• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright 2014-2020 The OpenSSL Project Authors. All Rights Reserved.
3  * Copyright (c) 2014, Intel Corporation. All Rights Reserved.
4  * Copyright (c) 2015, CloudFlare, Inc.
5  *
6  * Licensed under the OpenSSL license (the "License").  You may not use
7  * this file except in compliance with the License.  You can obtain a copy
8  * in the file LICENSE in the source distribution or at
9  * https://www.openssl.org/source/license.html
10  *
11  * Originally written by Shay Gueron (1, 2), and Vlad Krasnov (1, 3)
12  * (1) Intel Corporation, Israel Development Center, Haifa, Israel
13  * (2) University of Haifa, Israel
14  * (3) CloudFlare, Inc.
15  *
16  * Reference:
17  * S.Gueron and V.Krasnov, "Fast Prime Field Elliptic Curve Cryptography with
18  *                          256 Bit Primes"
19  */
20 
21 #include <string.h>
22 
23 #include "internal/cryptlib.h"
24 #include "crypto/bn.h"
25 #include "ec_local.h"
26 #include "internal/refcount.h"
27 
28 #if BN_BITS2 != 64
29 # define TOBN(hi,lo)    lo,hi
30 #else
31 # define TOBN(hi,lo)    ((BN_ULONG)hi<<32|lo)
32 #endif
33 
34 #if defined(__GNUC__)
35 # define ALIGN32        __attribute((aligned(32)))
36 #elif defined(_MSC_VER)
37 # define ALIGN32        __declspec(align(32))
38 #else
39 # define ALIGN32
40 #endif
41 
42 #define ALIGNPTR(p,N)   ((unsigned char *)p+N-(size_t)p%N)
43 #define P256_LIMBS      (256/BN_BITS2)
44 
45 typedef unsigned short u16;
46 
47 typedef struct {
48     BN_ULONG X[P256_LIMBS];
49     BN_ULONG Y[P256_LIMBS];
50     BN_ULONG Z[P256_LIMBS];
51 } P256_POINT;
52 
53 typedef struct {
54     BN_ULONG X[P256_LIMBS];
55     BN_ULONG Y[P256_LIMBS];
56 } P256_POINT_AFFINE;
57 
58 typedef P256_POINT_AFFINE PRECOMP256_ROW[64];
59 
60 /* structure for precomputed multiples of the generator */
61 struct nistz256_pre_comp_st {
62     const EC_GROUP *group;      /* Parent EC_GROUP object */
63     size_t w;                   /* Window size */
64     /*
65      * Constant time access to the X and Y coordinates of the pre-computed,
66      * generator multiplies, in the Montgomery domain. Pre-calculated
67      * multiplies are stored in affine form.
68      */
69     PRECOMP256_ROW *precomp;
70     void *precomp_storage;
71     CRYPTO_REF_COUNT references;
72     CRYPTO_RWLOCK *lock;
73 };
74 
75 /* Functions implemented in assembly */
76 /*
77  * Most of below mentioned functions *preserve* the property of inputs
78  * being fully reduced, i.e. being in [0, modulus) range. Simply put if
79  * inputs are fully reduced, then output is too. Note that reverse is
80  * not true, in sense that given partially reduced inputs output can be
81  * either, not unlikely reduced. And "most" in first sentence refers to
82  * the fact that given the calculations flow one can tolerate that
83  * addition, 1st function below, produces partially reduced result *if*
84  * multiplications by 2 and 3, which customarily use addition, fully
85  * reduce it. This effectively gives two options: a) addition produces
86  * fully reduced result [as long as inputs are, just like remaining
87  * functions]; b) addition is allowed to produce partially reduced
88  * result, but multiplications by 2 and 3 perform additional reduction
89  * step. Choice between the two can be platform-specific, but it was a)
90  * in all cases so far...
91  */
92 /* Modular add: res = a+b mod P   */
93 void ecp_nistz256_add(BN_ULONG res[P256_LIMBS],
94                       const BN_ULONG a[P256_LIMBS],
95                       const BN_ULONG b[P256_LIMBS]);
96 /* Modular mul by 2: res = 2*a mod P */
97 void ecp_nistz256_mul_by_2(BN_ULONG res[P256_LIMBS],
98                            const BN_ULONG a[P256_LIMBS]);
99 /* Modular mul by 3: res = 3*a mod P */
100 void ecp_nistz256_mul_by_3(BN_ULONG res[P256_LIMBS],
101                            const BN_ULONG a[P256_LIMBS]);
102 
103 /* Modular div by 2: res = a/2 mod P */
104 void ecp_nistz256_div_by_2(BN_ULONG res[P256_LIMBS],
105                            const BN_ULONG a[P256_LIMBS]);
106 /* Modular sub: res = a-b mod P   */
107 void ecp_nistz256_sub(BN_ULONG res[P256_LIMBS],
108                       const BN_ULONG a[P256_LIMBS],
109                       const BN_ULONG b[P256_LIMBS]);
110 /* Modular neg: res = -a mod P    */
111 void ecp_nistz256_neg(BN_ULONG res[P256_LIMBS], const BN_ULONG a[P256_LIMBS]);
112 /* Montgomery mul: res = a*b*2^-256 mod P */
113 void ecp_nistz256_mul_mont(BN_ULONG res[P256_LIMBS],
114                            const BN_ULONG a[P256_LIMBS],
115                            const BN_ULONG b[P256_LIMBS]);
116 /* Montgomery sqr: res = a*a*2^-256 mod P */
117 void ecp_nistz256_sqr_mont(BN_ULONG res[P256_LIMBS],
118                            const BN_ULONG a[P256_LIMBS]);
119 /* Convert a number from Montgomery domain, by multiplying with 1 */
120 void ecp_nistz256_from_mont(BN_ULONG res[P256_LIMBS],
121                             const BN_ULONG in[P256_LIMBS]);
122 /* Convert a number to Montgomery domain, by multiplying with 2^512 mod P*/
123 void ecp_nistz256_to_mont(BN_ULONG res[P256_LIMBS],
124                           const BN_ULONG in[P256_LIMBS]);
125 /* Functions that perform constant time access to the precomputed tables */
126 void ecp_nistz256_scatter_w5(P256_POINT *val,
127                              const P256_POINT *in_t, int idx);
128 void ecp_nistz256_gather_w5(P256_POINT *val,
129                             const P256_POINT *in_t, int idx);
130 void ecp_nistz256_scatter_w7(P256_POINT_AFFINE *val,
131                              const P256_POINT_AFFINE *in_t, int idx);
132 void ecp_nistz256_gather_w7(P256_POINT_AFFINE *val,
133                             const P256_POINT_AFFINE *in_t, int idx);
134 
135 /* One converted into the Montgomery domain */
136 static const BN_ULONG ONE[P256_LIMBS] = {
137     TOBN(0x00000000, 0x00000001), TOBN(0xffffffff, 0x00000000),
138     TOBN(0xffffffff, 0xffffffff), TOBN(0x00000000, 0xfffffffe)
139 };
140 
141 static NISTZ256_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group);
142 
143 /* Precomputed tables for the default generator */
144 extern const PRECOMP256_ROW ecp_nistz256_precomputed[37];
145 
146 /* Recode window to a signed digit, see ecp_nistputil.c for details */
_booth_recode_w5(unsigned int in)147 static unsigned int _booth_recode_w5(unsigned int in)
148 {
149     unsigned int s, d;
150 
151     s = ~((in >> 5) - 1);
152     d = (1 << 6) - in - 1;
153     d = (d & s) | (in & ~s);
154     d = (d >> 1) + (d & 1);
155 
156     return (d << 1) + (s & 1);
157 }
158 
_booth_recode_w7(unsigned int in)159 static unsigned int _booth_recode_w7(unsigned int in)
160 {
161     unsigned int s, d;
162 
163     s = ~((in >> 7) - 1);
164     d = (1 << 8) - in - 1;
165     d = (d & s) | (in & ~s);
166     d = (d >> 1) + (d & 1);
167 
168     return (d << 1) + (s & 1);
169 }
170 
copy_conditional(BN_ULONG dst[P256_LIMBS],const BN_ULONG src[P256_LIMBS],BN_ULONG move)171 static void copy_conditional(BN_ULONG dst[P256_LIMBS],
172                              const BN_ULONG src[P256_LIMBS], BN_ULONG move)
173 {
174     BN_ULONG mask1 = 0-move;
175     BN_ULONG mask2 = ~mask1;
176 
177     dst[0] = (src[0] & mask1) ^ (dst[0] & mask2);
178     dst[1] = (src[1] & mask1) ^ (dst[1] & mask2);
179     dst[2] = (src[2] & mask1) ^ (dst[2] & mask2);
180     dst[3] = (src[3] & mask1) ^ (dst[3] & mask2);
181     if (P256_LIMBS == 8) {
182         dst[4] = (src[4] & mask1) ^ (dst[4] & mask2);
183         dst[5] = (src[5] & mask1) ^ (dst[5] & mask2);
184         dst[6] = (src[6] & mask1) ^ (dst[6] & mask2);
185         dst[7] = (src[7] & mask1) ^ (dst[7] & mask2);
186     }
187 }
188 
is_zero(BN_ULONG in)189 static BN_ULONG is_zero(BN_ULONG in)
190 {
191     in |= (0 - in);
192     in = ~in;
193     in >>= BN_BITS2 - 1;
194     return in;
195 }
196 
is_equal(const BN_ULONG a[P256_LIMBS],const BN_ULONG b[P256_LIMBS])197 static BN_ULONG is_equal(const BN_ULONG a[P256_LIMBS],
198                          const BN_ULONG b[P256_LIMBS])
199 {
200     BN_ULONG res;
201 
202     res = a[0] ^ b[0];
203     res |= a[1] ^ b[1];
204     res |= a[2] ^ b[2];
205     res |= a[3] ^ b[3];
206     if (P256_LIMBS == 8) {
207         res |= a[4] ^ b[4];
208         res |= a[5] ^ b[5];
209         res |= a[6] ^ b[6];
210         res |= a[7] ^ b[7];
211     }
212 
213     return is_zero(res);
214 }
215 
is_one(const BIGNUM * z)216 static BN_ULONG is_one(const BIGNUM *z)
217 {
218     BN_ULONG res = 0;
219     BN_ULONG *a = bn_get_words(z);
220 
221     if (bn_get_top(z) == (P256_LIMBS - P256_LIMBS / 8)) {
222         res = a[0] ^ ONE[0];
223         res |= a[1] ^ ONE[1];
224         res |= a[2] ^ ONE[2];
225         res |= a[3] ^ ONE[3];
226         if (P256_LIMBS == 8) {
227             res |= a[4] ^ ONE[4];
228             res |= a[5] ^ ONE[5];
229             res |= a[6] ^ ONE[6];
230             /*
231              * no check for a[7] (being zero) on 32-bit platforms,
232              * because value of "one" takes only 7 limbs.
233              */
234         }
235         res = is_zero(res);
236     }
237 
238     return res;
239 }
240 
241 /*
242  * For reference, this macro is used only when new ecp_nistz256 assembly
243  * module is being developed.  For example, configure with
244  * -DECP_NISTZ256_REFERENCE_IMPLEMENTATION and implement only functions
245  * performing simplest arithmetic operations on 256-bit vectors. Then
246  * work on implementation of higher-level functions performing point
247  * operations. Then remove ECP_NISTZ256_REFERENCE_IMPLEMENTATION
248  * and never define it again. (The correct macro denoting presence of
249  * ecp_nistz256 module is ECP_NISTZ256_ASM.)
250  */
251 #ifndef ECP_NISTZ256_REFERENCE_IMPLEMENTATION
252 void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a);
253 void ecp_nistz256_point_add(P256_POINT *r,
254                             const P256_POINT *a, const P256_POINT *b);
255 void ecp_nistz256_point_add_affine(P256_POINT *r,
256                                    const P256_POINT *a,
257                                    const P256_POINT_AFFINE *b);
258 #else
259 /* Point double: r = 2*a */
ecp_nistz256_point_double(P256_POINT * r,const P256_POINT * a)260 static void ecp_nistz256_point_double(P256_POINT *r, const P256_POINT *a)
261 {
262     BN_ULONG S[P256_LIMBS];
263     BN_ULONG M[P256_LIMBS];
264     BN_ULONG Zsqr[P256_LIMBS];
265     BN_ULONG tmp0[P256_LIMBS];
266 
267     const BN_ULONG *in_x = a->X;
268     const BN_ULONG *in_y = a->Y;
269     const BN_ULONG *in_z = a->Z;
270 
271     BN_ULONG *res_x = r->X;
272     BN_ULONG *res_y = r->Y;
273     BN_ULONG *res_z = r->Z;
274 
275     ecp_nistz256_mul_by_2(S, in_y);
276 
277     ecp_nistz256_sqr_mont(Zsqr, in_z);
278 
279     ecp_nistz256_sqr_mont(S, S);
280 
281     ecp_nistz256_mul_mont(res_z, in_z, in_y);
282     ecp_nistz256_mul_by_2(res_z, res_z);
283 
284     ecp_nistz256_add(M, in_x, Zsqr);
285     ecp_nistz256_sub(Zsqr, in_x, Zsqr);
286 
287     ecp_nistz256_sqr_mont(res_y, S);
288     ecp_nistz256_div_by_2(res_y, res_y);
289 
290     ecp_nistz256_mul_mont(M, M, Zsqr);
291     ecp_nistz256_mul_by_3(M, M);
292 
293     ecp_nistz256_mul_mont(S, S, in_x);
294     ecp_nistz256_mul_by_2(tmp0, S);
295 
296     ecp_nistz256_sqr_mont(res_x, M);
297 
298     ecp_nistz256_sub(res_x, res_x, tmp0);
299     ecp_nistz256_sub(S, S, res_x);
300 
301     ecp_nistz256_mul_mont(S, S, M);
302     ecp_nistz256_sub(res_y, S, res_y);
303 }
304 
305 /* Point addition: r = a+b */
ecp_nistz256_point_add(P256_POINT * r,const P256_POINT * a,const P256_POINT * b)306 static void ecp_nistz256_point_add(P256_POINT *r,
307                                    const P256_POINT *a, const P256_POINT *b)
308 {
309     BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS];
310     BN_ULONG U1[P256_LIMBS], S1[P256_LIMBS];
311     BN_ULONG Z1sqr[P256_LIMBS];
312     BN_ULONG Z2sqr[P256_LIMBS];
313     BN_ULONG H[P256_LIMBS], R[P256_LIMBS];
314     BN_ULONG Hsqr[P256_LIMBS];
315     BN_ULONG Rsqr[P256_LIMBS];
316     BN_ULONG Hcub[P256_LIMBS];
317 
318     BN_ULONG res_x[P256_LIMBS];
319     BN_ULONG res_y[P256_LIMBS];
320     BN_ULONG res_z[P256_LIMBS];
321 
322     BN_ULONG in1infty, in2infty;
323 
324     const BN_ULONG *in1_x = a->X;
325     const BN_ULONG *in1_y = a->Y;
326     const BN_ULONG *in1_z = a->Z;
327 
328     const BN_ULONG *in2_x = b->X;
329     const BN_ULONG *in2_y = b->Y;
330     const BN_ULONG *in2_z = b->Z;
331 
332     /*
333      * Infinity in encoded as (,,0)
334      */
335     in1infty = (in1_z[0] | in1_z[1] | in1_z[2] | in1_z[3]);
336     if (P256_LIMBS == 8)
337         in1infty |= (in1_z[4] | in1_z[5] | in1_z[6] | in1_z[7]);
338 
339     in2infty = (in2_z[0] | in2_z[1] | in2_z[2] | in2_z[3]);
340     if (P256_LIMBS == 8)
341         in2infty |= (in2_z[4] | in2_z[5] | in2_z[6] | in2_z[7]);
342 
343     in1infty = is_zero(in1infty);
344     in2infty = is_zero(in2infty);
345 
346     ecp_nistz256_sqr_mont(Z2sqr, in2_z);        /* Z2^2 */
347     ecp_nistz256_sqr_mont(Z1sqr, in1_z);        /* Z1^2 */
348 
349     ecp_nistz256_mul_mont(S1, Z2sqr, in2_z);    /* S1 = Z2^3 */
350     ecp_nistz256_mul_mont(S2, Z1sqr, in1_z);    /* S2 = Z1^3 */
351 
352     ecp_nistz256_mul_mont(S1, S1, in1_y);       /* S1 = Y1*Z2^3 */
353     ecp_nistz256_mul_mont(S2, S2, in2_y);       /* S2 = Y2*Z1^3 */
354     ecp_nistz256_sub(R, S2, S1);                /* R = S2 - S1 */
355 
356     ecp_nistz256_mul_mont(U1, in1_x, Z2sqr);    /* U1 = X1*Z2^2 */
357     ecp_nistz256_mul_mont(U2, in2_x, Z1sqr);    /* U2 = X2*Z1^2 */
358     ecp_nistz256_sub(H, U2, U1);                /* H = U2 - U1 */
359 
360     /*
361      * The formulae are incorrect if the points are equal so we check for
362      * this and do doubling if this happens.
363      *
364      * Points here are in Jacobian projective coordinates (Xi, Yi, Zi)
365      * that are bound to the affine coordinates (xi, yi) by the following
366      * equations:
367      *     - xi = Xi / (Zi)^2
368      *     - y1 = Yi / (Zi)^3
369      *
370      * For the sake of optimization, the algorithm operates over
371      * intermediate variables U1, U2 and S1, S2 that are derived from
372      * the projective coordinates:
373      *     - U1 = X1 * (Z2)^2 ; U2 = X2 * (Z1)^2
374      *     - S1 = Y1 * (Z2)^3 ; S2 = Y2 * (Z1)^3
375      *
376      * It is easy to prove that is_equal(U1, U2) implies that the affine
377      * x-coordinates are equal, or either point is at infinity.
378      * Likewise is_equal(S1, S2) implies that the affine y-coordinates are
379      * equal, or either point is at infinity.
380      *
381      * The special case of either point being the point at infinity (Z1 or Z2
382      * is zero), is handled separately later on in this function, so we avoid
383      * jumping to point_double here in those special cases.
384      *
385      * When both points are inverse of each other, we know that the affine
386      * x-coordinates are equal, and the y-coordinates have different sign.
387      * Therefore since U1 = U2, we know H = 0, and therefore Z3 = H*Z1*Z2
388      * will equal 0, thus the result is infinity, if we simply let this
389      * function continue normally.
390      *
391      * We use bitwise operations to avoid potential side-channels introduced by
392      * the short-circuiting behaviour of boolean operators.
393      */
394     if (is_equal(U1, U2) & ~in1infty & ~in2infty & is_equal(S1, S2)) {
395         /*
396          * This is obviously not constant-time but it should never happen during
397          * single point multiplication, so there is no timing leak for ECDH or
398          * ECDSA signing.
399          */
400         ecp_nistz256_point_double(r, a);
401         return;
402     }
403 
404     ecp_nistz256_sqr_mont(Rsqr, R);             /* R^2 */
405     ecp_nistz256_mul_mont(res_z, H, in1_z);     /* Z3 = H*Z1*Z2 */
406     ecp_nistz256_sqr_mont(Hsqr, H);             /* H^2 */
407     ecp_nistz256_mul_mont(res_z, res_z, in2_z); /* Z3 = H*Z1*Z2 */
408     ecp_nistz256_mul_mont(Hcub, Hsqr, H);       /* H^3 */
409 
410     ecp_nistz256_mul_mont(U2, U1, Hsqr);        /* U1*H^2 */
411     ecp_nistz256_mul_by_2(Hsqr, U2);            /* 2*U1*H^2 */
412 
413     ecp_nistz256_sub(res_x, Rsqr, Hsqr);
414     ecp_nistz256_sub(res_x, res_x, Hcub);
415 
416     ecp_nistz256_sub(res_y, U2, res_x);
417 
418     ecp_nistz256_mul_mont(S2, S1, Hcub);
419     ecp_nistz256_mul_mont(res_y, R, res_y);
420     ecp_nistz256_sub(res_y, res_y, S2);
421 
422     copy_conditional(res_x, in2_x, in1infty);
423     copy_conditional(res_y, in2_y, in1infty);
424     copy_conditional(res_z, in2_z, in1infty);
425 
426     copy_conditional(res_x, in1_x, in2infty);
427     copy_conditional(res_y, in1_y, in2infty);
428     copy_conditional(res_z, in1_z, in2infty);
429 
430     memcpy(r->X, res_x, sizeof(res_x));
431     memcpy(r->Y, res_y, sizeof(res_y));
432     memcpy(r->Z, res_z, sizeof(res_z));
433 }
434 
435 /* Point addition when b is known to be affine: r = a+b */
ecp_nistz256_point_add_affine(P256_POINT * r,const P256_POINT * a,const P256_POINT_AFFINE * b)436 static void ecp_nistz256_point_add_affine(P256_POINT *r,
437                                           const P256_POINT *a,
438                                           const P256_POINT_AFFINE *b)
439 {
440     BN_ULONG U2[P256_LIMBS], S2[P256_LIMBS];
441     BN_ULONG Z1sqr[P256_LIMBS];
442     BN_ULONG H[P256_LIMBS], R[P256_LIMBS];
443     BN_ULONG Hsqr[P256_LIMBS];
444     BN_ULONG Rsqr[P256_LIMBS];
445     BN_ULONG Hcub[P256_LIMBS];
446 
447     BN_ULONG res_x[P256_LIMBS];
448     BN_ULONG res_y[P256_LIMBS];
449     BN_ULONG res_z[P256_LIMBS];
450 
451     BN_ULONG in1infty, in2infty;
452 
453     const BN_ULONG *in1_x = a->X;
454     const BN_ULONG *in1_y = a->Y;
455     const BN_ULONG *in1_z = a->Z;
456 
457     const BN_ULONG *in2_x = b->X;
458     const BN_ULONG *in2_y = b->Y;
459 
460     /*
461      * Infinity in encoded as (,,0)
462      */
463     in1infty = (in1_z[0] | in1_z[1] | in1_z[2] | in1_z[3]);
464     if (P256_LIMBS == 8)
465         in1infty |= (in1_z[4] | in1_z[5] | in1_z[6] | in1_z[7]);
466 
467     /*
468      * In affine representation we encode infinity as (0,0), which is
469      * not on the curve, so it is OK
470      */
471     in2infty = (in2_x[0] | in2_x[1] | in2_x[2] | in2_x[3] |
472                 in2_y[0] | in2_y[1] | in2_y[2] | in2_y[3]);
473     if (P256_LIMBS == 8)
474         in2infty |= (in2_x[4] | in2_x[5] | in2_x[6] | in2_x[7] |
475                      in2_y[4] | in2_y[5] | in2_y[6] | in2_y[7]);
476 
477     in1infty = is_zero(in1infty);
478     in2infty = is_zero(in2infty);
479 
480     ecp_nistz256_sqr_mont(Z1sqr, in1_z);        /* Z1^2 */
481 
482     ecp_nistz256_mul_mont(U2, in2_x, Z1sqr);    /* U2 = X2*Z1^2 */
483     ecp_nistz256_sub(H, U2, in1_x);             /* H = U2 - U1 */
484 
485     ecp_nistz256_mul_mont(S2, Z1sqr, in1_z);    /* S2 = Z1^3 */
486 
487     ecp_nistz256_mul_mont(res_z, H, in1_z);     /* Z3 = H*Z1*Z2 */
488 
489     ecp_nistz256_mul_mont(S2, S2, in2_y);       /* S2 = Y2*Z1^3 */
490     ecp_nistz256_sub(R, S2, in1_y);             /* R = S2 - S1 */
491 
492     ecp_nistz256_sqr_mont(Hsqr, H);             /* H^2 */
493     ecp_nistz256_sqr_mont(Rsqr, R);             /* R^2 */
494     ecp_nistz256_mul_mont(Hcub, Hsqr, H);       /* H^3 */
495 
496     ecp_nistz256_mul_mont(U2, in1_x, Hsqr);     /* U1*H^2 */
497     ecp_nistz256_mul_by_2(Hsqr, U2);            /* 2*U1*H^2 */
498 
499     ecp_nistz256_sub(res_x, Rsqr, Hsqr);
500     ecp_nistz256_sub(res_x, res_x, Hcub);
501     ecp_nistz256_sub(H, U2, res_x);
502 
503     ecp_nistz256_mul_mont(S2, in1_y, Hcub);
504     ecp_nistz256_mul_mont(H, H, R);
505     ecp_nistz256_sub(res_y, H, S2);
506 
507     copy_conditional(res_x, in2_x, in1infty);
508     copy_conditional(res_x, in1_x, in2infty);
509 
510     copy_conditional(res_y, in2_y, in1infty);
511     copy_conditional(res_y, in1_y, in2infty);
512 
513     copy_conditional(res_z, ONE, in1infty);
514     copy_conditional(res_z, in1_z, in2infty);
515 
516     memcpy(r->X, res_x, sizeof(res_x));
517     memcpy(r->Y, res_y, sizeof(res_y));
518     memcpy(r->Z, res_z, sizeof(res_z));
519 }
520 #endif
521 
522 /* r = in^-1 mod p */
ecp_nistz256_mod_inverse(BN_ULONG r[P256_LIMBS],const BN_ULONG in[P256_LIMBS])523 static void ecp_nistz256_mod_inverse(BN_ULONG r[P256_LIMBS],
524                                      const BN_ULONG in[P256_LIMBS])
525 {
526     /*
527      * The poly is ffffffff 00000001 00000000 00000000 00000000 ffffffff
528      * ffffffff ffffffff We use FLT and used poly-2 as exponent
529      */
530     BN_ULONG p2[P256_LIMBS];
531     BN_ULONG p4[P256_LIMBS];
532     BN_ULONG p8[P256_LIMBS];
533     BN_ULONG p16[P256_LIMBS];
534     BN_ULONG p32[P256_LIMBS];
535     BN_ULONG res[P256_LIMBS];
536     int i;
537 
538     ecp_nistz256_sqr_mont(res, in);
539     ecp_nistz256_mul_mont(p2, res, in);         /* 3*p */
540 
541     ecp_nistz256_sqr_mont(res, p2);
542     ecp_nistz256_sqr_mont(res, res);
543     ecp_nistz256_mul_mont(p4, res, p2);         /* f*p */
544 
545     ecp_nistz256_sqr_mont(res, p4);
546     ecp_nistz256_sqr_mont(res, res);
547     ecp_nistz256_sqr_mont(res, res);
548     ecp_nistz256_sqr_mont(res, res);
549     ecp_nistz256_mul_mont(p8, res, p4);         /* ff*p */
550 
551     ecp_nistz256_sqr_mont(res, p8);
552     for (i = 0; i < 7; i++)
553         ecp_nistz256_sqr_mont(res, res);
554     ecp_nistz256_mul_mont(p16, res, p8);        /* ffff*p */
555 
556     ecp_nistz256_sqr_mont(res, p16);
557     for (i = 0; i < 15; i++)
558         ecp_nistz256_sqr_mont(res, res);
559     ecp_nistz256_mul_mont(p32, res, p16);       /* ffffffff*p */
560 
561     ecp_nistz256_sqr_mont(res, p32);
562     for (i = 0; i < 31; i++)
563         ecp_nistz256_sqr_mont(res, res);
564     ecp_nistz256_mul_mont(res, res, in);
565 
566     for (i = 0; i < 32 * 4; i++)
567         ecp_nistz256_sqr_mont(res, res);
568     ecp_nistz256_mul_mont(res, res, p32);
569 
570     for (i = 0; i < 32; i++)
571         ecp_nistz256_sqr_mont(res, res);
572     ecp_nistz256_mul_mont(res, res, p32);
573 
574     for (i = 0; i < 16; i++)
575         ecp_nistz256_sqr_mont(res, res);
576     ecp_nistz256_mul_mont(res, res, p16);
577 
578     for (i = 0; i < 8; i++)
579         ecp_nistz256_sqr_mont(res, res);
580     ecp_nistz256_mul_mont(res, res, p8);
581 
582     ecp_nistz256_sqr_mont(res, res);
583     ecp_nistz256_sqr_mont(res, res);
584     ecp_nistz256_sqr_mont(res, res);
585     ecp_nistz256_sqr_mont(res, res);
586     ecp_nistz256_mul_mont(res, res, p4);
587 
588     ecp_nistz256_sqr_mont(res, res);
589     ecp_nistz256_sqr_mont(res, res);
590     ecp_nistz256_mul_mont(res, res, p2);
591 
592     ecp_nistz256_sqr_mont(res, res);
593     ecp_nistz256_sqr_mont(res, res);
594     ecp_nistz256_mul_mont(res, res, in);
595 
596     memcpy(r, res, sizeof(res));
597 }
598 
599 /*
600  * ecp_nistz256_bignum_to_field_elem copies the contents of |in| to |out| and
601  * returns one if it fits. Otherwise it returns zero.
602  */
ecp_nistz256_bignum_to_field_elem(BN_ULONG out[P256_LIMBS],const BIGNUM * in)603 __owur static int ecp_nistz256_bignum_to_field_elem(BN_ULONG out[P256_LIMBS],
604                                                     const BIGNUM *in)
605 {
606     return bn_copy_words(out, in, P256_LIMBS);
607 }
608 
609 /* r = sum(scalar[i]*point[i]) */
ecp_nistz256_windowed_mul(const EC_GROUP * group,P256_POINT * r,const BIGNUM ** scalar,const EC_POINT ** point,size_t num,BN_CTX * ctx)610 __owur static int ecp_nistz256_windowed_mul(const EC_GROUP *group,
611                                             P256_POINT *r,
612                                             const BIGNUM **scalar,
613                                             const EC_POINT **point,
614                                             size_t num, BN_CTX *ctx)
615 {
616     size_t i;
617     int j, ret = 0;
618     unsigned int idx;
619     unsigned char (*p_str)[33] = NULL;
620     const unsigned int window_size = 5;
621     const unsigned int mask = (1 << (window_size + 1)) - 1;
622     unsigned int wvalue;
623     P256_POINT *temp;           /* place for 5 temporary points */
624     const BIGNUM **scalars = NULL;
625     P256_POINT (*table)[16] = NULL;
626     void *table_storage = NULL;
627 
628     if ((num * 16 + 6) > OPENSSL_MALLOC_MAX_NELEMS(P256_POINT)
629         || (table_storage =
630             OPENSSL_malloc((num * 16 + 5) * sizeof(P256_POINT) + 64)) == NULL
631         || (p_str =
632             OPENSSL_malloc(num * 33 * sizeof(unsigned char))) == NULL
633         || (scalars = OPENSSL_malloc(num * sizeof(BIGNUM *))) == NULL) {
634         ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL, ERR_R_MALLOC_FAILURE);
635         goto err;
636     }
637 
638     table = (void *)ALIGNPTR(table_storage, 64);
639     temp = (P256_POINT *)(table + num);
640 
641     for (i = 0; i < num; i++) {
642         P256_POINT *row = table[i];
643 
644         /* This is an unusual input, we don't guarantee constant-timeness. */
645         if ((BN_num_bits(scalar[i]) > 256) || BN_is_negative(scalar[i])) {
646             BIGNUM *mod;
647 
648             if ((mod = BN_CTX_get(ctx)) == NULL)
649                 goto err;
650             if (!BN_nnmod(mod, scalar[i], group->order, ctx)) {
651                 ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL, ERR_R_BN_LIB);
652                 goto err;
653             }
654             scalars[i] = mod;
655         } else
656             scalars[i] = scalar[i];
657 
658         for (j = 0; j < bn_get_top(scalars[i]) * BN_BYTES; j += BN_BYTES) {
659             BN_ULONG d = bn_get_words(scalars[i])[j / BN_BYTES];
660 
661             p_str[i][j + 0] = (unsigned char)d;
662             p_str[i][j + 1] = (unsigned char)(d >> 8);
663             p_str[i][j + 2] = (unsigned char)(d >> 16);
664             p_str[i][j + 3] = (unsigned char)(d >>= 24);
665             if (BN_BYTES == 8) {
666                 d >>= 8;
667                 p_str[i][j + 4] = (unsigned char)d;
668                 p_str[i][j + 5] = (unsigned char)(d >> 8);
669                 p_str[i][j + 6] = (unsigned char)(d >> 16);
670                 p_str[i][j + 7] = (unsigned char)(d >> 24);
671             }
672         }
673         for (; j < 33; j++)
674             p_str[i][j] = 0;
675 
676         if (!ecp_nistz256_bignum_to_field_elem(temp[0].X, point[i]->X)
677             || !ecp_nistz256_bignum_to_field_elem(temp[0].Y, point[i]->Y)
678             || !ecp_nistz256_bignum_to_field_elem(temp[0].Z, point[i]->Z)) {
679             ECerr(EC_F_ECP_NISTZ256_WINDOWED_MUL,
680                   EC_R_COORDINATES_OUT_OF_RANGE);
681             goto err;
682         }
683 
684         /*
685          * row[0] is implicitly (0,0,0) (the point at infinity), therefore it
686          * is not stored. All other values are actually stored with an offset
687          * of -1 in table.
688          */
689 
690         ecp_nistz256_scatter_w5  (row, &temp[0], 1);
691         ecp_nistz256_point_double(&temp[1], &temp[0]);              /*1+1=2  */
692         ecp_nistz256_scatter_w5  (row, &temp[1], 2);
693         ecp_nistz256_point_add   (&temp[2], &temp[1], &temp[0]);    /*2+1=3  */
694         ecp_nistz256_scatter_w5  (row, &temp[2], 3);
695         ecp_nistz256_point_double(&temp[1], &temp[1]);              /*2*2=4  */
696         ecp_nistz256_scatter_w5  (row, &temp[1], 4);
697         ecp_nistz256_point_double(&temp[2], &temp[2]);              /*2*3=6  */
698         ecp_nistz256_scatter_w5  (row, &temp[2], 6);
699         ecp_nistz256_point_add   (&temp[3], &temp[1], &temp[0]);    /*4+1=5  */
700         ecp_nistz256_scatter_w5  (row, &temp[3], 5);
701         ecp_nistz256_point_add   (&temp[4], &temp[2], &temp[0]);    /*6+1=7  */
702         ecp_nistz256_scatter_w5  (row, &temp[4], 7);
703         ecp_nistz256_point_double(&temp[1], &temp[1]);              /*2*4=8  */
704         ecp_nistz256_scatter_w5  (row, &temp[1], 8);
705         ecp_nistz256_point_double(&temp[2], &temp[2]);              /*2*6=12 */
706         ecp_nistz256_scatter_w5  (row, &temp[2], 12);
707         ecp_nistz256_point_double(&temp[3], &temp[3]);              /*2*5=10 */
708         ecp_nistz256_scatter_w5  (row, &temp[3], 10);
709         ecp_nistz256_point_double(&temp[4], &temp[4]);              /*2*7=14 */
710         ecp_nistz256_scatter_w5  (row, &temp[4], 14);
711         ecp_nistz256_point_add   (&temp[2], &temp[2], &temp[0]);    /*12+1=13*/
712         ecp_nistz256_scatter_w5  (row, &temp[2], 13);
713         ecp_nistz256_point_add   (&temp[3], &temp[3], &temp[0]);    /*10+1=11*/
714         ecp_nistz256_scatter_w5  (row, &temp[3], 11);
715         ecp_nistz256_point_add   (&temp[4], &temp[4], &temp[0]);    /*14+1=15*/
716         ecp_nistz256_scatter_w5  (row, &temp[4], 15);
717         ecp_nistz256_point_add   (&temp[2], &temp[1], &temp[0]);    /*8+1=9  */
718         ecp_nistz256_scatter_w5  (row, &temp[2], 9);
719         ecp_nistz256_point_double(&temp[1], &temp[1]);              /*2*8=16 */
720         ecp_nistz256_scatter_w5  (row, &temp[1], 16);
721     }
722 
723     idx = 255;
724 
725     wvalue = p_str[0][(idx - 1) / 8];
726     wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
727 
728     /*
729      * We gather to temp[0], because we know it's position relative
730      * to table
731      */
732     ecp_nistz256_gather_w5(&temp[0], table[0], _booth_recode_w5(wvalue) >> 1);
733     memcpy(r, &temp[0], sizeof(temp[0]));
734 
735     while (idx >= 5) {
736         for (i = (idx == 255 ? 1 : 0); i < num; i++) {
737             unsigned int off = (idx - 1) / 8;
738 
739             wvalue = p_str[i][off] | p_str[i][off + 1] << 8;
740             wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
741 
742             wvalue = _booth_recode_w5(wvalue);
743 
744             ecp_nistz256_gather_w5(&temp[0], table[i], wvalue >> 1);
745 
746             ecp_nistz256_neg(temp[1].Y, temp[0].Y);
747             copy_conditional(temp[0].Y, temp[1].Y, (wvalue & 1));
748 
749             ecp_nistz256_point_add(r, r, &temp[0]);
750         }
751 
752         idx -= window_size;
753 
754         ecp_nistz256_point_double(r, r);
755         ecp_nistz256_point_double(r, r);
756         ecp_nistz256_point_double(r, r);
757         ecp_nistz256_point_double(r, r);
758         ecp_nistz256_point_double(r, r);
759     }
760 
761     /* Final window */
762     for (i = 0; i < num; i++) {
763         wvalue = p_str[i][0];
764         wvalue = (wvalue << 1) & mask;
765 
766         wvalue = _booth_recode_w5(wvalue);
767 
768         ecp_nistz256_gather_w5(&temp[0], table[i], wvalue >> 1);
769 
770         ecp_nistz256_neg(temp[1].Y, temp[0].Y);
771         copy_conditional(temp[0].Y, temp[1].Y, wvalue & 1);
772 
773         ecp_nistz256_point_add(r, r, &temp[0]);
774     }
775 
776     ret = 1;
777  err:
778     OPENSSL_free(table_storage);
779     OPENSSL_free(p_str);
780     OPENSSL_free(scalars);
781     return ret;
782 }
783 
784 /* Coordinates of G, for which we have precomputed tables */
785 static const BN_ULONG def_xG[P256_LIMBS] = {
786     TOBN(0x79e730d4, 0x18a9143c), TOBN(0x75ba95fc, 0x5fedb601),
787     TOBN(0x79fb732b, 0x77622510), TOBN(0x18905f76, 0xa53755c6)
788 };
789 
790 static const BN_ULONG def_yG[P256_LIMBS] = {
791     TOBN(0xddf25357, 0xce95560a), TOBN(0x8b4ab8e4, 0xba19e45c),
792     TOBN(0xd2e88688, 0xdd21f325), TOBN(0x8571ff18, 0x25885d85)
793 };
794 
795 /*
796  * ecp_nistz256_is_affine_G returns one if |generator| is the standard, P-256
797  * generator.
798  */
ecp_nistz256_is_affine_G(const EC_POINT * generator)799 static int ecp_nistz256_is_affine_G(const EC_POINT *generator)
800 {
801     return (bn_get_top(generator->X) == P256_LIMBS) &&
802         (bn_get_top(generator->Y) == P256_LIMBS) &&
803         is_equal(bn_get_words(generator->X), def_xG) &&
804         is_equal(bn_get_words(generator->Y), def_yG) &&
805         is_one(generator->Z);
806 }
807 
ecp_nistz256_mult_precompute(EC_GROUP * group,BN_CTX * ctx)808 __owur static int ecp_nistz256_mult_precompute(EC_GROUP *group, BN_CTX *ctx)
809 {
810     /*
811      * We precompute a table for a Booth encoded exponent (wNAF) based
812      * computation. Each table holds 64 values for safe access, with an
813      * implicit value of infinity at index zero. We use window of size 7, and
814      * therefore require ceil(256/7) = 37 tables.
815      */
816     const BIGNUM *order;
817     EC_POINT *P = NULL, *T = NULL;
818     const EC_POINT *generator;
819     NISTZ256_PRE_COMP *pre_comp;
820     BN_CTX *new_ctx = NULL;
821     int i, j, k, ret = 0;
822     size_t w;
823 
824     PRECOMP256_ROW *preComputedTable = NULL;
825     unsigned char *precomp_storage = NULL;
826 
827     /* if there is an old NISTZ256_PRE_COMP object, throw it away */
828     EC_pre_comp_free(group);
829     generator = EC_GROUP_get0_generator(group);
830     if (generator == NULL) {
831         ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, EC_R_UNDEFINED_GENERATOR);
832         return 0;
833     }
834 
835     if (ecp_nistz256_is_affine_G(generator)) {
836         /*
837          * No need to calculate tables for the standard generator because we
838          * have them statically.
839          */
840         return 1;
841     }
842 
843     if ((pre_comp = ecp_nistz256_pre_comp_new(group)) == NULL)
844         return 0;
845 
846     if (ctx == NULL) {
847         ctx = new_ctx = BN_CTX_new();
848         if (ctx == NULL)
849             goto err;
850     }
851 
852     BN_CTX_start(ctx);
853 
854     order = EC_GROUP_get0_order(group);
855     if (order == NULL)
856         goto err;
857 
858     if (BN_is_zero(order)) {
859         ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, EC_R_UNKNOWN_ORDER);
860         goto err;
861     }
862 
863     w = 7;
864 
865     if ((precomp_storage =
866          OPENSSL_malloc(37 * 64 * sizeof(P256_POINT_AFFINE) + 64)) == NULL) {
867         ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE, ERR_R_MALLOC_FAILURE);
868         goto err;
869     }
870 
871     preComputedTable = (void *)ALIGNPTR(precomp_storage, 64);
872 
873     P = EC_POINT_new(group);
874     T = EC_POINT_new(group);
875     if (P == NULL || T == NULL)
876         goto err;
877 
878     /*
879      * The zero entry is implicitly infinity, and we skip it, storing other
880      * values with -1 offset.
881      */
882     if (!EC_POINT_copy(T, generator))
883         goto err;
884 
885     for (k = 0; k < 64; k++) {
886         if (!EC_POINT_copy(P, T))
887             goto err;
888         for (j = 0; j < 37; j++) {
889             P256_POINT_AFFINE temp;
890             /*
891              * It would be faster to use EC_POINTs_make_affine and
892              * make multiple points affine at the same time.
893              */
894             if (!EC_POINT_make_affine(group, P, ctx))
895                 goto err;
896             if (!ecp_nistz256_bignum_to_field_elem(temp.X, P->X) ||
897                 !ecp_nistz256_bignum_to_field_elem(temp.Y, P->Y)) {
898                 ECerr(EC_F_ECP_NISTZ256_MULT_PRECOMPUTE,
899                       EC_R_COORDINATES_OUT_OF_RANGE);
900                 goto err;
901             }
902             ecp_nistz256_scatter_w7(preComputedTable[j], &temp, k);
903             for (i = 0; i < 7; i++) {
904                 if (!EC_POINT_dbl(group, P, P, ctx))
905                     goto err;
906             }
907         }
908         if (!EC_POINT_add(group, T, T, generator, ctx))
909             goto err;
910     }
911 
912     pre_comp->group = group;
913     pre_comp->w = w;
914     pre_comp->precomp = preComputedTable;
915     pre_comp->precomp_storage = precomp_storage;
916     precomp_storage = NULL;
917     SETPRECOMP(group, nistz256, pre_comp);
918     pre_comp = NULL;
919     ret = 1;
920 
921  err:
922     BN_CTX_end(ctx);
923     BN_CTX_free(new_ctx);
924 
925     EC_nistz256_pre_comp_free(pre_comp);
926     OPENSSL_free(precomp_storage);
927     EC_POINT_free(P);
928     EC_POINT_free(T);
929     return ret;
930 }
931 
ecp_nistz256_set_from_affine(EC_POINT * out,const EC_GROUP * group,const P256_POINT_AFFINE * in,BN_CTX * ctx)932 __owur static int ecp_nistz256_set_from_affine(EC_POINT *out, const EC_GROUP *group,
933                                                const P256_POINT_AFFINE *in,
934                                                BN_CTX *ctx)
935 {
936     int ret = 0;
937 
938     if ((ret = bn_set_words(out->X, in->X, P256_LIMBS))
939         && (ret = bn_set_words(out->Y, in->Y, P256_LIMBS))
940         && (ret = bn_set_words(out->Z, ONE, P256_LIMBS)))
941         out->Z_is_one = 1;
942 
943     return ret;
944 }
945 
946 /* r = scalar*G + sum(scalars[i]*points[i]) */
ecp_nistz256_points_mul(const EC_GROUP * group,EC_POINT * r,const BIGNUM * scalar,size_t num,const EC_POINT * points[],const BIGNUM * scalars[],BN_CTX * ctx)947 __owur static int ecp_nistz256_points_mul(const EC_GROUP *group,
948                                           EC_POINT *r,
949                                           const BIGNUM *scalar,
950                                           size_t num,
951                                           const EC_POINT *points[],
952                                           const BIGNUM *scalars[], BN_CTX *ctx)
953 {
954     int i = 0, ret = 0, no_precomp_for_generator = 0, p_is_infinity = 0;
955     unsigned char p_str[33] = { 0 };
956     const PRECOMP256_ROW *preComputedTable = NULL;
957     const NISTZ256_PRE_COMP *pre_comp = NULL;
958     const EC_POINT *generator = NULL;
959     const BIGNUM **new_scalars = NULL;
960     const EC_POINT **new_points = NULL;
961     unsigned int idx = 0;
962     const unsigned int window_size = 7;
963     const unsigned int mask = (1 << (window_size + 1)) - 1;
964     unsigned int wvalue;
965     ALIGN32 union {
966         P256_POINT p;
967         P256_POINT_AFFINE a;
968     } t, p;
969     BIGNUM *tmp_scalar;
970 
971     if ((num + 1) == 0 || (num + 1) > OPENSSL_MALLOC_MAX_NELEMS(void *)) {
972         ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
973         return 0;
974     }
975 
976     BN_CTX_start(ctx);
977 
978     if (scalar) {
979         generator = EC_GROUP_get0_generator(group);
980         if (generator == NULL) {
981             ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, EC_R_UNDEFINED_GENERATOR);
982             goto err;
983         }
984 
985         /* look if we can use precomputed multiples of generator */
986         pre_comp = group->pre_comp.nistz256;
987 
988         if (pre_comp) {
989             /*
990              * If there is a precomputed table for the generator, check that
991              * it was generated with the same generator.
992              */
993             EC_POINT *pre_comp_generator = EC_POINT_new(group);
994             if (pre_comp_generator == NULL)
995                 goto err;
996 
997             ecp_nistz256_gather_w7(&p.a, pre_comp->precomp[0], 1);
998             if (!ecp_nistz256_set_from_affine(pre_comp_generator,
999                                               group, &p.a, ctx)) {
1000                 EC_POINT_free(pre_comp_generator);
1001                 goto err;
1002             }
1003 
1004             if (0 == EC_POINT_cmp(group, generator, pre_comp_generator, ctx))
1005                 preComputedTable = (const PRECOMP256_ROW *)pre_comp->precomp;
1006 
1007             EC_POINT_free(pre_comp_generator);
1008         }
1009 
1010         if (preComputedTable == NULL && ecp_nistz256_is_affine_G(generator)) {
1011             /*
1012              * If there is no precomputed data, but the generator is the
1013              * default, a hardcoded table of precomputed data is used. This
1014              * is because applications, such as Apache, do not use
1015              * EC_KEY_precompute_mult.
1016              */
1017             preComputedTable = ecp_nistz256_precomputed;
1018         }
1019 
1020         if (preComputedTable) {
1021             BN_ULONG infty;
1022 
1023             if ((BN_num_bits(scalar) > 256)
1024                 || BN_is_negative(scalar)) {
1025                 if ((tmp_scalar = BN_CTX_get(ctx)) == NULL)
1026                     goto err;
1027 
1028                 if (!BN_nnmod(tmp_scalar, scalar, group->order, ctx)) {
1029                     ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_BN_LIB);
1030                     goto err;
1031                 }
1032                 scalar = tmp_scalar;
1033             }
1034 
1035             for (i = 0; i < bn_get_top(scalar) * BN_BYTES; i += BN_BYTES) {
1036                 BN_ULONG d = bn_get_words(scalar)[i / BN_BYTES];
1037 
1038                 p_str[i + 0] = (unsigned char)d;
1039                 p_str[i + 1] = (unsigned char)(d >> 8);
1040                 p_str[i + 2] = (unsigned char)(d >> 16);
1041                 p_str[i + 3] = (unsigned char)(d >>= 24);
1042                 if (BN_BYTES == 8) {
1043                     d >>= 8;
1044                     p_str[i + 4] = (unsigned char)d;
1045                     p_str[i + 5] = (unsigned char)(d >> 8);
1046                     p_str[i + 6] = (unsigned char)(d >> 16);
1047                     p_str[i + 7] = (unsigned char)(d >> 24);
1048                 }
1049             }
1050 
1051             for (; i < 33; i++)
1052                 p_str[i] = 0;
1053 
1054             /* First window */
1055             wvalue = (p_str[0] << 1) & mask;
1056             idx += window_size;
1057 
1058             wvalue = _booth_recode_w7(wvalue);
1059 
1060             ecp_nistz256_gather_w7(&p.a, preComputedTable[0],
1061                                    wvalue >> 1);
1062 
1063             ecp_nistz256_neg(p.p.Z, p.p.Y);
1064             copy_conditional(p.p.Y, p.p.Z, wvalue & 1);
1065 
1066             /*
1067              * Since affine infinity is encoded as (0,0) and
1068              * Jacobian is (,,0), we need to harmonize them
1069              * by assigning "one" or zero to Z.
1070              */
1071             infty = (p.p.X[0] | p.p.X[1] | p.p.X[2] | p.p.X[3] |
1072                      p.p.Y[0] | p.p.Y[1] | p.p.Y[2] | p.p.Y[3]);
1073             if (P256_LIMBS == 8)
1074                 infty |= (p.p.X[4] | p.p.X[5] | p.p.X[6] | p.p.X[7] |
1075                           p.p.Y[4] | p.p.Y[5] | p.p.Y[6] | p.p.Y[7]);
1076 
1077             infty = 0 - is_zero(infty);
1078             infty = ~infty;
1079 
1080             p.p.Z[0] = ONE[0] & infty;
1081             p.p.Z[1] = ONE[1] & infty;
1082             p.p.Z[2] = ONE[2] & infty;
1083             p.p.Z[3] = ONE[3] & infty;
1084             if (P256_LIMBS == 8) {
1085                 p.p.Z[4] = ONE[4] & infty;
1086                 p.p.Z[5] = ONE[5] & infty;
1087                 p.p.Z[6] = ONE[6] & infty;
1088                 p.p.Z[7] = ONE[7] & infty;
1089             }
1090 
1091             for (i = 1; i < 37; i++) {
1092                 unsigned int off = (idx - 1) / 8;
1093                 wvalue = p_str[off] | p_str[off + 1] << 8;
1094                 wvalue = (wvalue >> ((idx - 1) % 8)) & mask;
1095                 idx += window_size;
1096 
1097                 wvalue = _booth_recode_w7(wvalue);
1098 
1099                 ecp_nistz256_gather_w7(&t.a,
1100                                        preComputedTable[i], wvalue >> 1);
1101 
1102                 ecp_nistz256_neg(t.p.Z, t.a.Y);
1103                 copy_conditional(t.a.Y, t.p.Z, wvalue & 1);
1104 
1105                 ecp_nistz256_point_add_affine(&p.p, &p.p, &t.a);
1106             }
1107         } else {
1108             p_is_infinity = 1;
1109             no_precomp_for_generator = 1;
1110         }
1111     } else
1112         p_is_infinity = 1;
1113 
1114     if (no_precomp_for_generator) {
1115         /*
1116          * Without a precomputed table for the generator, it has to be
1117          * handled like a normal point.
1118          */
1119         new_scalars = OPENSSL_malloc((num + 1) * sizeof(BIGNUM *));
1120         if (new_scalars == NULL) {
1121             ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
1122             goto err;
1123         }
1124 
1125         new_points = OPENSSL_malloc((num + 1) * sizeof(EC_POINT *));
1126         if (new_points == NULL) {
1127             ECerr(EC_F_ECP_NISTZ256_POINTS_MUL, ERR_R_MALLOC_FAILURE);
1128             goto err;
1129         }
1130 
1131         memcpy(new_scalars, scalars, num * sizeof(BIGNUM *));
1132         new_scalars[num] = scalar;
1133         memcpy(new_points, points, num * sizeof(EC_POINT *));
1134         new_points[num] = generator;
1135 
1136         scalars = new_scalars;
1137         points = new_points;
1138         num++;
1139     }
1140 
1141     if (num) {
1142         P256_POINT *out = &t.p;
1143         if (p_is_infinity)
1144             out = &p.p;
1145 
1146         if (!ecp_nistz256_windowed_mul(group, out, scalars, points, num, ctx))
1147             goto err;
1148 
1149         if (!p_is_infinity)
1150             ecp_nistz256_point_add(&p.p, &p.p, out);
1151     }
1152 
1153     /* Not constant-time, but we're only operating on the public output. */
1154     if (!bn_set_words(r->X, p.p.X, P256_LIMBS) ||
1155         !bn_set_words(r->Y, p.p.Y, P256_LIMBS) ||
1156         !bn_set_words(r->Z, p.p.Z, P256_LIMBS)) {
1157         goto err;
1158     }
1159     r->Z_is_one = is_one(r->Z) & 1;
1160 
1161     ret = 1;
1162 
1163 err:
1164     BN_CTX_end(ctx);
1165     OPENSSL_free(new_points);
1166     OPENSSL_free(new_scalars);
1167     return ret;
1168 }
1169 
ecp_nistz256_get_affine(const EC_GROUP * group,const EC_POINT * point,BIGNUM * x,BIGNUM * y,BN_CTX * ctx)1170 __owur static int ecp_nistz256_get_affine(const EC_GROUP *group,
1171                                           const EC_POINT *point,
1172                                           BIGNUM *x, BIGNUM *y, BN_CTX *ctx)
1173 {
1174     BN_ULONG z_inv2[P256_LIMBS];
1175     BN_ULONG z_inv3[P256_LIMBS];
1176     BN_ULONG x_aff[P256_LIMBS];
1177     BN_ULONG y_aff[P256_LIMBS];
1178     BN_ULONG point_x[P256_LIMBS], point_y[P256_LIMBS], point_z[P256_LIMBS];
1179     BN_ULONG x_ret[P256_LIMBS], y_ret[P256_LIMBS];
1180 
1181     if (EC_POINT_is_at_infinity(group, point)) {
1182         ECerr(EC_F_ECP_NISTZ256_GET_AFFINE, EC_R_POINT_AT_INFINITY);
1183         return 0;
1184     }
1185 
1186     if (!ecp_nistz256_bignum_to_field_elem(point_x, point->X) ||
1187         !ecp_nistz256_bignum_to_field_elem(point_y, point->Y) ||
1188         !ecp_nistz256_bignum_to_field_elem(point_z, point->Z)) {
1189         ECerr(EC_F_ECP_NISTZ256_GET_AFFINE, EC_R_COORDINATES_OUT_OF_RANGE);
1190         return 0;
1191     }
1192 
1193     ecp_nistz256_mod_inverse(z_inv3, point_z);
1194     ecp_nistz256_sqr_mont(z_inv2, z_inv3);
1195     ecp_nistz256_mul_mont(x_aff, z_inv2, point_x);
1196 
1197     if (x != NULL) {
1198         ecp_nistz256_from_mont(x_ret, x_aff);
1199         if (!bn_set_words(x, x_ret, P256_LIMBS))
1200             return 0;
1201     }
1202 
1203     if (y != NULL) {
1204         ecp_nistz256_mul_mont(z_inv3, z_inv3, z_inv2);
1205         ecp_nistz256_mul_mont(y_aff, z_inv3, point_y);
1206         ecp_nistz256_from_mont(y_ret, y_aff);
1207         if (!bn_set_words(y, y_ret, P256_LIMBS))
1208             return 0;
1209     }
1210 
1211     return 1;
1212 }
1213 
ecp_nistz256_pre_comp_new(const EC_GROUP * group)1214 static NISTZ256_PRE_COMP *ecp_nistz256_pre_comp_new(const EC_GROUP *group)
1215 {
1216     NISTZ256_PRE_COMP *ret = NULL;
1217 
1218     if (!group)
1219         return NULL;
1220 
1221     ret = OPENSSL_zalloc(sizeof(*ret));
1222 
1223     if (ret == NULL) {
1224         ECerr(EC_F_ECP_NISTZ256_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
1225         return ret;
1226     }
1227 
1228     ret->group = group;
1229     ret->w = 6;                 /* default */
1230     ret->references = 1;
1231 
1232     ret->lock = CRYPTO_THREAD_lock_new();
1233     if (ret->lock == NULL) {
1234         ECerr(EC_F_ECP_NISTZ256_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
1235         OPENSSL_free(ret);
1236         return NULL;
1237     }
1238     return ret;
1239 }
1240 
EC_nistz256_pre_comp_dup(NISTZ256_PRE_COMP * p)1241 NISTZ256_PRE_COMP *EC_nistz256_pre_comp_dup(NISTZ256_PRE_COMP *p)
1242 {
1243     int i;
1244     if (p != NULL)
1245         CRYPTO_UP_REF(&p->references, &i, p->lock);
1246     return p;
1247 }
1248 
EC_nistz256_pre_comp_free(NISTZ256_PRE_COMP * pre)1249 void EC_nistz256_pre_comp_free(NISTZ256_PRE_COMP *pre)
1250 {
1251     int i;
1252 
1253     if (pre == NULL)
1254         return;
1255 
1256     CRYPTO_DOWN_REF(&pre->references, &i, pre->lock);
1257     REF_PRINT_COUNT("EC_nistz256", x);
1258     if (i > 0)
1259         return;
1260     REF_ASSERT_ISNT(i < 0);
1261 
1262     OPENSSL_free(pre->precomp_storage);
1263     CRYPTO_THREAD_lock_free(pre->lock);
1264     OPENSSL_free(pre);
1265 }
1266 
1267 
ecp_nistz256_window_have_precompute_mult(const EC_GROUP * group)1268 static int ecp_nistz256_window_have_precompute_mult(const EC_GROUP *group)
1269 {
1270     /* There is a hard-coded table for the default generator. */
1271     const EC_POINT *generator = EC_GROUP_get0_generator(group);
1272 
1273     if (generator != NULL && ecp_nistz256_is_affine_G(generator)) {
1274         /* There is a hard-coded table for the default generator. */
1275         return 1;
1276     }
1277 
1278     return HAVEPRECOMP(group, nistz256);
1279 }
1280 
1281 #if defined(__x86_64) || defined(__x86_64__) || \
1282     defined(_M_AMD64) || defined(_M_X64) || \
1283     defined(__powerpc64__) || defined(_ARCH_PP64) || \
1284     defined(__aarch64__)
1285 /*
1286  * Montgomery mul modulo Order(P): res = a*b*2^-256 mod Order(P)
1287  */
1288 void ecp_nistz256_ord_mul_mont(BN_ULONG res[P256_LIMBS],
1289                                const BN_ULONG a[P256_LIMBS],
1290                                const BN_ULONG b[P256_LIMBS]);
1291 void ecp_nistz256_ord_sqr_mont(BN_ULONG res[P256_LIMBS],
1292                                const BN_ULONG a[P256_LIMBS],
1293                                int rep);
1294 
ecp_nistz256_inv_mod_ord(const EC_GROUP * group,BIGNUM * r,const BIGNUM * x,BN_CTX * ctx)1295 static int ecp_nistz256_inv_mod_ord(const EC_GROUP *group, BIGNUM *r,
1296                                     const BIGNUM *x, BN_CTX *ctx)
1297 {
1298     /* RR = 2^512 mod ord(p256) */
1299     static const BN_ULONG RR[P256_LIMBS]  = {
1300         TOBN(0x83244c95,0xbe79eea2), TOBN(0x4699799c,0x49bd6fa6),
1301         TOBN(0x2845b239,0x2b6bec59), TOBN(0x66e12d94,0xf3d95620)
1302     };
1303     /* The constant 1 (unlike ONE that is one in Montgomery representation) */
1304     static const BN_ULONG one[P256_LIMBS] = {
1305         TOBN(0,1), TOBN(0,0), TOBN(0,0), TOBN(0,0)
1306     };
1307     /*
1308      * We don't use entry 0 in the table, so we omit it and address
1309      * with -1 offset.
1310      */
1311     BN_ULONG table[15][P256_LIMBS];
1312     BN_ULONG out[P256_LIMBS], t[P256_LIMBS];
1313     int i, ret = 0;
1314     enum {
1315         i_1 = 0, i_10,     i_11,     i_101, i_111, i_1010, i_1111,
1316         i_10101, i_101010, i_101111, i_x6,  i_x8,  i_x16,  i_x32
1317     };
1318 
1319     /*
1320      * Catch allocation failure early.
1321      */
1322     if (bn_wexpand(r, P256_LIMBS) == NULL) {
1323         ECerr(EC_F_ECP_NISTZ256_INV_MOD_ORD, ERR_R_BN_LIB);
1324         goto err;
1325     }
1326 
1327     if ((BN_num_bits(x) > 256) || BN_is_negative(x)) {
1328         BIGNUM *tmp;
1329 
1330         if ((tmp = BN_CTX_get(ctx)) == NULL
1331             || !BN_nnmod(tmp, x, group->order, ctx)) {
1332             ECerr(EC_F_ECP_NISTZ256_INV_MOD_ORD, ERR_R_BN_LIB);
1333             goto err;
1334         }
1335         x = tmp;
1336     }
1337 
1338     if (!ecp_nistz256_bignum_to_field_elem(t, x)) {
1339         ECerr(EC_F_ECP_NISTZ256_INV_MOD_ORD, EC_R_COORDINATES_OUT_OF_RANGE);
1340         goto err;
1341     }
1342 
1343     ecp_nistz256_ord_mul_mont(table[0], t, RR);
1344 #if 0
1345     /*
1346      * Original sparse-then-fixed-window algorithm, retained for reference.
1347      */
1348     for (i = 2; i < 16; i += 2) {
1349         ecp_nistz256_ord_sqr_mont(table[i-1], table[i/2-1], 1);
1350         ecp_nistz256_ord_mul_mont(table[i], table[i-1], table[0]);
1351     }
1352 
1353     /*
1354      * The top 128bit of the exponent are highly redudndant, so we
1355      * perform an optimized flow
1356      */
1357     ecp_nistz256_ord_sqr_mont(t, table[15-1], 4);   /* f0 */
1358     ecp_nistz256_ord_mul_mont(t, t, table[15-1]);   /* ff */
1359 
1360     ecp_nistz256_ord_sqr_mont(out, t, 8);           /* ff00 */
1361     ecp_nistz256_ord_mul_mont(out, out, t);         /* ffff */
1362 
1363     ecp_nistz256_ord_sqr_mont(t, out, 16);          /* ffff0000 */
1364     ecp_nistz256_ord_mul_mont(t, t, out);           /* ffffffff */
1365 
1366     ecp_nistz256_ord_sqr_mont(out, t, 64);          /* ffffffff0000000000000000 */
1367     ecp_nistz256_ord_mul_mont(out, out, t);         /* ffffffff00000000ffffffff */
1368 
1369     ecp_nistz256_ord_sqr_mont(out, out, 32);        /* ffffffff00000000ffffffff00000000 */
1370     ecp_nistz256_ord_mul_mont(out, out, t);         /* ffffffff00000000ffffffffffffffff */
1371 
1372     /*
1373      * The bottom 128 bit of the exponent are processed with fixed 4-bit window
1374      */
1375     for(i = 0; i < 32; i++) {
1376         /* expLo - the low 128 bits of the exponent we use (ord(p256) - 2),
1377          * split into nibbles */
1378         static const unsigned char expLo[32]  = {
1379             0xb,0xc,0xe,0x6,0xf,0xa,0xa,0xd,0xa,0x7,0x1,0x7,0x9,0xe,0x8,0x4,
1380             0xf,0x3,0xb,0x9,0xc,0xa,0xc,0x2,0xf,0xc,0x6,0x3,0x2,0x5,0x4,0xf
1381         };
1382 
1383         ecp_nistz256_ord_sqr_mont(out, out, 4);
1384         /* The exponent is public, no need in constant-time access */
1385         ecp_nistz256_ord_mul_mont(out, out, table[expLo[i]-1]);
1386     }
1387 #else
1388     /*
1389      * https://briansmith.org/ecc-inversion-addition-chains-01#p256_scalar_inversion
1390      *
1391      * Even though this code path spares 12 squarings, 4.5%, and 13
1392      * multiplications, 25%, on grand scale sign operation is not that
1393      * much faster, not more that 2%...
1394      */
1395 
1396     /* pre-calculate powers */
1397     ecp_nistz256_ord_sqr_mont(table[i_10], table[i_1], 1);
1398 
1399     ecp_nistz256_ord_mul_mont(table[i_11], table[i_1], table[i_10]);
1400 
1401     ecp_nistz256_ord_mul_mont(table[i_101], table[i_11], table[i_10]);
1402 
1403     ecp_nistz256_ord_mul_mont(table[i_111], table[i_101], table[i_10]);
1404 
1405     ecp_nistz256_ord_sqr_mont(table[i_1010], table[i_101], 1);
1406 
1407     ecp_nistz256_ord_mul_mont(table[i_1111], table[i_1010], table[i_101]);
1408 
1409     ecp_nistz256_ord_sqr_mont(table[i_10101], table[i_1010], 1);
1410     ecp_nistz256_ord_mul_mont(table[i_10101], table[i_10101], table[i_1]);
1411 
1412     ecp_nistz256_ord_sqr_mont(table[i_101010], table[i_10101], 1);
1413 
1414     ecp_nistz256_ord_mul_mont(table[i_101111], table[i_101010], table[i_101]);
1415 
1416     ecp_nistz256_ord_mul_mont(table[i_x6], table[i_101010], table[i_10101]);
1417 
1418     ecp_nistz256_ord_sqr_mont(table[i_x8], table[i_x6], 2);
1419     ecp_nistz256_ord_mul_mont(table[i_x8], table[i_x8], table[i_11]);
1420 
1421     ecp_nistz256_ord_sqr_mont(table[i_x16], table[i_x8], 8);
1422     ecp_nistz256_ord_mul_mont(table[i_x16], table[i_x16], table[i_x8]);
1423 
1424     ecp_nistz256_ord_sqr_mont(table[i_x32], table[i_x16], 16);
1425     ecp_nistz256_ord_mul_mont(table[i_x32], table[i_x32], table[i_x16]);
1426 
1427     /* calculations */
1428     ecp_nistz256_ord_sqr_mont(out, table[i_x32], 64);
1429     ecp_nistz256_ord_mul_mont(out, out, table[i_x32]);
1430 
1431     for (i = 0; i < 27; i++) {
1432         static const struct { unsigned char p, i; } chain[27] = {
1433             { 32, i_x32 }, { 6,  i_101111 }, { 5,  i_111    },
1434             { 4,  i_11  }, { 5,  i_1111   }, { 5,  i_10101  },
1435             { 4,  i_101 }, { 3,  i_101    }, { 3,  i_101    },
1436             { 5,  i_111 }, { 9,  i_101111 }, { 6,  i_1111   },
1437             { 2,  i_1   }, { 5,  i_1      }, { 6,  i_1111   },
1438             { 5,  i_111 }, { 4,  i_111    }, { 5,  i_111    },
1439             { 5,  i_101 }, { 3,  i_11     }, { 10, i_101111 },
1440             { 2,  i_11  }, { 5,  i_11     }, { 5,  i_11     },
1441             { 3,  i_1   }, { 7,  i_10101  }, { 6,  i_1111   }
1442         };
1443 
1444         ecp_nistz256_ord_sqr_mont(out, out, chain[i].p);
1445         ecp_nistz256_ord_mul_mont(out, out, table[chain[i].i]);
1446     }
1447 #endif
1448     ecp_nistz256_ord_mul_mont(out, out, one);
1449 
1450     /*
1451      * Can't fail, but check return code to be consistent anyway.
1452      */
1453     if (!bn_set_words(r, out, P256_LIMBS))
1454         goto err;
1455 
1456     ret = 1;
1457 err:
1458     return ret;
1459 }
1460 #else
1461 # define ecp_nistz256_inv_mod_ord NULL
1462 #endif
1463 
EC_GFp_nistz256_method(void)1464 const EC_METHOD *EC_GFp_nistz256_method(void)
1465 {
1466     static const EC_METHOD ret = {
1467         EC_FLAGS_DEFAULT_OCT,
1468         NID_X9_62_prime_field,
1469         ec_GFp_mont_group_init,
1470         ec_GFp_mont_group_finish,
1471         ec_GFp_mont_group_clear_finish,
1472         ec_GFp_mont_group_copy,
1473         ec_GFp_mont_group_set_curve,
1474         ec_GFp_simple_group_get_curve,
1475         ec_GFp_simple_group_get_degree,
1476         ec_group_simple_order_bits,
1477         ec_GFp_simple_group_check_discriminant,
1478         ec_GFp_simple_point_init,
1479         ec_GFp_simple_point_finish,
1480         ec_GFp_simple_point_clear_finish,
1481         ec_GFp_simple_point_copy,
1482         ec_GFp_simple_point_set_to_infinity,
1483         ec_GFp_simple_set_Jprojective_coordinates_GFp,
1484         ec_GFp_simple_get_Jprojective_coordinates_GFp,
1485         ec_GFp_simple_point_set_affine_coordinates,
1486         ecp_nistz256_get_affine,
1487         0, 0, 0,
1488         ec_GFp_simple_add,
1489         ec_GFp_simple_dbl,
1490         ec_GFp_simple_invert,
1491         ec_GFp_simple_is_at_infinity,
1492         ec_GFp_simple_is_on_curve,
1493         ec_GFp_simple_cmp,
1494         ec_GFp_simple_make_affine,
1495         ec_GFp_simple_points_make_affine,
1496         ecp_nistz256_points_mul,                    /* mul */
1497         ecp_nistz256_mult_precompute,               /* precompute_mult */
1498         ecp_nistz256_window_have_precompute_mult,   /* have_precompute_mult */
1499         ec_GFp_mont_field_mul,
1500         ec_GFp_mont_field_sqr,
1501         0,                                          /* field_div */
1502         ec_GFp_mont_field_inv,
1503         ec_GFp_mont_field_encode,
1504         ec_GFp_mont_field_decode,
1505         ec_GFp_mont_field_set_to_one,
1506         ec_key_simple_priv2oct,
1507         ec_key_simple_oct2priv,
1508         0, /* set private */
1509         ec_key_simple_generate_key,
1510         ec_key_simple_check_key,
1511         ec_key_simple_generate_public_key,
1512         0, /* keycopy */
1513         0, /* keyfinish */
1514         ecdh_simple_compute_key,
1515         ecp_nistz256_inv_mod_ord,                   /* can be #define-d NULL */
1516         0,                                          /* blind_coordinates */
1517         0,                                          /* ladder_pre */
1518         0,                                          /* ladder_step */
1519         0                                           /* ladder_post */
1520     };
1521 
1522     return &ret;
1523 }
1524