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 };
40 #define TWISTED_D ((EDWARDS_D)-1)
42 #define EFF_D (-(TWISTED_D))
45 /* End of template stuff */
47 #define WBITS DECAF_WORD_BITS /* NB this may be different from ARCH_WORD_BITS */
49 /* Projective Niels coordinates */
50 typedef struct { gf a, b, c; } niels_s, niels_t[1];
51 typedef struct { niels_t n; gf z; } VECTOR_ALIGNED pniels_s, pniels_t[1];
53 /* Precomputed base */
54 struct curve448_precomputed_s { niels_t table [COMBS_N<<(COMBS_T-1)]; };
56 extern const gf curve448_precomputed_base_as_fe[];
57 const curve448_precomputed_s *curve448_precomputed_base =
58 (const curve448_precomputed_s *) &curve448_precomputed_base_as_fe;
62 gf_invert(gf y, const gf x, int assert_nonzero) {
66 gf_sqr(t1, x); /* o^2 */
67 ret = gf_isr(t2, t1); /* +-1/sqrt(o^2) = +-1/o */
69 if (assert_nonzero) assert(ret);
71 gf_mul(t2, t1, x); /* not direct to y in case of alias. */
75 /** identity = (0,1) */
76 const curve448_point_t curve448_point_identity = {{{{{0}}},{{{1}}},{{{1}}},{{{0}}}}};
79 point_double_internal (
81 const curve448_point_t q,
87 gf_add_nr ( d, c, a ); /* 2+e */
88 gf_add_nr ( p->t, q->y, q->x ); /* 2+e */
90 gf_subx_nr ( b, b, d, 3 ); /* 4+e */
91 gf_sub_nr ( p->t, a, c ); /* 3+e */
92 gf_sqr ( p->x, q->z );
93 gf_add_nr ( p->z, p->x, p->x ); /* 2+e */
94 gf_subx_nr ( a, p->z, p->t, 4 ); /* 6+e */
95 if (GF_HEADROOM == 5) gf_weak_reduce(a); /* or 1+e */
96 gf_mul ( p->x, a, b );
97 gf_mul ( p->z, p->t, a );
98 gf_mul ( p->y, p->t, d );
99 if (!before_double) gf_mul ( p->t, b, d );
102 void curve448_point_double(curve448_point_t p, const curve448_point_t q) {
103 point_double_internal(p,q,0);
106 /* Operations on [p]niels */
107 static ossl_inline void
112 gf_cond_swap(n->a, n->b, neg);
113 gf_cond_neg(n->c, neg);
116 static void pt_to_pniels (
118 const curve448_point_t a
120 gf_sub ( b->n->a, a->y, a->x );
121 gf_add ( b->n->b, a->x, a->y );
122 gf_mulw ( b->n->c, a->t, 2*TWISTED_D );
123 gf_add ( b->z, a->z, a->z );
126 static void pniels_to_pt (
131 gf_add ( eu, d->n->b, d->n->a );
132 gf_sub ( e->y, d->n->b, d->n->a );
133 gf_mul ( e->t, e->y, eu);
134 gf_mul ( e->x, d->z, e->y );
135 gf_mul ( e->y, d->z, eu );
136 gf_sqr ( e->z, d->z );
144 gf_add ( e->y, n->b, n->a );
145 gf_sub ( e->x, n->b, n->a );
146 gf_mul ( e->t, e->y, e->x );
147 gf_copy ( e->z, ONE );
157 gf_sub_nr ( b, d->y, d->x ); /* 3+e */
158 gf_mul ( a, e->a, b );
159 gf_add_nr ( b, d->x, d->y ); /* 2+e */
160 gf_mul ( d->y, e->b, b );
161 gf_mul ( d->x, e->c, d->t );
162 gf_add_nr ( c, a, d->y ); /* 2+e */
163 gf_sub_nr ( b, d->y, a ); /* 3+e */
164 gf_sub_nr ( d->y, d->z, d->x ); /* 3+e */
165 gf_add_nr ( a, d->x, d->z ); /* 2+e */
166 gf_mul ( d->z, a, d->y );
167 gf_mul ( d->x, d->y, b );
168 gf_mul ( d->y, a, c );
169 if (!before_double) gf_mul ( d->t, b, c );
179 gf_sub_nr ( b, d->y, d->x ); /* 3+e */
180 gf_mul ( a, e->b, b );
181 gf_add_nr ( b, d->x, d->y ); /* 2+e */
182 gf_mul ( d->y, e->a, b );
183 gf_mul ( d->x, e->c, d->t );
184 gf_add_nr ( c, a, d->y ); /* 2+e */
185 gf_sub_nr ( b, d->y, a ); /* 3+e */
186 gf_add_nr ( d->y, d->z, d->x ); /* 2+e */
187 gf_sub_nr ( a, d->z, d->x ); /* 3+e */
188 gf_mul ( d->z, a, d->y );
189 gf_mul ( d->x, d->y, b );
190 gf_mul ( d->y, a, c );
191 if (!before_double) gf_mul ( d->t, b, c );
201 gf_mul ( L0, p->z, pn->z );
202 gf_copy ( p->z, L0 );
203 add_niels_to_pt( p, pn->n, before_double );
213 gf_mul ( L0, p->z, pn->z );
214 gf_copy ( p->z, L0 );
215 sub_niels_from_pt( p, pn->n, before_double );
218 decaf_bool_t curve448_point_eq ( const curve448_point_t p, const curve448_point_t q ) {
221 /* equality mod 2-torsion compares x/y */
223 gf_mul ( a, p->y, q->x );
224 gf_mul ( b, q->y, p->x );
227 return mask_to_bool(succ);
230 decaf_bool_t curve448_point_valid (
231 const curve448_point_t p
243 gf_mulw(c,b,TWISTED_D);
247 out &= ~gf_eq(p->z,ZERO);
248 return mask_to_bool(out);
251 static ossl_inline void
252 constant_time_lookup_niels (
253 niels_s *__restrict__ ni,
254 const niels_t *table,
258 constant_time_lookup(ni, table, sizeof(niels_s), nelts, idx);
261 void curve448_precomputed_scalarmul (
262 curve448_point_t out,
263 const curve448_precomputed_s *table,
264 const curve448_scalar_t scalar
268 const unsigned int n = COMBS_N, t = COMBS_T, s = COMBS_S;
271 curve448_scalar_t scalar1x;
272 curve448_scalar_add(scalar1x, scalar, precomputed_scalarmul_adjustment);
273 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++) {
282 for (k=0; k<t; k++) {
283 unsigned int bit = i + s*(k + j*t);
284 if (bit < DECAF_448_SCALAR_BITS) {
285 tab |= (scalar1x->limb[bit/WBITS] >> (bit%WBITS) & 1) << k;
289 invert = (tab>>(t-1))-1;
291 tab &= (1<<(t-1)) - 1;
293 constant_time_lookup_niels(ni, &table->table[j<<(t-1)], 1<<(t-1), tab);
295 cond_neg_niels(ni, invert);
296 if ((i!=(int)s-1)||j) {
297 add_niels_to_pt(out, ni, j==n-1 && i);
299 niels_to_pt(out, ni);
304 OPENSSL_cleanse(ni,sizeof(ni));
305 OPENSSL_cleanse(scalar1x,sizeof(scalar1x));
308 void curve448_point_mul_by_ratio_and_encode_like_eddsa (
309 uint8_t enc[DECAF_EDDSA_448_PUBLIC_BYTES],
310 const curve448_point_t p
313 /* The point is now on the twisted curve. Move it to untwisted. */
316 curve448_point_copy(q,p);
319 /* 4-isogeny: 2xy/(y^+x^2), (y^2-x^2)/(2z^2-y^2+x^2) */
324 gf_add( z, q->y, q->x );
334 OPENSSL_cleanse(u,sizeof(u));
343 enc[DECAF_EDDSA_448_PRIVATE_BYTES-1] = 0;
344 gf_serialize(enc, x, 1);
345 enc[DECAF_EDDSA_448_PRIVATE_BYTES-1] |= 0x80 & gf_lobit(t);
347 OPENSSL_cleanse(x,sizeof(x));
348 OPENSSL_cleanse(y,sizeof(y));
349 OPENSSL_cleanse(z,sizeof(z));
350 OPENSSL_cleanse(t,sizeof(t));
351 curve448_point_destroy(q);
355 decaf_error_t curve448_point_decode_like_eddsa_and_mul_by_ratio (
357 const uint8_t enc[DECAF_EDDSA_448_PUBLIC_BYTES]
359 uint8_t enc2[DECAF_EDDSA_448_PUBLIC_BYTES];
363 memcpy(enc2,enc,sizeof(enc2));
365 low = ~word_is_zero(enc2[DECAF_EDDSA_448_PRIVATE_BYTES-1] & 0x80);
366 enc2[DECAF_EDDSA_448_PRIVATE_BYTES-1] &= ~0x80;
368 succ = gf_deserialize(p->y, enc2, 1, 0);
370 succ &= word_is_zero(enc2[DECAF_EDDSA_448_PRIVATE_BYTES-1]);
374 gf_sub(p->z,ONE,p->x); /* num = 1-y^2 */
375 gf_mulw(p->t,p->x,EDWARDS_D); /* dy^2 */
376 gf_sub(p->t,ONE,p->t); /* denom = 1-dy^2 or 1-d + dy^2 */
378 gf_mul(p->x,p->z,p->t);
379 succ &= gf_isr(p->t,p->x); /* 1/sqrt(num * denom) */
381 gf_mul(p->x,p->t,p->z); /* sqrt(num / denom) */
382 gf_cond_neg(p->x,gf_lobit(p->x)^low);
386 /* 4-isogeny 2xy/(y^2-ax^2), (y^2+ax^2)/(2-y^2-ax^2) */
391 gf_add ( p->t, p->y, p->x );
394 gf_sub ( p->t, a, c );
395 gf_sqr ( p->x, p->z );
396 gf_add ( p->z, p->x, p->x );
397 gf_sub ( a, p->z, d );
398 gf_mul ( p->x, a, b );
399 gf_mul ( p->z, p->t, a );
400 gf_mul ( p->y, p->t, d );
401 gf_mul ( p->t, b, d );
402 OPENSSL_cleanse(a,sizeof(a));
403 OPENSSL_cleanse(b,sizeof(b));
404 OPENSSL_cleanse(c,sizeof(c));
405 OPENSSL_cleanse(d,sizeof(d));
408 OPENSSL_cleanse(enc2,sizeof(enc2));
409 assert(curve448_point_valid(p) || ~succ);
411 return decaf_succeed_if(mask_to_bool(succ));
414 decaf_error_t decaf_x448 (
415 uint8_t out[X_PUBLIC_BYTES],
416 const uint8_t base[X_PUBLIC_BYTES],
417 const uint8_t scalar[X_PRIVATE_BYTES]
419 gf x1, x2, z2, x3, z3, t1, t2;
424 ignore_result(gf_deserialize(x1,base,1,0));
430 for (t = X_PRIVATE_BITS-1; t>=0; t--) {
431 uint8_t sb = scalar[t/8];
434 /* Scalar conditioning */
435 if (t/8==0) sb &= -(uint8_t)COFACTOR;
436 else if (t == X_PRIVATE_BITS-1) sb = -1;
438 k_t = (sb>>(t%8)) & 1;
439 k_t = -k_t; /* set to all 0s or all 1s */
442 gf_cond_swap(x2,x3,swap);
443 gf_cond_swap(z2,z3,swap);
446 gf_add_nr(t1,x2,z2); /* A = x2 + z2 */ /* 2+e */
447 gf_sub_nr(t2,x2,z2); /* B = x2 - z2 */ /* 3+e */
448 gf_sub_nr(z2,x3,z3); /* D = x3 - z3 */ /* 3+e */
449 gf_mul(x2,t1,z2); /* DA */
450 gf_add_nr(z2,z3,x3); /* C = x3 + z3 */ /* 2+e */
451 gf_mul(x3,t2,z2); /* CB */
452 gf_sub_nr(z3,x2,x3); /* DA-CB */ /* 3+e */
453 gf_sqr(z2,z3); /* (DA-CB)^2 */
454 gf_mul(z3,x1,z2); /* z3 = x1(DA-CB)^2 */
455 gf_add_nr(z2,x2,x3); /* (DA+CB) */ /* 2+e */
456 gf_sqr(x3,z2); /* x3 = (DA+CB)^2 */
458 gf_sqr(z2,t1); /* AA = A^2 */
459 gf_sqr(t1,t2); /* BB = B^2 */
460 gf_mul(x2,z2,t1); /* x2 = AA*BB */
461 gf_sub_nr(t2,z2,t1); /* E = AA-BB */ /* 3+e */
463 gf_mulw(t1,t2,-EDWARDS_D); /* E*-d = a24*E */
464 gf_add_nr(t1,t1,z2); /* AA + a24*E */ /* 2+e */
465 gf_mul(z2,t2,t1); /* z2 = E(AA+a24*E) */
469 gf_cond_swap(x2,x3,swap);
470 gf_cond_swap(z2,z3,swap);
473 gf_serialize(out,x1,1);
474 nz = ~gf_eq(x1,ZERO);
476 OPENSSL_cleanse(x1,sizeof(x1));
477 OPENSSL_cleanse(x2,sizeof(x2));
478 OPENSSL_cleanse(z2,sizeof(z2));
479 OPENSSL_cleanse(x3,sizeof(x3));
480 OPENSSL_cleanse(z3,sizeof(z3));
481 OPENSSL_cleanse(t1,sizeof(t1));
482 OPENSSL_cleanse(t2,sizeof(t2));
484 return decaf_succeed_if(mask_to_bool(nz));
487 /* Thanks Johan Pascal */
488 void decaf_ed448_convert_public_key_to_x448 (
489 uint8_t x[DECAF_X448_PUBLIC_BYTES],
490 const uint8_t ed[DECAF_EDDSA_448_PUBLIC_BYTES]
493 const uint8_t mask = (uint8_t)(0xFE<<(7));
494 ignore_result(gf_deserialize(y, ed, 1, mask));
499 /* u = y^2 * (1-dy^2) / (1-y^2) */
500 gf_sqr(n,y); /* y^2*/
501 gf_sub(d,ONE,n); /* 1-y^2*/
502 gf_invert(d,d,0); /* 1/(1-y^2)*/
503 gf_mul(y,n,d); /* y^2 / (1-y^2) */
504 gf_mulw(d,n,EDWARDS_D); /* dy^2*/
505 gf_sub(d, ONE, d); /* 1-dy^2*/
506 gf_mul(n, y, d); /* y^2 * (1-dy^2) / (1-y^2) */
509 OPENSSL_cleanse(y,sizeof(y));
510 OPENSSL_cleanse(n,sizeof(n));
511 OPENSSL_cleanse(d,sizeof(d));
515 void curve448_point_mul_by_ratio_and_encode_like_x448 (
516 uint8_t out[X_PUBLIC_BYTES],
517 const curve448_point_t p
520 curve448_point_copy(q,p);
521 gf_invert(q->t,q->x,0); /* 1/x */
522 gf_mul(q->z,q->t,q->y); /* y/x */
523 gf_sqr(q->y,q->z); /* (y/x)^2 */
524 gf_serialize(out,q->y,1);
525 curve448_point_destroy(q);
528 void decaf_x448_derive_public_key (
529 uint8_t out[X_PUBLIC_BYTES],
530 const uint8_t scalar[X_PRIVATE_BYTES]
532 /* Scalar conditioning */
533 uint8_t scalar2[X_PRIVATE_BYTES];
534 curve448_scalar_t the_scalar;
538 memcpy(scalar2,scalar,sizeof(scalar2));
539 scalar2[0] &= -(uint8_t)COFACTOR;
541 scalar2[X_PRIVATE_BYTES-1] &= ~(-1u<<((X_PRIVATE_BITS+7)%8));
542 scalar2[X_PRIVATE_BYTES-1] |= 1<<((X_PRIVATE_BITS+7)%8);
544 curve448_scalar_decode_long(the_scalar,scalar2,sizeof(scalar2));
546 /* Compensate for the encoding ratio */
547 for (i=1; i<DECAF_X448_ENCODE_RATIO; i<<=1) {
548 curve448_scalar_halve(the_scalar,the_scalar);
550 curve448_precomputed_scalarmul(p,curve448_precomputed_base,the_scalar);
551 curve448_point_mul_by_ratio_and_encode_like_x448(out,p);
552 curve448_point_destroy(p);
557 * Control for variable-time scalar multiply algorithms.
559 struct smvt_control {
563 static int recode_wnaf (
564 struct smvt_control *control, /* [nbits/(table_bits+1) + 3] */
565 const curve448_scalar_t scalar,
566 unsigned int table_bits
568 unsigned int table_size = DECAF_448_SCALAR_BITS/(table_bits+1) + 3;
569 int position = table_size - 1; /* at the end */
570 uint64_t current = scalar->limb[0] & 0xFFFF;
571 uint32_t mask = (1<<(table_bits+1))-1;
573 const unsigned int B_OVER_16 = sizeof(scalar->limb[0]) / 2;
576 /* place the end marker */
577 control[position].power = -1;
578 control[position].addend = 0;
581 /* PERF: Could negate scalar if it's large. But then would need more cases
582 * in the actual code that uses it, all for an expected reduction of like 1/5 op.
583 * Probably not worth it.
586 for (w = 1; w<(DECAF_448_SCALAR_BITS-1)/16+3; w++) {
587 if (w < (DECAF_448_SCALAR_BITS-1)/16+1) {
588 /* Refill the 16 high bits of current */
589 current += (uint32_t)((scalar->limb[w/B_OVER_16]>>(16*(w%B_OVER_16)))<<16);
592 while (current & 0xFFFF) {
593 uint32_t pos = __builtin_ctz((uint32_t)current), odd = (uint32_t)current >> pos;
594 int32_t delta = odd & mask;
596 assert(position >= 0);
597 if (odd & 1<<(table_bits+1)) delta -= (1<<(table_bits+1));
598 current -= delta << pos;
599 control[position].power = pos + 16*(w-1);
600 control[position].addend = delta;
608 n = table_size - position;
609 for (i=0; i<n; i++) {
610 control[i] = control[i+position];
618 const curve448_point_t working,
621 curve448_point_t tmp;
625 pt_to_pniels(output[0], working);
627 if (tbits == 0) return;
629 curve448_point_double(tmp,working);
630 pt_to_pniels(twop, tmp);
632 add_pniels_to_pt(tmp, output[0],0);
633 pt_to_pniels(output[1], tmp);
635 for (i=2; i < 1<<tbits; i++) {
636 add_pniels_to_pt(tmp, twop,0);
637 pt_to_pniels(output[i], tmp);
640 curve448_point_destroy(tmp);
641 OPENSSL_cleanse(twop,sizeof(twop));
644 extern const gf curve448_precomputed_wnaf_as_fe[];
645 static const niels_t *curve448_wnaf_base = (const niels_t *)curve448_precomputed_wnaf_as_fe;
647 void curve448_base_double_scalarmul_non_secret (
648 curve448_point_t combo,
649 const curve448_scalar_t scalar1,
650 const curve448_point_t base2,
651 const curve448_scalar_t scalar2
653 const int table_bits_var = DECAF_WNAF_VAR_TABLE_BITS,
654 table_bits_pre = DECAF_WNAF_FIXED_TABLE_BITS;
655 struct smvt_control control_var[DECAF_448_SCALAR_BITS/(DECAF_WNAF_VAR_TABLE_BITS+1)+3];
656 struct smvt_control control_pre[DECAF_448_SCALAR_BITS/(DECAF_WNAF_FIXED_TABLE_BITS+1)+3];
657 int ncb_pre = recode_wnaf(control_pre, scalar1, table_bits_pre);
658 int ncb_var = recode_wnaf(control_var, scalar2, table_bits_var);
659 pniels_t precmp_var[1<<DECAF_WNAF_VAR_TABLE_BITS];
660 int contp=0, contv=0, i;
662 prepare_wnaf_table(precmp_var, base2, table_bits_var);
663 i = control_var[0].power;
666 curve448_point_copy(combo, curve448_point_identity);
668 } else if (i > control_pre[0].power) {
669 pniels_to_pt(combo, precmp_var[control_var[0].addend >> 1]);
671 } else if (i == control_pre[0].power && i >=0 ) {
672 pniels_to_pt(combo, precmp_var[control_var[0].addend >> 1]);
673 add_niels_to_pt(combo, curve448_wnaf_base[control_pre[0].addend >> 1], i);
676 i = control_pre[0].power;
677 niels_to_pt(combo, curve448_wnaf_base[control_pre[0].addend >> 1]);
681 for (i--; i >= 0; i--) {
682 int cv = (i==control_var[contv].power), cp = (i==control_pre[contp].power);
683 point_double_internal(combo,combo,i && !(cv||cp));
686 assert(control_var[contv].addend);
688 if (control_var[contv].addend > 0) {
689 add_pniels_to_pt(combo, precmp_var[control_var[contv].addend >> 1], i&&!cp);
691 sub_pniels_from_pt(combo, precmp_var[(-control_var[contv].addend) >> 1], i&&!cp);
697 assert(control_pre[contp].addend);
699 if (control_pre[contp].addend > 0) {
700 add_niels_to_pt(combo, curve448_wnaf_base[control_pre[contp].addend >> 1], i);
702 sub_niels_from_pt(combo, curve448_wnaf_base[(-control_pre[contp].addend) >> 1], i);
708 /* This function is non-secret, but whatever this is cheap. */
709 OPENSSL_cleanse(control_var,sizeof(control_var));
710 OPENSSL_cleanse(control_pre,sizeof(control_pre));
711 OPENSSL_cleanse(precmp_var,sizeof(precmp_var));
713 assert(contv == ncb_var); (void)ncb_var;
714 assert(contp == ncb_pre); (void)ncb_pre;
717 void curve448_point_destroy (
718 curve448_point_t point
720 OPENSSL_cleanse(point, sizeof(curve448_point_t));
723 int X448(uint8_t out_shared_key[56], const uint8_t private_key[56],
724 const uint8_t peer_public_value[56])
726 return decaf_x448(out_shared_key, peer_public_value, private_key)
730 void X448_public_from_private(uint8_t out_public_value[56],
731 const uint8_t private_key[56])
733 decaf_x448_derive_public_key(out_public_value, private_key);