Thanks Johan Pascal */
482 void decaf_ed448_convert_public_key_to_x448 (
483 uint8_t x[DECAF_X448_PUBLIC_BYTES],
484 const uint8_t ed[DECAF_EDDSA_448_PUBLIC_BYTES]
487 const uint8_t mask = (uint8_t)(0xFE<<(7));
488 ignore_result(gf_deserialize(y, ed, 1, mask));
493 /* u = y^2 * (1-dy^2) / (1-y^2) */
494 gf_sqr(n,y); /* y^2*/
495 gf_sub(d,ONE,n); /* 1-y^2*/
496 gf_invert(d,d,0); /* 1/(1-y^2)*/
497 gf_mul(y,n,d); /* y^2 / (1-y^2) */
498 gf_mulw(d,n,EDWARDS_D); /* dy^2*/
499 gf_sub(d, ONE, d); /* 1-dy^2*/
500 gf_mul(n, y, d); /* y^2 * (1-dy^2) / (1-y^2) */
503 OPENSSL_cleanse(y,sizeof(y));
504 OPENSSL_cleanse(n,sizeof(n));
505 OPENSSL_cleanse(d,sizeof(d));
509 void curve448_point_mul_by_ratio_and_encode_like_x448 (
510 uint8_t out[X_PUBLIC_BYTES],
511 const curve448_point_t p
514 curve448_point_copy(q,p);
515 gf_invert(q->t,q->x,0); /* 1/x */
516 gf_mul(q->z,q->t,q->y); /* y/x */
517 gf_sqr(q->y,q->z); /* (y/x)^2 */
518 gf_serialize(out,q->y,1);
519 curve448_point_destroy(q);
522 void decaf_x448_derive_public_key (
523 uint8_t out[X_PUBLIC_BYTES],
524 const uint8_t scalar[X_PRIVATE_BYTES]
526 /* Scalar conditioning */
527 uint8_t scalar2[X_PRIVATE_BYTES];
528 memcpy(scalar2,scalar,sizeof(scalar2));
529 scalar2[0] &= -(uint8_t)COFACTOR;
531 scalar2[X_PRIVATE_BYTES-1] &= ~(-1u<<((X_PRIVATE_BITS+7)%8));
532 scalar2[X_PRIVATE_BYTES-1] |= 1<<((X_PRIVATE_BITS+7)%8);
534 curve448_scalar_t the_scalar;
535 curve448_scalar_decode_long(the_scalar,scalar2,sizeof(scalar2));
537 /* Compensate for the encoding ratio */
538 for (unsigned i=1; i<DECAF_X448_ENCODE_RATIO; i<<=1) {
539 curve448_scalar_halve(the_scalar,the_scalar);
542 curve448_precomputed_scalarmul(p,curve448_precomputed_base,the_scalar);
543 curve448_point_mul_by_ratio_and_encode_like_x448(out,p);
544 curve448_point_destroy(p);
549 * Control for variable-time scalar multiply algorithms.
551 struct smvt_control {
555 static int recode_wnaf (
556 struct smvt_control *control, /* [nbits/(table_bits+1) + 3] */
557 const curve448_scalar_t scalar,
558 unsigned int table_bits
560 unsigned int table_size = DECAF_448_SCALAR_BITS/(table_bits+1) + 3;
561 int position = table_size - 1; /* at the end */
563 /* place the end marker */
564 control[position].power = -1;
565 control[position].addend = 0;
568 /* PERF: Could negate scalar if it's large. But then would need more cases
569 * in the actual code that uses it, all for an expected reduction of like 1/5 op.
570 * Probably not worth it.
573 uint64_t current = scalar->limb[0] & 0xFFFF;
574 uint32_t mask = (1<<(table_bits+1))-1;
577 const unsigned int B_OVER_16 = sizeof(scalar->limb[0]) / 2;
578 for (w = 1; w<(DECAF_448_SCALAR_BITS-1)/16+3; w++) {
579 if (w < (DECAF_448_SCALAR_BITS-1)/16+1) {
580 /* Refill the 16 high bits of current */
581 current += (uint32_t)((scalar->limb[w/B_OVER_16]>>(16*(w%B_OVER_16)))<<16);
584 while (current & 0xFFFF) {
585 assert(position >= 0);
586 uint32_t pos = __builtin_ctz((uint32_t)current), odd = (uint32_t)current >> pos;
587 int32_t delta = odd & mask;
588 if (odd & 1<<(table_bits+1)) delta -= (1<<(table_bits+1));
589 current -= delta << pos;
590 control[position].power = pos + 16*(w-1);
591 control[position].addend = delta;
599 unsigned int n = table_size - position;
601 for (i=0; i<n; i++) {
602 control[i] = control[i+position];
610 const curve448_point_t working,
613 curve448_point_t tmp;
615 pt_to_pniels(output[0], working);
617 if (tbits == 0) return;
619 curve448_point_double(tmp,working);
621 pt_to_pniels(twop, tmp);
623 add_pniels_to_pt(tmp, output[0],0);
624 pt_to_pniels(output[1], tmp);
626 for (i=2; i < 1<<tbits; i++) {
627 add_pniels_to_pt(tmp, twop,0);
628 pt_to_pniels(output[i], tmp);
631 curve448_point_destroy(tmp);
632 OPENSSL_cleanse(twop,sizeof(twop));
635 extern const gf curve448_precomputed_wnaf_as_fe[];
636 static const niels_t *curve448_wnaf_base = (const niels_t *)curve448_precomputed_wnaf_as_fe;
638 void curve448_base_double_scalarmul_non_secret (
639 curve448_point_t combo,
640 const curve448_scalar_t scalar1,
641 const curve448_point_t base2,
642 const curve448_scalar_t scalar2
644 const int table_bits_var = DECAF_WNAF_VAR_TABLE_BITS,
645 table_bits_pre = DECAF_WNAF_FIXED_TABLE_BITS;
646 struct smvt_control control_var[DECAF_448_SCALAR_BITS/(table_bits_var+1)+3];
647 struct smvt_control control_pre[DECAF_448_SCALAR_BITS/(table_bits_pre+1)+3];
649 int ncb_pre = recode_wnaf(control_pre, scalar1, table_bits_pre);
650 int ncb_var = recode_wnaf(control_var, scalar2, table_bits_var);
652 pniels_t precmp_var[1<<table_bits_var];
653 prepare_wnaf_table(precmp_var, base2, table_bits_var);
655 int contp=0, contv=0, i = control_var[0].power;
658 curve448_point_copy(combo, curve448_point_identity);
660 } else if (i > control_pre[0].power) {
661 pniels_to_pt(combo, precmp_var[control_var[0].addend >> 1]);
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, curve448_wnaf_base[control_pre[0].addend >> 1], i);
668 i = control_pre[0].power;
669 niels_to_pt(combo, curve448_wnaf_base[control_pre[0].addend >> 1]);
673 for (i--; i >= 0; i--) {
674 int cv = (i==control_var[contv].power), cp = (i==control_pre[contp].power);
675 point_double_internal(combo,combo,i && !(cv||cp));
678 assert(control_var[contv].addend);
680 if (control_var[contv].addend > 0) {
681 add_pniels_to_pt(combo, precmp_var[control_var[contv].addend >> 1], i&&!cp);
683 sub_pniels_from_pt(combo, precmp_var[(-control_var[contv].addend) >> 1], i&&!cp);
689 assert(control_pre[contp].addend);
691 if (control_pre[contp].addend > 0) {
692 add_niels_to_pt(combo, curve448_wnaf_base[control_pre[contp].addend >> 1], i);
694 sub_niels_from_pt(combo, curve448_wnaf_base[(-control_pre[contp].addend) >> 1], i);
700 /* This function is non-secret, but whatever this is cheap. */
701 OPENSSL_cleanse(control_var,sizeof(control_var));
702 OPENSSL_cleanse(control_pre,sizeof(control_pre));
703 OPENSSL_cleanse(precmp_var,sizeof(precmp_var));
705 assert(contv == ncb_var); (void)ncb_var;
706 assert(contp == ncb_pre); (void)ncb_pre;
709 void curve448_point_destroy (
710 curve448_point_t point
712 OPENSSL_cleanse(point, sizeof(curve448_point_t));
715 int X448(uint8_t out_shared_key[56], const uint8_t private_key[56],
716 const uint8_t peer_public_value[56])
718 return decaf_x448(out_shared_key, peer_public_value, private_key)
722 void X448_public_from_private(uint8_t out_public_value[56],
723 const uint8_t private_key[56])
725 decaf_x448_derive_public_key(out_public_value, private_key);