crypto/ec/ecp_nistp224.c

   1 /* crypto/ec/ecp_nistp224.c */
   2 /*
   3  * Written by Emilia Kasper (Google) for the OpenSSL project.
   4  */
   5 /* Copyright 2011 Google Inc.
   6  *
   7  * Licensed under the Apache License, Version 2.0 (the "License");
   8  *
   9  * you may not use this file except in compliance with the License.
  10  * You may obtain a copy of the License at
  11  *
  12  *     http://www.apache.org/licenses/LICENSE-2.0
  13  *
  14  *  Unless required by applicable law or agreed to in writing, software
  15  *  distributed under the License is distributed on an "AS IS" BASIS,
  16  *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  17  *  See the License for the specific language governing permissions and
  18  *  limitations under the License.
  19  */
  20
  21 /*
  22  * A 64-bit implementation of the NIST P-224 elliptic curve point multiplication
  23  *
  24  * Inspired by Daniel J. Bernstein's public domain nistp224 implementation
  25  * and Adam Langley's public domain 64-bit C implementation of curve25519
  26  */
  27
  28 #include <openssl/opensslconf.h>
  29 #ifndef OPENSSL_NO_EC_NISTP_64_GCC_128
  30
  31 #include <stdint.h>
  32 #include <string.h>
  33 #include <openssl/err.h>
  34 #include "ec_lcl.h"
  35
  36 #if defined(__GNUC__) && (__GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ >= 1))
  37   /* even with gcc, the typedef won't work for 32-bit platforms */
  38   typedef __uint128_t uint128_t; /* nonstandard; implemented by gcc on 64-bit platforms */
  39 #else
  40   #error "Need GCC 3.1 or later to define type uint128_t"
  41 #endif
  42
  43 typedef uint8_t u8;
  44 typedef uint64_t u64;
  45 typedef int64_t s64;
  46
  47
  48 /******************************************************************************/
  49 /*-
  50  * INTERNAL REPRESENTATION OF FIELD ELEMENTS
  51  *
  52  * Field elements are represented as a_0 + 2^56*a_1 + 2^112*a_2 + 2^168*a_3
  53  * using 64-bit coefficients called 'limbs',
  54  * and sometimes (for multiplication results) as
  55  * b_0 + 2^56*b_1 + 2^112*b_2 + 2^168*b_3 + 2^224*b_4 + 2^280*b_5 + 2^336*b_6
  56  * using 128-bit coefficients called 'widelimbs'.
  57  * A 4-limb representation is an 'felem';
  58  * a 7-widelimb representation is a 'widefelem'.
  59  * Even within felems, bits of adjacent limbs overlap, and we don't always
  60  * reduce the representations: we ensure that inputs to each felem
  61  * multiplication satisfy a_i < 2^60, so outputs satisfy b_i < 4*2^60*2^60,
  62  * and fit into a 128-bit word without overflow. The coefficients are then
  63  * again partially reduced to obtain an felem satisfying a_i < 2^57.
  64  * We only reduce to the unique minimal representation at the end of the
  65  * computation.
  66  */
  67
  68 typedef uint64_t limb;
  69 typedef uint128_t widelimb;
  70
  71 typedef limb felem[4];
  72 typedef widelimb widefelem[7];
  73
  74 /* Field element represented as a byte arrary.
  75  * 28*8 = 224 bits is also the group order size for the elliptic curve,
  76  * and we also use this type for scalars for point multiplication.
  77   */
  78 typedef u8 felem_bytearray[28];
  79
  80 static const felem_bytearray nistp224_curve_params[5] = {
  81         {0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,    /* p */
  82          0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0x00,0x00,0x00,0x00,
  83          0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x01},
  84         {0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,    /* a */
  85          0xFF,0xFF,0xFF,0xFF,0xFF,0xFE,0xFF,0xFF,0xFF,0xFF,
  86          0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFE},
  87         {0xB4,0x05,0x0A,0x85,0x0C,0x04,0xB3,0xAB,0xF5,0x41,    /* b */
  88          0x32,0x56,0x50,0x44,0xB0,0xB7,0xD7,0xBF,0xD8,0xBA,
  89          0x27,0x0B,0x39,0x43,0x23,0x55,0xFF,0xB4},
  90         {0xB7,0x0E,0x0C,0xBD,0x6B,0xB4,0xBF,0x7F,0x32,0x13,    /* x */
  91          0x90,0xB9,0x4A,0x03,0xC1,0xD3,0x56,0xC2,0x11,0x22,
  92          0x34,0x32,0x80,0xD6,0x11,0x5C,0x1D,0x21},
  93         {0xbd,0x37,0x63,0x88,0xb5,0xf7,0x23,0xfb,0x4c,0x22,    /* y */
  94          0xdf,0xe6,0xcd,0x43,0x75,0xa0,0x5a,0x07,0x47,0x64,
  95          0x44,0xd5,0x81,0x99,0x85,0x00,0x7e,0x34}
  96 };
  97
  98 /*-
  99  * Precomputed multiples of the standard generator
 100  * Points are given in coordinates (X, Y, Z) where Z normally is 1
 101  * (0 for the point at infinity).
 102  * For each field element, slice a_0 is word 0, etc.
 103  *
 104  * The table has 2 * 16 elements, starting with the following:
 105  * index | bits    | point
 106  * ------+---------+------------------------------
 107  *     0 | 0 0 0 0 | 0G
 108  *     1 | 0 0 0 1 | 1G
 109  *     2 | 0 0 1 0 | 2^56G
 110  *     3 | 0 0 1 1 | (2^56 + 1)G
 111  *     4 | 0 1 0 0 | 2^112G
 112  *     5 | 0 1 0 1 | (2^112 + 1)G
 113  *     6 | 0 1 1 0 | (2^112 + 2^56)G
 114  *     7 | 0 1 1 1 | (2^112 + 2^56 + 1)G
 115  *     8 | 1 0 0 0 | 2^168G
 116  *     9 | 1 0 0 1 | (2^168 + 1)G
 117  *    10 | 1 0 1 0 | (2^168 + 2^56)G
 118  *    11 | 1 0 1 1 | (2^168 + 2^56 + 1)G
 119  *    12 | 1 1 0 0 | (2^168 + 2^112)G
 120  *    13 | 1 1 0 1 | (2^168 + 2^112 + 1)G
 121  *    14 | 1 1 1 0 | (2^168 + 2^112 + 2^56)G
 122  *    15 | 1 1 1 1 | (2^168 + 2^112 + 2^56 + 1)G
 123  * followed by a copy of this with each element multiplied by 2^28.
 124  *
 125  * The reason for this is so that we can clock bits into four different
 126  * locations when doing simple scalar multiplies against the base point,
 127  * and then another four locations using the second 16 elements.
 128  */
 129 static const felem gmul[2][16][3] =
 130 {{{{0, 0, 0, 0},
 131    {0, 0, 0, 0},
 132    {0, 0, 0, 0}},
 133   {{0x3280d6115c1d21, 0xc1d356c2112234, 0x7f321390b94a03, 0xb70e0cbd6bb4bf},
 134    {0xd5819985007e34, 0x75a05a07476444, 0xfb4c22dfe6cd43, 0xbd376388b5f723},
 135    {1, 0, 0, 0}},
 136   {{0xfd9675666ebbe9, 0xbca7664d40ce5e, 0x2242df8d8a2a43, 0x1f49bbb0f99bc5},
 137    {0x29e0b892dc9c43, 0xece8608436e662, 0xdc858f185310d0, 0x9812dd4eb8d321},
 138    {1, 0, 0, 0}},
 139   {{0x6d3e678d5d8eb8, 0x559eed1cb362f1, 0x16e9a3bbce8a3f, 0xeedcccd8c2a748},
 140    {0xf19f90ed50266d, 0xabf2b4bf65f9df, 0x313865468fafec, 0x5cb379ba910a17},
 141    {1, 0, 0, 0}},
 142   {{0x0641966cab26e3, 0x91fb2991fab0a0, 0xefec27a4e13a0b, 0x0499aa8a5f8ebe},
 143    {0x7510407766af5d, 0x84d929610d5450, 0x81d77aae82f706, 0x6916f6d4338c5b},
 144    {1, 0, 0, 0}},
 145   {{0xea95ac3b1f15c6, 0x086000905e82d4, 0xdd323ae4d1c8b1, 0x932b56be7685a3},
 146    {0x9ef93dea25dbbf, 0x41665960f390f0, 0xfdec76dbe2a8a7, 0x523e80f019062a},
 147    {1, 0, 0, 0}},
 148   {{0x822fdd26732c73, 0xa01c83531b5d0f, 0x363f37347c1ba4, 0xc391b45c84725c},
 149    {0xbbd5e1b2d6ad24, 0xddfbcde19dfaec, 0xc393da7e222a7f, 0x1efb7890ede244},
 150    {1, 0, 0, 0}},
 151   {{0x4c9e90ca217da1, 0xd11beca79159bb, 0xff8d33c2c98b7c, 0x2610b39409f849},
 152    {0x44d1352ac64da0, 0xcdbb7b2c46b4fb, 0x966c079b753c89, 0xfe67e4e820b112},
 153    {1, 0, 0, 0}},
 154   {{0xe28cae2df5312d, 0xc71b61d16f5c6e, 0x79b7619a3e7c4c, 0x05c73240899b47},
 155    {0x9f7f6382c73e3a, 0x18615165c56bda, 0x641fab2116fd56, 0x72855882b08394},
 156    {1, 0, 0, 0}},
 157   {{0x0469182f161c09, 0x74a98ca8d00fb5, 0xb89da93489a3e0, 0x41c98768fb0c1d},
 158    {0xe5ea05fb32da81, 0x3dce9ffbca6855, 0x1cfe2d3fbf59e6, 0x0e5e03408738a7},
 159    {1, 0, 0, 0}},
 160   {{0xdab22b2333e87f, 0x4430137a5dd2f6, 0xe03ab9f738beb8, 0xcb0c5d0dc34f24},
 161    {0x764a7df0c8fda5, 0x185ba5c3fa2044, 0x9281d688bcbe50, 0xc40331df893881},
 162    {1, 0, 0, 0}},
 163   {{0xb89530796f0f60, 0xade92bd26909a3, 0x1a0c83fb4884da, 0x1765bf22a5a984},
 164    {0x772a9ee75db09e, 0x23bc6c67cec16f, 0x4c1edba8b14e2f, 0xe2a215d9611369},
 165    {1, 0, 0, 0}},
 166   {{0x571e509fb5efb3, 0xade88696410552, 0xc8ae85fada74fe, 0x6c7e4be83bbde3},
 167    {0xff9f51160f4652, 0xb47ce2495a6539, 0xa2946c53b582f4, 0x286d2db3ee9a60},
 168    {1, 0, 0, 0}},
 169   {{0x40bbd5081a44af, 0x0995183b13926c, 0xbcefba6f47f6d0, 0x215619e9cc0057},
 170    {0x8bc94d3b0df45e, 0xf11c54a3694f6f, 0x8631b93cdfe8b5, 0xe7e3f4b0982db9},
 171    {1, 0, 0, 0}},
 172   {{0xb17048ab3e1c7b, 0xac38f36ff8a1d8, 0x1c29819435d2c6, 0xc813132f4c07e9},
 173    {0x2891425503b11f, 0x08781030579fea, 0xf5426ba5cc9674, 0x1e28ebf18562bc},
 174    {1, 0, 0, 0}},
 175   {{0x9f31997cc864eb, 0x06cd91d28b5e4c, 0xff17036691a973, 0xf1aef351497c58},
 176    {0xdd1f2d600564ff, 0xdead073b1402db, 0x74a684435bd693, 0xeea7471f962558},
 177    {1, 0, 0, 0}}},
 178  {{{0, 0, 0, 0},
 179    {0, 0, 0, 0},
 180    {0, 0, 0, 0}},
 181   {{0x9665266dddf554, 0x9613d78b60ef2d, 0xce27a34cdba417, 0xd35ab74d6afc31},
 182    {0x85ccdd22deb15e, 0x2137e5783a6aab, 0xa141cffd8c93c6, 0x355a1830e90f2d},
 183    {1, 0, 0, 0}},
 184   {{0x1a494eadaade65, 0xd6da4da77fe53c, 0xe7992996abec86, 0x65c3553c6090e3},
 185    {0xfa610b1fb09346, 0xf1c6540b8a4aaf, 0xc51a13ccd3cbab, 0x02995b1b18c28a},
 186    {1, 0, 0, 0}},
 187   {{0x7874568e7295ef, 0x86b419fbe38d04, 0xdc0690a7550d9a, 0xd3966a44beac33},
 188    {0x2b7280ec29132f, 0xbeaa3b6a032df3, 0xdc7dd88ae41200, 0xd25e2513e3a100},
 189    {1, 0, 0, 0}},
 190   {{0x924857eb2efafd, 0xac2bce41223190, 0x8edaa1445553fc, 0x825800fd3562d5},
 191    {0x8d79148ea96621, 0x23a01c3dd9ed8d, 0xaf8b219f9416b5, 0xd8db0cc277daea},
 192    {1, 0, 0, 0}},
 193   {{0x76a9c3b1a700f0, 0xe9acd29bc7e691, 0x69212d1a6b0327, 0x6322e97fe154be},
 194    {0x469fc5465d62aa, 0x8d41ed18883b05, 0x1f8eae66c52b88, 0xe4fcbe9325be51},
 195    {1, 0, 0, 0}},
 196   {{0x825fdf583cac16, 0x020b857c7b023a, 0x683c17744b0165, 0x14ffd0a2daf2f1},
 197    {0x323b36184218f9, 0x4944ec4e3b47d4, 0xc15b3080841acf, 0x0bced4b01a28bb},
 198    {1, 0, 0, 0}},
 199   {{0x92ac22230df5c4, 0x52f33b4063eda8, 0xcb3f19870c0c93, 0x40064f2ba65233},
 200    {0xfe16f0924f8992, 0x012da25af5b517, 0x1a57bb24f723a6, 0x06f8bc76760def},
 201    {1, 0, 0, 0}},
 202   {{0x4a7084f7817cb9, 0xbcab0738ee9a78, 0x3ec11e11d9c326, 0xdc0fe90e0f1aae},
 203    {0xcf639ea5f98390, 0x5c350aa22ffb74, 0x9afae98a4047b7, 0x956ec2d617fc45},
 204    {1, 0, 0, 0}},
 205   {{0x4306d648c1be6a, 0x9247cd8bc9a462, 0xf5595e377d2f2e, 0xbd1c3caff1a52e},
 206    {0x045e14472409d0, 0x29f3e17078f773, 0x745a602b2d4f7d, 0x191837685cdfbb},
 207    {1, 0, 0, 0}},
 208   {{0x5b6ee254a8cb79, 0x4953433f5e7026, 0xe21faeb1d1def4, 0xc4c225785c09de},
 209    {0x307ce7bba1e518, 0x31b125b1036db8, 0x47e91868839e8f, 0xc765866e33b9f3},
 210    {1, 0, 0, 0}},
 211   {{0x3bfece24f96906, 0x4794da641e5093, 0xde5df64f95db26, 0x297ecd89714b05},
 212    {0x701bd3ebb2c3aa, 0x7073b4f53cb1d5, 0x13c5665658af16, 0x9895089d66fe58},
 213    {1, 0, 0, 0}},
 214   {{0x0fef05f78c4790, 0x2d773633b05d2e, 0x94229c3a951c94, 0xbbbd70df4911bb},
 215    {0xb2c6963d2c1168, 0x105f47a72b0d73, 0x9fdf6111614080, 0x7b7e94b39e67b0},
 216    {1, 0, 0, 0}},
 217   {{0xad1a7d6efbe2b3, 0xf012482c0da69d, 0x6b3bdf12438345, 0x40d7558d7aa4d9},
 218    {0x8a09fffb5c6d3d, 0x9a356e5d9ffd38, 0x5973f15f4f9b1c, 0xdcd5f59f63c3ea},
 219    {1, 0, 0, 0}},
 220   {{0xacf39f4c5ca7ab, 0x4c8071cc5fd737, 0xc64e3602cd1184, 0x0acd4644c9abba},
 221    {0x6c011a36d8bf6e, 0xfecd87ba24e32a, 0x19f6f56574fad8, 0x050b204ced9405},
 222    {1, 0, 0, 0}},
 223   {{0xed4f1cae7d9a96, 0x5ceef7ad94c40a, 0x778e4a3bf3ef9b, 0x7405783dc3b55e},
 224    {0x32477c61b6e8c6, 0xb46a97570f018b, 0x91176d0a7e95d1, 0x3df90fbc4c7d0e},
 225    {1, 0, 0, 0}}}};
 226
 227 /* Precomputation for the group generator. */
 228 typedef struct {
 229         felem g_pre_comp[2][16][3];
 230         int references;
 231 } NISTP224_PRE_COMP;
 232
 233 const EC_METHOD *EC_GFp_nistp224_method(void)
 234         {
 235         static const EC_METHOD ret = {
 236                 EC_FLAGS_DEFAULT_OCT,
 237                 NID_X9_62_prime_field,
 238                 ec_GFp_nistp224_group_init,
 239                 ec_GFp_simple_group_finish,
 240                 ec_GFp_simple_group_clear_finish,
 241                 ec_GFp_nist_group_copy,
 242                 ec_GFp_nistp224_group_set_curve,
 243                 ec_GFp_simple_group_get_curve,
 244                 ec_GFp_simple_group_get_degree,
 245                 ec_GFp_simple_group_check_discriminant,
 246                 ec_GFp_simple_point_init,
 247                 ec_GFp_simple_point_finish,
 248                 ec_GFp_simple_point_clear_finish,
 249                 ec_GFp_simple_point_copy,
 250                 ec_GFp_simple_point_set_to_infinity,
 251                 ec_GFp_simple_set_Jprojective_coordinates_GFp,
 252                 ec_GFp_simple_get_Jprojective_coordinates_GFp,
 253                 ec_GFp_simple_point_set_affine_coordinates,
 254                 ec_GFp_nistp224_point_get_affine_coordinates,
 255                 0 /* point_set_compressed_coordinates */,
 256                 0 /* point2oct */,
 257                 0 /* oct2point */,
 258                 ec_GFp_simple_add,
 259                 ec_GFp_simple_dbl,
 260                 ec_GFp_simple_invert,
 261                 ec_GFp_simple_is_at_infinity,
 262                 ec_GFp_simple_is_on_curve,
 263                 ec_GFp_simple_cmp,
 264                 ec_GFp_simple_make_affine,
 265                 ec_GFp_simple_points_make_affine,
 266                 ec_GFp_nistp224_points_mul,
 267                 ec_GFp_nistp224_precompute_mult,
 268                 ec_GFp_nistp224_have_precompute_mult,
 269                 ec_GFp_nist_field_mul,
 270                 ec_GFp_nist_field_sqr,
 271                 0 /* field_div */,
 272                 0 /* field_encode */,
 273                 0 /* field_decode */,
 274                 0 /* field_set_to_one */ };
 275
 276         return &ret;
 277         }
 278
 279 /* Helper functions to convert field elements to/from internal representation */
 280 static void bin28_to_felem(felem out, const u8 in[28])
 281         {
 282         out[0] = *((const uint64_t *)(in)) & 0x00ffffffffffffff;
 283         out[1] = (*((const uint64_t *)(in+7))) & 0x00ffffffffffffff;
 284         out[2] = (*((const uint64_t *)(in+14))) & 0x00ffffffffffffff;
 285         out[3] = (*((const uint64_t *)(in+21))) & 0x00ffffffffffffff;
 286         }
 287
 288 static void felem_to_bin28(u8 out[28], const felem in)
 289         {
 290         unsigned i;
 291         for (i = 0; i < 7; ++i)
 292                 {
 293                 out[i]    = in[0]>>(8*i);
 294                 out[i+7]  = in[1]>>(8*i);
 295                 out[i+14] = in[2]>>(8*i);
 296                 out[i+21] = in[3]>>(8*i);
 297                 }
 298         }
 299
 300 /* To preserve endianness when using BN_bn2bin and BN_bin2bn */
 301 static void flip_endian(u8 *out, const u8 *in, unsigned len)
 302         {
 303         unsigned i;
 304         for (i = 0; i < len; ++i)
 305                 out[i] = in[len-1-i];
 306         }
 307
 308 /* From OpenSSL BIGNUM to internal representation */
 309 static int BN_to_felem(felem out, const BIGNUM *bn)
 310         {
 311         felem_bytearray b_in;
 312         felem_bytearray b_out;
 313         unsigned num_bytes;
 314
 315         /* BN_bn2bin eats leading zeroes */
 316         memset(b_out, 0, sizeof b_out);
 317         num_bytes = BN_num_bytes(bn);
 318         if (num_bytes > sizeof b_out)
 319                 {
 320                 ECerr(EC_F_BN_TO_FELEM, EC_R_BIGNUM_OUT_OF_RANGE);
 321                 return 0;
 322                 }
 323         if (BN_is_negative(bn))
 324                 {
 325                 ECerr(EC_F_BN_TO_FELEM, EC_R_BIGNUM_OUT_OF_RANGE);
 326                 return 0;
 327                 }
 328         num_bytes = BN_bn2bin(bn, b_in);
 329         flip_endian(b_out, b_in, num_bytes);
 330         bin28_to_felem(out, b_out);
 331         return 1;
 332         }
 333
 334 /* From internal representation to OpenSSL BIGNUM */
 335 static BIGNUM *felem_to_BN(BIGNUM *out, const felem in)
 336         {
 337         felem_bytearray b_in, b_out;
 338         felem_to_bin28(b_in, in);
 339         flip_endian(b_out, b_in, sizeof b_out);
 340         return BN_bin2bn(b_out, sizeof b_out, out);
 341         }
 342
 343 /******************************************************************************/
 344 /*-
 345  *                              FIELD OPERATIONS
 346  *
 347  * Field operations, using the internal representation of field elements.
 348  * NB! These operations are specific to our point multiplication and cannot be
 349  * expected to be correct in general - e.g., multiplication with a large scalar
 350  * will cause an overflow.
 351  *
 352  */
 353
 354 static void felem_one(felem out)
 355         {
 356         out[0] = 1;
 357         out[1] = 0;
 358         out[2] = 0;
 359         out[3] = 0;
 360         }
 361
 362 static void felem_assign(felem out, const felem in)
 363         {
 364         out[0] = in[0];
 365         out[1] = in[1];
 366         out[2] = in[2];
 367         out[3] = in[3];
 368         }
 369
 370 /* Sum two field elements: out += in */
 371 static void felem_sum(felem out, const felem in)
 372         {
 373         out[0] += in[0];
 374         out[1] += in[1];
 375         out[2] += in[2];
 376         out[3] += in[3];
 377         }
 378
 379 /* Get negative value: out = -in */
 380 /* Assumes in[i] < 2^57 */
 381 static void felem_neg(felem out, const felem in)
 382         {
 383         static const limb two58p2 = (((limb) 1) << 58) + (((limb) 1) << 2);
 384         static const limb two58m2 = (((limb) 1) << 58) - (((limb) 1) << 2);
 385         static const limb two58m42m2 = (((limb) 1) << 58) -
 386             (((limb) 1) << 42) - (((limb) 1) << 2);
 387
 388         /* Set to 0 mod 2^224-2^96+1 to ensure out > in */
 389         out[0] = two58p2 - in[0];
 390         out[1] = two58m42m2 - in[1];
 391         out[2] = two58m2 - in[2];
 392         out[3] = two58m2 - in[3];
 393         }
 394
 395 /* Subtract field elements: out -= in */
 396 /* Assumes in[i] < 2^57 */
 397 static void felem_diff(felem out, const felem in)
 398         {
 399         static const limb two58p2 = (((limb) 1) << 58) + (((limb) 1) << 2);
 400         static const limb two58m2 = (((limb) 1) << 58) - (((limb) 1) << 2);
 401         static const limb two58m42m2 = (((limb) 1) << 58) -
 402             (((limb) 1) << 42) - (((limb) 1) << 2);
 403
 404         /* Add 0 mod 2^224-2^96+1 to ensure out > in */
 405         out[0] += two58p2;
 406         out[1] += two58m42m2;
 407         out[2] += two58m2;
 408         out[3] += two58m2;
 409
 410         out[0] -= in[0];
 411         out[1] -= in[1];
 412         out[2] -= in[2];
 413         out[3] -= in[3];
 414         }
 415
 416 /* Subtract in unreduced 128-bit mode: out -= in */
 417 /* Assumes in[i] < 2^119 */
 418 static void widefelem_diff(widefelem out, const widefelem in)
 419         {
 420         static const widelimb two120 = ((widelimb) 1) << 120;
 421         static const widelimb two120m64 = (((widelimb) 1) << 120) -
 422                 (((widelimb) 1) << 64);
 423         static const widelimb two120m104m64 = (((widelimb) 1) << 120) -
 424                 (((widelimb) 1) << 104) - (((widelimb) 1) << 64);
 425
 426         /* Add 0 mod 2^224-2^96+1 to ensure out > in */
 427         out[0] += two120;
 428         out[1] += two120m64;
 429         out[2] += two120m64;
 430         out[3] += two120;
 431         out[4] += two120m104m64;
 432         out[5] += two120m64;
 433         out[6] += two120m64;
 434
 435         out[0] -= in[0];
 436         out[1] -= in[1];
 437         out[2] -= in[2];
 438         out[3] -= in[3];
 439         out[4] -= in[4];
 440         out[5] -= in[5];
 441         out[6] -= in[6];
 442         }
 443
 444 /* Subtract in mixed mode: out128 -= in64 */
 445 /* in[i] < 2^63 */
 446 static void felem_diff_128_64(widefelem out, const felem in)
 447         {
 448         static const widelimb two64p8 = (((widelimb) 1) << 64) +
 449                 (((widelimb) 1) << 8);
 450         static const widelimb two64m8 = (((widelimb) 1) << 64) -
 451                 (((widelimb) 1) << 8);
 452         static const widelimb two64m48m8 = (((widelimb) 1) << 64) -
 453                 (((widelimb) 1) << 48) - (((widelimb) 1) << 8);
 454
 455         /* Add 0 mod 2^224-2^96+1 to ensure out > in */
 456         out[0] += two64p8;
 457         out[1] += two64m48m8;
 458         out[2] += two64m8;
 459         out[3] += two64m8;
 460
 461         out[0] -= in[0];
 462         out[1] -= in[1];
 463         out[2] -= in[2];
 464         out[3] -= in[3];
 465         }
 466
 467 /* Multiply a field element by a scalar: out = out * scalar
 468  * The scalars we actually use are small, so results fit without overflow */
 469 static void felem_scalar(felem out, const limb scalar)
 470         {
 471         out[0] *= scalar;
 472         out[1] *= scalar;
 473         out[2] *= scalar;
 474         out[3] *= scalar;
 475         }
 476
 477 /* Multiply an unreduced field element by a scalar: out = out * scalar
 478  * The scalars we actually use are small, so results fit without overflow */
 479 static void widefelem_scalar(widefelem out, const widelimb scalar)
 480         {
 481         out[0] *= scalar;
 482         out[1] *= scalar;
 483         out[2] *= scalar;
 484         out[3] *= scalar;
 485         out[4] *= scalar;
 486         out[5] *= scalar;
 487         out[6] *= scalar;
 488         }
 489
 490 /* Square a field element: out = in^2 */
 491 static void felem_square(widefelem out, const felem in)
 492         {
 493         limb tmp0, tmp1, tmp2;
 494         tmp0 = 2 * in[0]; tmp1 = 2 * in[1]; tmp2 = 2 * in[2];
 495         out[0] = ((widelimb) in[0]) * in[0];
 496         out[1] = ((widelimb) in[0]) * tmp1;
 497         out[2] = ((widelimb) in[0]) * tmp2 + ((widelimb) in[1]) * in[1];
 498         out[3] = ((widelimb) in[3]) * tmp0 +
 499                 ((widelimb) in[1]) * tmp2;
 500         out[4] = ((widelimb) in[3]) * tmp1 + ((widelimb) in[2]) * in[2];
 501         out[5] = ((widelimb) in[3]) * tmp2;
 502         out[6] = ((widelimb) in[3]) * in[3];
 503         }
 504
 505 /* Multiply two field elements: out = in1 * in2 */
 506 static void felem_mul(widefelem out, const felem in1, const felem in2)
 507         {
 508         out[0] = ((widelimb) in1[0]) * in2[0];
 509         out[1] = ((widelimb) in1[0]) * in2[1] + ((widelimb) in1[1]) * in2[0];
 510         out[2] = ((widelimb) in1[0]) * in2[2] + ((widelimb) in1[1]) * in2[1] +
 511                 ((widelimb) in1[2]) * in2[0];
 512         out[3] = ((widelimb) in1[0]) * in2[3] + ((widelimb) in1[1]) * in2[2] +
 513                 ((widelimb) in1[2]) * in2[1] + ((widelimb) in1[3]) * in2[0];
 514         out[4] = ((widelimb) in1[1]) * in2[3] + ((widelimb) in1[2]) * in2[2] +
 515                 ((widelimb) in1[3]) * in2[1];
 516         out[5] = ((widelimb) in1[2]) * in2[3] + ((widelimb) in1[3]) * in2[2];
 517         out[6] = ((widelimb) in1[3]) * in2[3];
 518         }
 519
 520 /*-
 521  * Reduce seven 128-bit coefficients to four 64-bit coefficients.
 522  * Requires in[i] < 2^126,
 523  * ensures out[0] < 2^56, out[1] < 2^56, out[2] < 2^56, out[3] <= 2^56 + 2^16 */
 524 static void felem_reduce(felem out, const widefelem in)
 525         {
 526         static const widelimb two127p15 = (((widelimb) 1) << 127) +
 527                 (((widelimb) 1) << 15);
 528         static const widelimb two127m71 = (((widelimb) 1) << 127) -
 529                 (((widelimb) 1) << 71);
 530         static const widelimb two127m71m55 = (((widelimb) 1) << 127) -
 531                 (((widelimb) 1) << 71) - (((widelimb) 1) << 55);
 532         widelimb output[5];
 533
 534         /* Add 0 mod 2^224-2^96+1 to ensure all differences are positive */
 535         output[0] = in[0] + two127p15;
 536         output[1] = in[1] + two127m71m55;
 537         output[2] = in[2] + two127m71;
 538         output[3] = in[3];
 539         output[4] = in[4];
 540
 541         /* Eliminate in[4], in[5], in[6] */
 542         output[4] += in[6] >> 16;
 543         output[3] += (in[6] & 0xffff) << 40;
 544         output[2] -= in[6];
 545
 546         output[3] += in[5] >> 16;
 547         output[2] += (in[5] & 0xffff) << 40;
 548         output[1] -= in[5];
 549
 550         output[2] += output[4] >> 16;
 551         output[1] += (output[4] & 0xffff) << 40;
 552         output[0] -= output[4];
 553
 554         /* Carry 2 -> 3 -> 4 */
 555         output[3] += output[2] >> 56;
 556         output[2] &= 0x00ffffffffffffff;
 557
 558         output[4] = output[3] >> 56;
 559         output[3] &= 0x00ffffffffffffff;
 560
 561         /* Now output[2] < 2^56, output[3] < 2^56, output[4] < 2^72 */
 562
 563         /* Eliminate output[4] */
 564         output[2] += output[4] >> 16;
 565         /* output[2] < 2^56 + 2^56 = 2^57 */
 566         output[1] += (output[4] & 0xffff) << 40;
 567         output[0] -= output[4];
 568
 569         /* Carry 0 -> 1 -> 2 -> 3 */
 570         output[1] += output[0] >> 56;
 571         out[0] = output[0] & 0x00ffffffffffffff;
 572
 573         output[2] += output[1] >> 56;
 574         /* output[2] < 2^57 + 2^72 */
 575         out[1] = output[1] & 0x00ffffffffffffff;
 576         output[3] += output[2] >> 56;
 577         /* output[3] <= 2^56 + 2^16 */
 578         out[2] = output[2] & 0x00ffffffffffffff;
 579
 580         /*-
 581          * out[0] < 2^56, out[1] < 2^56, out[2] < 2^56,
 582          * out[3] <= 2^56 + 2^16 (due to final carry),
 583          * so out < 2*p
 584          */
 585         out[3] = output[3];
 586         }
 587
 588 static void felem_square_reduce(felem out, const felem in)
 589         {
 590         widefelem tmp;
 591         felem_square(tmp, in);
 592         felem_reduce(out, tmp);
 593         }
 594
 595 static void felem_mul_reduce(felem out, const felem in1, const felem in2)
 596         {
 597         widefelem tmp;
 598         felem_mul(tmp, in1, in2);
 599         felem_reduce(out, tmp);
 600         }
 601
 602 /* Reduce to unique minimal representation.
 603  * Requires 0 <= in < 2*p (always call felem_reduce first) */
 604 static void felem_contract(felem out, const felem in)
 605         {
 606         static const int64_t two56 = ((limb) 1) << 56;
 607         /* 0 <= in < 2*p, p = 2^224 - 2^96 + 1 */
 608         /* if in > p , reduce in = in - 2^224 + 2^96 - 1 */
 609         int64_t tmp[4], a;
 610         tmp[0] = in[0];
 611         tmp[1] = in[1];
 612         tmp[2] = in[2];
 613         tmp[3] = in[3];
 614         /* Case 1: a = 1 iff in >= 2^224 */
 615         a = (in[3] >> 56);
 616         tmp[0] -= a;
 617         tmp[1] += a << 40;
 618         tmp[3] &= 0x00ffffffffffffff;
 619         /* Case 2: a = 0 iff p <= in < 2^224, i.e.,
 620          * the high 128 bits are all 1 and the lower part is non-zero */
 621         a = ((in[3] & in[2] & (in[1] | 0x000000ffffffffff)) + 1) |
 622                 (((int64_t)(in[0] + (in[1] & 0x000000ffffffffff)) - 1) >> 63);
 623         a &= 0x00ffffffffffffff;
 624         /* turn a into an all-one mask (if a = 0) or an all-zero mask */
 625         a = (a - 1) >> 63;
 626         /* subtract 2^224 - 2^96 + 1 if a is all-one*/
 627         tmp[3] &= a ^ 0xffffffffffffffff;
 628         tmp[2] &= a ^ 0xffffffffffffffff;
 629         tmp[1] &= (a ^ 0xffffffffffffffff) | 0x000000ffffffffff;
 630         tmp[0] -= 1 & a;
 631
 632         /* eliminate negative coefficients: if tmp[0] is negative, tmp[1] must
 633          * be non-zero, so we only need one step */
 634         a = tmp[0] >> 63;
 635         tmp[0] += two56 & a;
 636         tmp[1] -= 1 & a;
 637
 638         /* carry 1 -> 2 -> 3 */
 639         tmp[2] += tmp[1] >> 56;
 640         tmp[1] &= 0x00ffffffffffffff;
 641
 642         tmp[3] += tmp[2] >> 56;
 643         tmp[2] &= 0x00ffffffffffffff;
 644
 645         /* Now 0 <= out < p */
 646         out[0] = tmp[0];
 647         out[1] = tmp[1];
 648         out[2] = tmp[2];
 649         out[3] = tmp[3];
 650         }
 651
 652 /* Zero-check: returns 1 if input is 0, and 0 otherwise.
 653  * We know that field elements are reduced to in < 2^225,
 654  * so we only need to check three cases: 0, 2^224 - 2^96 + 1,
 655  * and 2^225 - 2^97 + 2 */
 656 static limb felem_is_zero(const felem in)
 657         {
 658         limb zero, two224m96p1, two225m97p2;
 659
 660         zero = in[0] | in[1] | in[2] | in[3];
 661         zero = (((int64_t)(zero) - 1) >> 63) & 1;
 662         two224m96p1 = (in[0] ^ 1) | (in[1] ^ 0x00ffff0000000000)
 663                 | (in[2] ^ 0x00ffffffffffffff) | (in[3] ^ 0x00ffffffffffffff);
 664         two224m96p1 = (((int64_t)(two224m96p1) - 1) >> 63) & 1;
 665         two225m97p2 = (in[0] ^ 2) | (in[1] ^ 0x00fffe0000000000)
 666                 | (in[2] ^ 0x00ffffffffffffff) | (in[3] ^ 0x01ffffffffffffff);
 667         two225m97p2 = (((int64_t)(two225m97p2) - 1) >> 63) & 1;
 668         return (zero | two224m96p1 | two225m97p2);
 669         }
 670
 671 static limb felem_is_zero_int(const felem in)
 672         {
 673         return (int) (felem_is_zero(in) & ((limb)1));
 674         }
 675
 676 /* Invert a field element */
 677 /* Computation chain copied from djb's code */
 678 static void felem_inv(felem out, const felem in)
 679         {
 680         felem ftmp, ftmp2, ftmp3, ftmp4;
 681         widefelem tmp;
 682         unsigned i;
 683
 684         felem_square(tmp, in); felem_reduce(ftmp, tmp);         /* 2 */
 685         felem_mul(tmp, in, ftmp); felem_reduce(ftmp, tmp);      /* 2^2 - 1 */
 686         felem_square(tmp, ftmp); felem_reduce(ftmp, tmp);       /* 2^3 - 2 */
 687         felem_mul(tmp, in, ftmp); felem_reduce(ftmp, tmp);      /* 2^3 - 1 */
 688         felem_square(tmp, ftmp); felem_reduce(ftmp2, tmp);      /* 2^4 - 2 */
 689         felem_square(tmp, ftmp2); felem_reduce(ftmp2, tmp);     /* 2^5 - 4 */
 690         felem_square(tmp, ftmp2); felem_reduce(ftmp2, tmp);     /* 2^6 - 8 */
 691         felem_mul(tmp, ftmp2, ftmp); felem_reduce(ftmp, tmp);   /* 2^6 - 1 */
 692         felem_square(tmp, ftmp); felem_reduce(ftmp2, tmp);      /* 2^7 - 2 */
 693         for (i = 0; i < 5; ++i)                                 /* 2^12 - 2^6 */
 694                 {
 695                 felem_square(tmp, ftmp2); felem_reduce(ftmp2, tmp);
 696                 }
 697         felem_mul(tmp, ftmp2, ftmp); felem_reduce(ftmp2, tmp);  /* 2^12 - 1 */
 698         felem_square(tmp, ftmp2); felem_reduce(ftmp3, tmp);     /* 2^13 - 2 */
 699         for (i = 0; i < 11; ++i)                                /* 2^24 - 2^12 */
 700                 {
 701                 felem_square(tmp, ftmp3); felem_reduce(ftmp3, tmp);
 702                 }
 703         felem_mul(tmp, ftmp3, ftmp2); felem_reduce(ftmp2, tmp); /* 2^24 - 1 */
 704         felem_square(tmp, ftmp2); felem_reduce(ftmp3, tmp);     /* 2^25 - 2 */
 705         for (i = 0; i < 23; ++i)                                /* 2^48 - 2^24 */
 706                 {
 707                 felem_square(tmp, ftmp3); felem_reduce(ftmp3, tmp);
 708                 }
 709         felem_mul(tmp, ftmp3, ftmp2); felem_reduce(ftmp3, tmp); /* 2^48 - 1 */
 710         felem_square(tmp, ftmp3); felem_reduce(ftmp4, tmp);     /* 2^49 - 2 */
 711         for (i = 0; i < 47; ++i)                                /* 2^96 - 2^48 */
 712                 {
 713                 felem_square(tmp, ftmp4); felem_reduce(ftmp4, tmp);
 714                 }
 715         felem_mul(tmp, ftmp3, ftmp4); felem_reduce(ftmp3, tmp); /* 2^96 - 1 */
 716         felem_square(tmp, ftmp3); felem_reduce(ftmp4, tmp);     /* 2^97 - 2 */
 717         for (i = 0; i < 23; ++i)                                /* 2^120 - 2^24 */
 718                 {
 719                 felem_square(tmp, ftmp4); felem_reduce(ftmp4, tmp);
 720                 }
 721         felem_mul(tmp, ftmp2, ftmp4); felem_reduce(ftmp2, tmp); /* 2^120 - 1 */
 722         for (i = 0; i < 6; ++i)                                 /* 2^126 - 2^6 */
 723                 {
 724                 felem_square(tmp, ftmp2); felem_reduce(ftmp2, tmp);
 725                 }
 726         felem_mul(tmp, ftmp2, ftmp); felem_reduce(ftmp, tmp);   /* 2^126 - 1 */
 727         felem_square(tmp, ftmp); felem_reduce(ftmp, tmp);       /* 2^127 - 2 */
 728         felem_mul(tmp, ftmp, in); felem_reduce(ftmp, tmp);      /* 2^127 - 1 */
 729         for (i = 0; i < 97; ++i)                                /* 2^224 - 2^97 */
 730                 {
 731                 felem_square(tmp, ftmp); felem_reduce(ftmp, tmp);
 732                 }
 733         felem_mul(tmp, ftmp, ftmp3); felem_reduce(out, tmp);    /* 2^224 - 2^96 - 1 */
 734         }
 735
 736 /* Copy in constant time:
 737  * if icopy == 1, copy in to out,
 738  * if icopy == 0, copy out to itself. */
 739 static void
 740 copy_conditional(felem out, const felem in, limb icopy)
 741         {
 742         unsigned i;
 743         /* icopy is a (64-bit) 0 or 1, so copy is either all-zero or all-one */
 744         const limb copy = -icopy;
 745         for (i = 0; i < 4; ++i)
 746                 {
 747                 const limb tmp = copy & (in[i] ^ out[i]);
 748                 out[i] ^= tmp;
 749                 }
 750         }
 751
 752 /******************************************************************************/
 753 /*-
 754  *                       ELLIPTIC CURVE POINT OPERATIONS
 755  *
 756  * Points are represented in Jacobian projective coordinates:
 757  * (X, Y, Z) corresponds to the affine point (X/Z^2, Y/Z^3),
 758  * or to the point at infinity if Z == 0.
 759  *
 760  */
 761
 762 /*-
 763  * Double an elliptic curve point:
 764  * (X', Y', Z') = 2 * (X, Y, Z), where
 765  * X' = (3 * (X - Z^2) * (X + Z^2))^2 - 8 * X * Y^2
 766  * Y' = 3 * (X - Z^2) * (X + Z^2) * (4 * X * Y^2 - X') - 8 * Y^2
 767  * Z' = (Y + Z)^2 - Y^2 - Z^2 = 2 * Y * Z
 768  * Outputs can equal corresponding inputs, i.e., x_out == x_in is allowed,
 769  * while x_out == y_in is not (maybe this works, but it's not tested).
 770  */
 771 static void
 772 point_double(felem x_out, felem y_out, felem z_out,
 773              const felem x_in, const felem y_in, const felem z_in)
 774         {
 775         widefelem tmp, tmp2;
 776         felem delta, gamma, beta, alpha, ftmp, ftmp2;
 777
 778         felem_assign(ftmp, x_in);
 779         felem_assign(ftmp2, x_in);
 780
 781         /* delta = z^2 */
 782         felem_square(tmp, z_in);
 783         felem_reduce(delta, tmp);
 784
 785         /* gamma = y^2 */
 786         felem_square(tmp, y_in);
 787         felem_reduce(gamma, tmp);
 788
 789         /* beta = x*gamma */
 790         felem_mul(tmp, x_in, gamma);
 791         felem_reduce(beta, tmp);
 792
 793         /* alpha = 3*(x-delta)*(x+delta) */
 794         felem_diff(ftmp, delta);
 795         /* ftmp[i] < 2^57 + 2^58 + 2 < 2^59 */
 796         felem_sum(ftmp2, delta);
 797         /* ftmp2[i] < 2^57 + 2^57 = 2^58 */
 798         felem_scalar(ftmp2, 3);
 799         /* ftmp2[i] < 3 * 2^58 < 2^60 */
 800         felem_mul(tmp, ftmp, ftmp2);
 801         /* tmp[i] < 2^60 * 2^59 * 4 = 2^121 */
 802         felem_reduce(alpha, tmp);
 803
 804         /* x' = alpha^2 - 8*beta */
 805         felem_square(tmp, alpha);
 806         /* tmp[i] < 4 * 2^57 * 2^57 = 2^116 */
 807         felem_assign(ftmp, beta);
 808         felem_scalar(ftmp, 8);
 809         /* ftmp[i] < 8 * 2^57 = 2^60 */
 810         felem_diff_128_64(tmp, ftmp);
 811         /* tmp[i] < 2^116 + 2^64 + 8 < 2^117 */
 812         felem_reduce(x_out, tmp);
 813
 814         /* z' = (y + z)^2 - gamma - delta */
 815         felem_sum(delta, gamma);
 816         /* delta[i] < 2^57 + 2^57 = 2^58 */
 817         felem_assign(ftmp, y_in);
 818         felem_sum(ftmp, z_in);
 819         /* ftmp[i] < 2^57 + 2^57 = 2^58 */
 820         felem_square(tmp, ftmp);
 821         /* tmp[i] < 4 * 2^58 * 2^58 = 2^118 */
 822         felem_diff_128_64(tmp, delta);
 823         /* tmp[i] < 2^118 + 2^64 + 8 < 2^119 */
 824         felem_reduce(z_out, tmp);
 825
 826         /* y' = alpha*(4*beta - x') - 8*gamma^2 */
 827         felem_scalar(beta, 4);
 828         /* beta[i] < 4 * 2^57 = 2^59 */
 829         felem_diff(beta, x_out);
 830         /* beta[i] < 2^59 + 2^58 + 2 < 2^60 */
 831         felem_mul(tmp, alpha, beta);
 832         /* tmp[i] < 4 * 2^57 * 2^60 = 2^119 */
 833         felem_square(tmp2, gamma);
 834         /* tmp2[i] < 4 * 2^57 * 2^57 = 2^116 */
 835         widefelem_scalar(tmp2, 8);
 836         /* tmp2[i] < 8 * 2^116 = 2^119 */
 837         widefelem_diff(tmp, tmp2);
 838         /* tmp[i] < 2^119 + 2^120 < 2^121 */
 839         felem_reduce(y_out, tmp);
 840         }
 841
 842 /*-
 843  * Add two elliptic curve points:
 844  * (X_1, Y_1, Z_1) + (X_2, Y_2, Z_2) = (X_3, Y_3, Z_3), where
 845  * X_3 = (Z_1^3 * Y_2 - Z_2^3 * Y_1)^2 - (Z_1^2 * X_2 - Z_2^2 * X_1)^3 -
 846  * 2 * Z_2^2 * X_1 * (Z_1^2 * X_2 - Z_2^2 * X_1)^2
 847  * Y_3 = (Z_1^3 * Y_2 - Z_2^3 * Y_1) * (Z_2^2 * X_1 * (Z_1^2 * X_2 - Z_2^2 * X_1)^2 - X_3) -
 848  *        Z_2^3 * Y_1 * (Z_1^2 * X_2 - Z_2^2 * X_1)^3
 849  * Z_3 = (Z_1^2 * X_2 - Z_2^2 * X_1) * (Z_1 * Z_2)
 850  *
 851  * This runs faster if 'mixed' is set, which requires Z_2 = 1 or Z_2 = 0.
 852  */
 853
 854 /* This function is not entirely constant-time:
 855  * it includes a branch for checking whether the two input points are equal,
 856  * (while not equal to the point at infinity).
 857  * This case never happens during single point multiplication,
 858  * so there is no timing leak for ECDH or ECDSA signing. */
 859 static void point_add(felem x3, felem y3, felem z3,
 860         const felem x1, const felem y1, const felem z1,
 861         const int mixed, const felem x2, const felem y2, const felem z2)
 862         {
 863         felem ftmp, ftmp2, ftmp3, ftmp4, ftmp5, x_out, y_out, z_out;
 864         widefelem tmp, tmp2;
 865         limb z1_is_zero, z2_is_zero, x_equal, y_equal;
 866
 867         if (!mixed)
 868                 {
 869                 /* ftmp2 = z2^2 */
 870                 felem_square(tmp, z2);
 871                 felem_reduce(ftmp2, tmp);
 872
 873                 /* ftmp4 = z2^3 */
 874                 felem_mul(tmp, ftmp2, z2);
 875                 felem_reduce(ftmp4, tmp);
 876
 877                 /* ftmp4 = z2^3*y1 */
 878                 felem_mul(tmp2, ftmp4, y1);
 879                 felem_reduce(ftmp4, tmp2);
 880
 881                 /* ftmp2 = z2^2*x1 */
 882                 felem_mul(tmp2, ftmp2, x1);
 883                 felem_reduce(ftmp2, tmp2);
 884                 }
 885         else
 886                 {
 887                 /* We'll assume z2 = 1 (special case z2 = 0 is handled later) */
 888
 889                 /* ftmp4 = z2^3*y1 */
 890                 felem_assign(ftmp4, y1);
 891
 892                 /* ftmp2 = z2^2*x1 */
 893                 felem_assign(ftmp2, x1);
 894                 }
 895
 896         /* ftmp = z1^2 */
 897         felem_square(tmp, z1);
 898         felem_reduce(ftmp, tmp);
 899
 900         /* ftmp3 = z1^3 */
 901         felem_mul(tmp, ftmp, z1);
 902         felem_reduce(ftmp3, tmp);
 903
 904         /* tmp = z1^3*y2 */
 905         felem_mul(tmp, ftmp3, y2);
 906         /* tmp[i] < 4 * 2^57 * 2^57 = 2^116 */
 907
 908         /* ftmp3 = z1^3*y2 - z2^3*y1 */
 909         felem_diff_128_64(tmp, ftmp4);
 910         /* tmp[i] < 2^116 + 2^64 + 8 < 2^117 */
 911         felem_reduce(ftmp3, tmp);
 912
 913         /* tmp = z1^2*x2 */
 914         felem_mul(tmp, ftmp, x2);
 915         /* tmp[i] < 4 * 2^57 * 2^57 = 2^116 */
 916
 917         /* ftmp = z1^2*x2 - z2^2*x1 */
 918         felem_diff_128_64(tmp, ftmp2);
 919         /* tmp[i] < 2^116 + 2^64 + 8 < 2^117 */
 920         felem_reduce(ftmp, tmp);
 921
 922         /* the formulae are incorrect if the points are equal
 923          * so we check for this and do doubling if this happens */
 924         x_equal = felem_is_zero(ftmp);
 925         y_equal = felem_is_zero(ftmp3);
 926         z1_is_zero = felem_is_zero(z1);
 927         z2_is_zero = felem_is_zero(z2);
 928         /* In affine coordinates, (X_1, Y_1) == (X_2, Y_2) */
 929         if (x_equal && y_equal && !z1_is_zero && !z2_is_zero)
 930                 {
 931                 point_double(x3, y3, z3, x1, y1, z1);
 932                 return;
 933                 }
 934
 935         /* ftmp5 = z1*z2 */
 936         if (!mixed)
 937                 {
 938                 felem_mul(tmp, z1, z2);
 939                 felem_reduce(ftmp5, tmp);
 940                 }
 941         else
 942                 {
 943                 /* special case z2 = 0 is handled later */
 944                 felem_assign(ftmp5, z1);
 945                 }
 946
 947         /* z_out = (z1^2*x2 - z2^2*x1)*(z1*z2) */
 948         felem_mul(tmp, ftmp, ftmp5);
 949         felem_reduce(z_out, tmp);
 950
 951         /* ftmp = (z1^2*x2 - z2^2*x1)^2 */
 952         felem_assign(ftmp5, ftmp);
 953         felem_square(tmp, ftmp);
 954         felem_reduce(ftmp, tmp);
 955
 956         /* ftmp5 = (z1^2*x2 - z2^2*x1)^3 */
 957         felem_mul(tmp, ftmp, ftmp5);
 958         felem_reduce(ftmp5, tmp);
 959
 960         /* ftmp2 = z2^2*x1*(z1^2*x2 - z2^2*x1)^2 */
 961         felem_mul(tmp, ftmp2, ftmp);
 962         felem_reduce(ftmp2, tmp);
 963
 964         /* tmp = z2^3*y1*(z1^2*x2 - z2^2*x1)^3 */
 965         felem_mul(tmp, ftmp4, ftmp5);
 966         /* tmp[i] < 4 * 2^57 * 2^57 = 2^116 */
 967
 968         /* tmp2 = (z1^3*y2 - z2^3*y1)^2 */
 969         felem_square(tmp2, ftmp3);
 970         /* tmp2[i] < 4 * 2^57 * 2^57 < 2^116 */
 971
 972         /* tmp2 = (z1^3*y2 - z2^3*y1)^2 - (z1^2*x2 - z2^2*x1)^3 */
 973         felem_diff_128_64(tmp2, ftmp5);
 974         /* tmp2[i] < 2^116 + 2^64 + 8 < 2^117 */
 975
 976         /* ftmp5 = 2*z2^2*x1*(z1^2*x2 - z2^2*x1)^2 */
 977         felem_assign(ftmp5, ftmp2);
 978         felem_scalar(ftmp5, 2);
 979         /* ftmp5[i] < 2 * 2^57 = 2^58 */
 980
 981         /*-
 982          * x_out = (z1^3*y2 - z2^3*y1)^2 - (z1^2*x2 - z2^2*x1)^3 -
 983          *  2*z2^2*x1*(z1^2*x2 - z2^2*x1)^2
 984          */
 985         felem_diff_128_64(tmp2, ftmp5);
 986         /* tmp2[i] < 2^117 + 2^64 + 8 < 2^118 */
 987         felem_reduce(x_out, tmp2);
 988
 989         /* ftmp2 = z2^2*x1*(z1^2*x2 - z2^2*x1)^2 - x_out */
 990         felem_diff(ftmp2, x_out);
 991         /* ftmp2[i] < 2^57 + 2^58 + 2 < 2^59 */
 992
 993         /* tmp2 = (z1^3*y2 - z2^3*y1)*(z2^2*x1*(z1^2*x2 - z2^2*x1)^2 - x_out) */
 994         felem_mul(tmp2, ftmp3, ftmp2);
 995         /* tmp2[i] < 4 * 2^57 * 2^59 = 2^118 */
 996
 997         /*-
 998          * y_out = (z1^3*y2 - z2^3*y1)*(z2^2*x1*(z1^2*x2 - z2^2*x1)^2 - x_out) -
 999          *  z2^3*y1*(z1^2*x2 - z2^2*x1)^3
1000          */
1001         widefelem_diff(tmp2, tmp);
1002         /* tmp2[i] < 2^118 + 2^120 < 2^121 */
1003         felem_reduce(y_out, tmp2);
1004
1005         /* the result (x_out, y_out, z_out) is incorrect if one of the inputs is
1006          * the point at infinity, so we need to check for this separately */
1007
1008         /* if point 1 is at infinity, copy point 2 to output, and vice versa */
1009         copy_conditional(x_out, x2, z1_is_zero);
1010         copy_conditional(x_out, x1, z2_is_zero);
1011         copy_conditional(y_out, y2, z1_is_zero);
1012         copy_conditional(y_out, y1, z2_is_zero);
1013         copy_conditional(z_out, z2, z1_is_zero);
1014         copy_conditional(z_out, z1, z2_is_zero);
1015         felem_assign(x3, x_out);
1016         felem_assign(y3, y_out);
1017         felem_assign(z3, z_out);
1018         }
1019
1020 /* select_point selects the |idx|th point from a precomputation table and
1021  * copies it to out. */
1022 static void select_point(const u64 idx, unsigned int size, const felem pre_comp[/*size*/][3], felem out[3])
1023         {
1024         unsigned i, j;
1025         limb *outlimbs = &out[0][0];
1026         memset(outlimbs, 0, 3 * sizeof(felem));
1027
1028         for (i = 0; i < size; i++)
1029                 {
1030                 const limb *inlimbs = &pre_comp[i][0][0];
1031                 u64 mask = i ^ idx;
1032                 mask |= mask >> 4;
1033                 mask |= mask >> 2;
1034                 mask |= mask >> 1;
1035                 mask &= 1;
1036                 mask--;
1037                 for (j = 0; j < 4 * 3; j++)
1038                         outlimbs[j] |= inlimbs[j] & mask;
1039                 }
1040         }
1041
1042 /* get_bit returns the |i|th bit in |in| */
1043 static char get_bit(const felem_bytearray in, unsigned i)
1044         {
1045         if (i >= 224)
1046                 return 0;
1047         return (in[i >> 3] >> (i & 7)) & 1;
1048         }
1049
1050 /* Interleaved point multiplication using precomputed point multiples:
1051  * The small point multiples 0*P, 1*P, ..., 16*P are in pre_comp[],
1052  * the scalars in scalars[]. If g_scalar is non-NULL, we also add this multiple
1053  * of the generator, using certain (large) precomputed multiples in g_pre_comp.
1054  * Output point (X, Y, Z) is stored in x_out, y_out, z_out */
1055 static void batch_mul(felem x_out, felem y_out, felem z_out,
1056         const felem_bytearray scalars[], const unsigned num_points, const u8 *g_scalar,
1057         const int mixed, const felem pre_comp[][17][3], const felem g_pre_comp[2][16][3])
1058         {
1059         int i, skip;
1060         unsigned num;
1061         unsigned gen_mul = (g_scalar != NULL);
1062         felem nq[3], tmp[4];
1063         u64 bits;
1064         u8 sign, digit;
1065
1066         /* set nq to the point at infinity */
1067         memset(nq, 0, 3 * sizeof(felem));
1068
1069         /* Loop over all scalars msb-to-lsb, interleaving additions
1070          * of multiples of the generator (two in each of the last 28 rounds)
1071          * and additions of other points multiples (every 5th round).
1072          */
1073         skip = 1; /* save two point operations in the first round */
1074         for (i = (num_points ? 220 : 27); i >= 0; --i)
1075                 {
1076                 /* double */
1077                 if (!skip)
1078                         point_double(nq[0], nq[1], nq[2], nq[0], nq[1], nq[2]);
1079
1080                 /* add multiples of the generator */
1081                 if (gen_mul && (i <= 27))
1082                         {
1083                         /* first, look 28 bits upwards */
1084                         bits = get_bit(g_scalar, i + 196) << 3;
1085                         bits |= get_bit(g_scalar, i + 140) << 2;
1086                         bits |= get_bit(g_scalar, i + 84) << 1;
1087                         bits |= get_bit(g_scalar, i + 28);
1088                         /* select the point to add, in constant time */
1089                         select_point(bits, 16, g_pre_comp[1], tmp);
1090
1091                         if (!skip)
1092                                 {
1093                                 point_add(nq[0], nq[1], nq[2],
1094                                         nq[0], nq[1], nq[2],
1095                                         1 /* mixed */, tmp[0], tmp[1], tmp[2]);
1096                                 }
1097                         else
1098                                 {
1099                                 memcpy(nq, tmp, 3 * sizeof(felem));
1100                                 skip = 0;
1101                                 }
1102
1103                         /* second, look at the current position */
1104                         bits = get_bit(g_scalar, i + 168) << 3;
1105                         bits |= get_bit(g_scalar, i + 112) << 2;
1106                         bits |= get_bit(g_scalar, i + 56) << 1;
1107                         bits |= get_bit(g_scalar, i);
1108                         /* select the point to add, in constant time */
1109                         select_point(bits, 16, g_pre_comp[0], tmp);
1110                         point_add(nq[0], nq[1], nq[2],
1111                                 nq[0], nq[1], nq[2],
1112                                 1 /* mixed */, tmp[0], tmp[1], tmp[2]);
1113                         }
1114
1115                 /* do other additions every 5 doublings */
1116                 if (num_points && (i % 5 == 0))
1117                         {
1118                         /* loop over all scalars */
1119                         for (num = 0; num < num_points; ++num)
1120                                 {
1121                                 bits = get_bit(scalars[num], i + 4) << 5;
1122                                 bits |= get_bit(scalars[num], i + 3) << 4;
1123                                 bits |= get_bit(scalars[num], i + 2) << 3;
1124                                 bits |= get_bit(scalars[num], i + 1) << 2;
1125                                 bits |= get_bit(scalars[num], i) << 1;
1126                                 bits |= get_bit(scalars[num], i - 1);
1127                                 ec_GFp_nistp_recode_scalar_bits(&sign, &digit, bits);
1128
1129                                 /* select the point to add or subtract */
1130                                 select_point(digit, 17, pre_comp[num], tmp);
1131                                 felem_neg(tmp[3], tmp[1]); /* (X, -Y, Z) is the negative point */
1132                                 copy_conditional(tmp[1], tmp[3], sign);
1133
1134                                 if (!skip)
1135                                         {
1136                                         point_add(nq[0], nq[1], nq[2],
1137                                                 nq[0], nq[1], nq[2],
1138                                                 mixed, tmp[0], tmp[1], tmp[2]);
1139                                         }
1140                                 else
1141                                         {
1142                                         memcpy(nq, tmp, 3 * sizeof(felem));
1143                                         skip = 0;
1144                                         }
1145                                 }
1146                         }
1147                 }
1148         felem_assign(x_out, nq[0]);
1149         felem_assign(y_out, nq[1]);
1150         felem_assign(z_out, nq[2]);
1151         }
1152
1153 /******************************************************************************/
1154 /*                     FUNCTIONS TO MANAGE PRECOMPUTATION
1155  */
1156
1157 static NISTP224_PRE_COMP *nistp224_pre_comp_new()
1158         {
1159         NISTP224_PRE_COMP *ret = NULL;
1160         ret = (NISTP224_PRE_COMP *) OPENSSL_malloc(sizeof *ret);
1161         if (!ret)
1162                 {
1163                 ECerr(EC_F_NISTP224_PRE_COMP_NEW, ERR_R_MALLOC_FAILURE);
1164                 return ret;
1165                 }
1166         memset(ret->g_pre_comp, 0, sizeof(ret->g_pre_comp));
1167         ret->references = 1;
1168         return ret;
1169         }
1170
1171 static void *nistp224_pre_comp_dup(void *src_)
1172         {
1173         NISTP224_PRE_COMP *src = src_;
1174
1175         /* no need to actually copy, these objects never change! */
1176         CRYPTO_add(&src->references, 1, CRYPTO_LOCK_EC_PRE_COMP);
1177
1178         return src_;
1179         }
1180
1181 static void nistp224_pre_comp_free(void *pre_)
1182         {
1183         int i;
1184         NISTP224_PRE_COMP *pre = pre_;
1185
1186         if (!pre)
1187                 return;
1188
1189         i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP);
1190         if (i > 0)
1191                 return;
1192
1193         OPENSSL_free(pre);
1194         }
1195
1196 static void nistp224_pre_comp_clear_free(void *pre_)
1197         {
1198         int i;
1199         NISTP224_PRE_COMP *pre = pre_;
1200
1201         if (!pre)
1202                 return;
1203
1204         i = CRYPTO_add(&pre->references, -1, CRYPTO_LOCK_EC_PRE_COMP);
1205         if (i > 0)
1206                 return;
1207
1208         OPENSSL_cleanse(pre, sizeof *pre);
1209         OPENSSL_free(pre);
1210         }
1211
1212 /******************************************************************************/
1213 /*                         OPENSSL EC_METHOD FUNCTIONS
1214  */
1215
1216 int ec_GFp_nistp224_group_init(EC_GROUP *group)
1217         {
1218         int ret;
1219         ret = ec_GFp_simple_group_init(group);
1220         group->a_is_minus3 = 1;
1221         return ret;
1222         }
1223
1224 int ec_GFp_nistp224_group_set_curve(EC_GROUP *group, const BIGNUM *p,
1225         const BIGNUM *a, const BIGNUM *b, BN_CTX *ctx)
1226         {
1227         int ret = 0;
1228         BN_CTX *new_ctx = NULL;
1229         BIGNUM *curve_p, *curve_a, *curve_b;
1230
1231         if (ctx == NULL)
1232                 if ((ctx = new_ctx = BN_CTX_new()) == NULL) return 0;
1233         BN_CTX_start(ctx);
1234         if (((curve_p = BN_CTX_get(ctx)) == NULL) ||
1235                 ((curve_a = BN_CTX_get(ctx)) == NULL) ||
1236                 ((curve_b = BN_CTX_get(ctx)) == NULL)) goto err;
1237         BN_bin2bn(nistp224_curve_params[0], sizeof(felem_bytearray), curve_p);
1238         BN_bin2bn(nistp224_curve_params[1], sizeof(felem_bytearray), curve_a);
1239         BN_bin2bn(nistp224_curve_params[2], sizeof(felem_bytearray), curve_b);
1240         if ((BN_cmp(curve_p, p)) || (BN_cmp(curve_a, a)) ||
1241                 (BN_cmp(curve_b, b)))
1242                 {
1243                 ECerr(EC_F_EC_GFP_NISTP224_GROUP_SET_CURVE,
1244                         EC_R_WRONG_CURVE_PARAMETERS);
1245                 goto err;
1246                 }
1247         group->field_mod_func = BN_nist_mod_224;
1248         ret = ec_GFp_simple_group_set_curve(group, p, a, b, ctx);
1249 err:
1250         BN_CTX_end(ctx);
1251         if (new_ctx != NULL)
1252                 BN_CTX_free(new_ctx);
1253         return ret;
1254         }
1255
1256 /* Takes the Jacobian coordinates (X, Y, Z) of a point and returns
1257  * (X', Y') = (X/Z^2, Y/Z^3) */
1258 int ec_GFp_nistp224_point_get_affine_coordinates(const EC_GROUP *group,
1259         const EC_POINT *point, BIGNUM *x, BIGNUM *y, BN_CTX *ctx)
1260         {
1261         felem z1, z2, x_in, y_in, x_out, y_out;
1262         widefelem tmp;
1263
1264         if (EC_POINT_is_at_infinity(group, point))
1265                 {
1266                 ECerr(EC_F_EC_GFP_NISTP224_POINT_GET_AFFINE_COORDINATES,
1267                         EC_R_POINT_AT_INFINITY);
1268                 return 0;
1269                 }
1270         if ((!BN_to_felem(x_in, &point->X)) || (!BN_to_felem(y_in, &point->Y)) ||
1271                 (!BN_to_felem(z1, &point->Z))) return 0;
1272         felem_inv(z2, z1);
1273         felem_square(tmp, z2); felem_reduce(z1, tmp);
1274         felem_mul(tmp, x_in, z1); felem_reduce(x_in, tmp);
1275         felem_contract(x_out, x_in);
1276         if (x != NULL)
1277                 {
1278                 if (!felem_to_BN(x, x_out)) {
1279                 ECerr(EC_F_EC_GFP_NISTP224_POINT_GET_AFFINE_COORDINATES,
1280                         ERR_R_BN_LIB);
1281                 return 0;
1282                 }
1283                 }
1284         felem_mul(tmp, z1, z2); felem_reduce(z1, tmp);
1285         felem_mul(tmp, y_in, z1); felem_reduce(y_in, tmp);
1286         felem_contract(y_out, y_in);
1287         if (y != NULL)
1288                 {
1289                 if (!felem_to_BN(y, y_out)) {
1290                 ECerr(EC_F_EC_GFP_NISTP224_POINT_GET_AFFINE_COORDINATES,
1291                         ERR_R_BN_LIB);
1292                 return 0;
1293                 }
1294                 }
1295         return 1;
1296         }
1297
1298 static void make_points_affine(size_t num, felem points[/*num*/][3], felem tmp_felems[/*num+1*/])
1299         {
1300         /* Runs in constant time, unless an input is the point at infinity
1301          * (which normally shouldn't happen). */
1302         ec_GFp_nistp_points_make_affine_internal(
1303                 num,
1304                 points,
1305                 sizeof(felem),
1306                 tmp_felems,
1307                 (void (*)(void *)) felem_one,
1308                 (int (*)(const void *)) felem_is_zero_int,
1309                 (void (*)(void *, const void *)) felem_assign,
1310                 (void (*)(void *, const void *)) felem_square_reduce,
1311                 (void (*)(void *, const void *, const void *)) felem_mul_reduce,
1312                 (void (*)(void *, const void *)) felem_inv,
1313                 (void (*)(void *, const void *)) felem_contract);
1314         }
1315
1316 /* Computes scalar*generator + \sum scalars[i]*points[i], ignoring NULL values
1317  * Result is stored in r (r can equal one of the inputs). */
1318 int ec_GFp_nistp224_points_mul(const EC_GROUP *group, EC_POINT *r,
1319         const BIGNUM *scalar, size_t num, const EC_POINT *points[],
1320         const BIGNUM *scalars[], BN_CTX *ctx)
1321         {
1322         int ret = 0;
1323         int j;
1324         unsigned i;
1325         int mixed = 0;
1326         BN_CTX *new_ctx = NULL;
1327         BIGNUM *x, *y, *z, *tmp_scalar;
1328         felem_bytearray g_secret;
1329         felem_bytearray *secrets = NULL;
1330         felem (*pre_comp)[17][3] = NULL;
1331         felem *tmp_felems = NULL;
1332         felem_bytearray tmp;
1333         unsigned num_bytes;
1334         int have_pre_comp = 0;
1335         size_t num_points = num;
1336         felem x_in, y_in, z_in, x_out, y_out, z_out;
1337         NISTP224_PRE_COMP *pre = NULL;
1338         const felem (*g_pre_comp)[16][3] = NULL;
1339         EC_POINT *generator = NULL;
1340         const EC_POINT *p = NULL;
1341         const BIGNUM *p_scalar = NULL;
1342
1343         if (ctx == NULL)
1344                 if ((ctx = new_ctx = BN_CTX_new()) == NULL) return 0;
1345         BN_CTX_start(ctx);
1346         if (((x = BN_CTX_get(ctx)) == NULL) ||
1347                 ((y = BN_CTX_get(ctx)) == NULL) ||
1348                 ((z = BN_CTX_get(ctx)) == NULL) ||
1349                 ((tmp_scalar = BN_CTX_get(ctx)) == NULL))
1350                 goto err;
1351
1352         if (scalar != NULL)
1353                 {
1354                 pre = EC_EX_DATA_get_data(group->extra_data,
1355                         nistp224_pre_comp_dup, nistp224_pre_comp_free,
1356                         nistp224_pre_comp_clear_free);
1357                 if (pre)
1358                         /* we have precomputation, try to use it */
1359                         g_pre_comp = (const felem (*)[16][3]) pre->g_pre_comp;
1360                 else
1361                         /* try to use the standard precomputation */
1362                         g_pre_comp = &gmul[0];
1363                 generator = EC_POINT_new(group);
1364                 if (generator == NULL)
1365                         goto err;
1366                 /* get the generator from precomputation */
1367                 if (!felem_to_BN(x, g_pre_comp[0][1][0]) ||
1368                         !felem_to_BN(y, g_pre_comp[0][1][1]) ||
1369                         !felem_to_BN(z, g_pre_comp[0][1][2]))
1370                         {
1371                         ECerr(EC_F_EC_GFP_NISTP224_POINTS_MUL, ERR_R_BN_LIB);
1372                         goto err;
1373                         }
1374                 if (!EC_POINT_set_Jprojective_coordinates_GFp(group,
1375                                 generator, x, y, z, ctx))
1376                         goto err;
1377                 if (0 == EC_POINT_cmp(group, generator, group->generator, ctx))
1378                         /* precomputation matches generator */
1379                         have_pre_comp = 1;
1380                 else
1381                         /* we don't have valid precomputation:
1382                          * treat the generator as a random point */
1383                         num_points = num_points + 1;
1384                 }
1385
1386         if (num_points > 0)
1387                 {
1388                 if (num_points >= 3)
1389                         {
1390                         /* unless we precompute multiples for just one or two points,
1391                          * converting those into affine form is time well spent  */
1392                         mixed = 1;
1393                         }
1394                 secrets = OPENSSL_malloc(num_points * sizeof(felem_bytearray));
1395                 pre_comp = OPENSSL_malloc(num_points * 17 * 3 * sizeof(felem));
1396                 if (mixed)
1397                         tmp_felems = OPENSSL_malloc((num_points * 17 + 1) * sizeof(felem));
1398                 if ((secrets == NULL) || (pre_comp == NULL) || (mixed && (tmp_felems == NULL)))
1399                         {
1400                         ECerr(EC_F_EC_GFP_NISTP224_POINTS_MUL, ERR_R_MALLOC_FAILURE);
1401                         goto err;
1402                         }
1403
1404                 /* we treat NULL scalars as 0, and NULL points as points at infinity,
1405                  * i.e., they contribute nothing to the linear combination */
1406                 memset(secrets, 0, num_points * sizeof(felem_bytearray));
1407                 memset(pre_comp, 0, num_points * 17 * 3 * sizeof(felem));
1408                 for (i = 0; i < num_points; ++i)
1409                         {
1410                         if (i == num)
1411                                 /* the generator */
1412                                 {
1413                                 p = EC_GROUP_get0_generator(group);
1414                                 p_scalar = scalar;
1415                                 }
1416                         else
1417                                 /* the i^th point */
1418                                 {
1419                                 p = points[i];
1420                                 p_scalar = scalars[i];
1421                                 }
1422                         if ((p_scalar != NULL) && (p != NULL))
1423                                 {
1424                                 /* reduce scalar to 0 <= scalar < 2^224 */
1425                                 if ((BN_num_bits(p_scalar) > 224) || (BN_is_negative(p_scalar)))
1426                                         {
1427                                         /* this is an unusual input, and we don't guarantee
1428                                          * constant-timeness */
1429                                         if (!BN_nnmod(tmp_scalar, p_scalar, &group->order, ctx))
1430                                                 {
1431                                                 ECerr(EC_F_EC_GFP_NISTP224_POINTS_MUL, ERR_R_BN_LIB);
1432                                                 goto err;
1433                                                 }
1434                                         num_bytes = BN_bn2bin(tmp_scalar, tmp);
1435                                         }
1436                                 else
1437                                         num_bytes = BN_bn2bin(p_scalar, tmp);
1438                                 flip_endian(secrets[i], tmp, num_bytes);
1439                                 /* precompute multiples */
1440                                 if ((!BN_to_felem(x_out, &p->X)) ||
1441                                         (!BN_to_felem(y_out, &p->Y)) ||
1442                                         (!BN_to_felem(z_out, &p->Z))) goto err;
1443                                 felem_assign(pre_comp[i][1][0], x_out);
1444                                 felem_assign(pre_comp[i][1][1], y_out);
1445                                 felem_assign(pre_comp[i][1][2], z_out);
1446                                 for (j = 2; j <= 16; ++j)
1447                                         {
1448                                         if (j & 1)
1449                                                 {
1450                                                 point_add(
1451                                                         pre_comp[i][j][0], pre_comp[i][j][1], pre_comp[i][j][2],
1452                                                         pre_comp[i][1][0], pre_comp[i][1][1], pre_comp[i][1][2],
1453                                                         0, pre_comp[i][j-1][0], pre_comp[i][j-1][1], pre_comp[i][j-1][2]);
1454                                                 }
1455                                         else
1456                                                 {
1457                                                 point_double(
1458                                                         pre_comp[i][j][0], pre_comp[i][j][1], pre_comp[i][j][2],
1459                                                         pre_comp[i][j/2][0], pre_comp[i][j/2][1], pre_comp[i][j/2][2]);
1460                                                 }
1461                                         }
1462                                 }
1463                         }
1464                 if (mixed)
1465                         make_points_affine(num_points * 17, pre_comp[0], tmp_felems);
1466                 }
1467
1468         /* the scalar for the generator */
1469         if ((scalar != NULL) && (have_pre_comp))
1470                 {
1471                 memset(g_secret, 0, sizeof g_secret);
1472                 /* reduce scalar to 0 <= scalar < 2^224 */
1473                 if ((BN_num_bits(scalar) > 224) || (BN_is_negative(scalar)))
1474                         {
1475                         /* this is an unusual input, and we don't guarantee
1476                          * constant-timeness */
1477                         if (!BN_nnmod(tmp_scalar, scalar, &group->order, ctx))
1478                                 {
1479                                 ECerr(EC_F_EC_GFP_NISTP224_POINTS_MUL, ERR_R_BN_LIB);
1480                                 goto err;
1481                                 }
1482                         num_bytes = BN_bn2bin(tmp_scalar, tmp);
1483                         }
1484                 else
1485                         num_bytes = BN_bn2bin(scalar, tmp);
1486                 flip_endian(g_secret, tmp, num_bytes);
1487                 /* do the multiplication with generator precomputation*/
1488                 batch_mul(x_out, y_out, z_out,
1489                         (const felem_bytearray (*)) secrets, num_points,
1490                         g_secret,
1491                         mixed, (const felem (*)[17][3]) pre_comp,
1492                         g_pre_comp);
1493                 }
1494         else
1495                 /* do the multiplication without generator precomputation */
1496                 batch_mul(x_out, y_out, z_out,
1497                         (const felem_bytearray (*)) secrets, num_points,
1498                         NULL, mixed, (const felem (*)[17][3]) pre_comp, NULL);
1499         /* reduce the output to its unique minimal representation */
1500         felem_contract(x_in, x_out);
1501         felem_contract(y_in, y_out);
1502         felem_contract(z_in, z_out);
1503         if ((!felem_to_BN(x, x_in)) || (!felem_to_BN(y, y_in)) ||
1504                 (!felem_to_BN(z, z_in)))
1505                 {
1506                 ECerr(EC_F_EC_GFP_NISTP224_POINTS_MUL, ERR_R_BN_LIB);
1507                 goto err;
1508                 }
1509         ret = EC_POINT_set_Jprojective_coordinates_GFp(group, r, x, y, z, ctx);
1510
1511 err:
1512         BN_CTX_end(ctx);
1513         if (generator != NULL)
1514                 EC_POINT_free(generator);
1515         if (new_ctx != NULL)
1516                 BN_CTX_free(new_ctx);
1517         if (secrets != NULL)
1518                 OPENSSL_free(secrets);
1519         if (pre_comp != NULL)
1520                 OPENSSL_free(pre_comp);
1521         if (tmp_felems != NULL)
1522                 OPENSSL_free(tmp_felems);
1523         return ret;
1524         }
1525
1526 int ec_GFp_nistp224_precompute_mult(EC_GROUP *group, BN_CTX *ctx)
1527         {
1528         int ret = 0;
1529         NISTP224_PRE_COMP *pre = NULL;
1530         int i, j;
1531         BN_CTX *new_ctx = NULL;
1532         BIGNUM *x, *y;
1533         EC_POINT *generator = NULL;
1534         felem tmp_felems[32];
1535
1536         /* throw away old precomputation */
1537         EC_EX_DATA_free_data(&group->extra_data, nistp224_pre_comp_dup,
1538                 nistp224_pre_comp_free, nistp224_pre_comp_clear_free);
1539         if (ctx == NULL)
1540                 if ((ctx = new_ctx = BN_CTX_new()) == NULL) return 0;
1541         BN_CTX_start(ctx);
1542         if (((x = BN_CTX_get(ctx)) == NULL) ||
1543                 ((y = BN_CTX_get(ctx)) == NULL))
1544                 goto err;
1545         /* get the generator */
1546         if (group->generator == NULL) goto err;
1547         generator = EC_POINT_new(group);
1548         if (generator == NULL)
1549                 goto err;
1550         BN_bin2bn(nistp224_curve_params[3], sizeof (felem_bytearray), x);
1551         BN_bin2bn(nistp224_curve_params[4], sizeof (felem_bytearray), y);
1552         if (!EC_POINT_set_affine_coordinates_GFp(group, generator, x, y, ctx))
1553                 goto err;
1554         if ((pre = nistp224_pre_comp_new()) == NULL)
1555                 goto err;
1556         /* if the generator is the standard one, use built-in precomputation */
1557         if (0 == EC_POINT_cmp(group, generator, group->generator, ctx))
1558                 {
1559                 memcpy(pre->g_pre_comp, gmul, sizeof(pre->g_pre_comp));
1560                 ret = 1;
1561                 goto err;
1562                 }
1563         if ((!BN_to_felem(pre->g_pre_comp[0][1][0], &group->generator->X)) ||
1564                 (!BN_to_felem(pre->g_pre_comp[0][1][1], &group->generator->Y)) ||
1565                 (!BN_to_felem(pre->g_pre_comp[0][1][2], &group->generator->Z)))
1566                 goto err;
1567         /* compute 2^56*G, 2^112*G, 2^168*G for the first table,
1568          * 2^28*G, 2^84*G, 2^140*G, 2^196*G for the second one
1569          */
1570         for (i = 1; i <= 8; i <<= 1)
1571                 {
1572                 point_double(
1573                         pre->g_pre_comp[1][i][0], pre->g_pre_comp[1][i][1], pre->g_pre_comp[1][i][2],
1574                         pre->g_pre_comp[0][i][0], pre->g_pre_comp[0][i][1], pre->g_pre_comp[0][i][2]);
1575                 for (j = 0; j < 27; ++j)
1576                         {
1577                         point_double(
1578                                 pre->g_pre_comp[1][i][0], pre->g_pre_comp[1][i][1], pre->g_pre_comp[1][i][2],
1579                                 pre->g_pre_comp[1][i][0], pre->g_pre_comp[1][i][1], pre->g_pre_comp[1][i][2]);
1580                         }
1581                 if (i == 8)
1582                         break;
1583                 point_double(
1584                         pre->g_pre_comp[0][2*i][0], pre->g_pre_comp[0][2*i][1], pre->g_pre_comp[0][2*i][2],
1585                         pre->g_pre_comp[1][i][0], pre->g_pre_comp[1][i][1], pre->g_pre_comp[1][i][2]);
1586                 for (j = 0; j < 27; ++j)
1587                         {
1588                         point_double(
1589                                 pre->g_pre_comp[0][2*i][0], pre->g_pre_comp[0][2*i][1], pre->g_pre_comp[0][2*i][2],
1590                                 pre->g_pre_comp[0][2*i][0], pre->g_pre_comp[0][2*i][1], pre->g_pre_comp[0][2*i][2]);
1591                         }
1592                 }
1593         for (i = 0; i < 2; i++)
1594                 {
1595                 /* g_pre_comp[i][0] is the point at infinity */
1596                 memset(pre->g_pre_comp[i][0], 0, sizeof(pre->g_pre_comp[i][0]));
1597                 /* the remaining multiples */
1598                 /* 2^56*G + 2^112*G resp. 2^84*G + 2^140*G */
1599                 point_add(
1600                         pre->g_pre_comp[i][6][0], pre->g_pre_comp[i][6][1],
1601                         pre->g_pre_comp[i][6][2], pre->g_pre_comp[i][4][0],
1602                         pre->g_pre_comp[i][4][1], pre->g_pre_comp[i][4][2],
1603                         0, pre->g_pre_comp[i][2][0], pre->g_pre_comp[i][2][1],
1604                         pre->g_pre_comp[i][2][2]);
1605                 /* 2^56*G + 2^168*G resp. 2^84*G + 2^196*G */
1606                 point_add(
1607                         pre->g_pre_comp[i][10][0], pre->g_pre_comp[i][10][1],
1608                         pre->g_pre_comp[i][10][2], pre->g_pre_comp[i][8][0],
1609                         pre->g_pre_comp[i][8][1], pre->g_pre_comp[i][8][2],
1610                         0, pre->g_pre_comp[i][2][0], pre->g_pre_comp[i][2][1],
1611                         pre->g_pre_comp[i][2][2]);
1612                 /* 2^112*G + 2^168*G resp. 2^140*G + 2^196*G */
1613                 point_add(
1614                         pre->g_pre_comp[i][12][0], pre->g_pre_comp[i][12][1],
1615                         pre->g_pre_comp[i][12][2], pre->g_pre_comp[i][8][0],
1616                         pre->g_pre_comp[i][8][1], pre->g_pre_comp[i][8][2],
1617                         0, pre->g_pre_comp[i][4][0], pre->g_pre_comp[i][4][1],
1618                         pre->g_pre_comp[i][4][2]);
1619                 /* 2^56*G + 2^112*G + 2^168*G resp. 2^84*G + 2^140*G + 2^196*G */
1620                 point_add(
1621                         pre->g_pre_comp[i][14][0], pre->g_pre_comp[i][14][1],
1622                         pre->g_pre_comp[i][14][2], pre->g_pre_comp[i][12][0],
1623                         pre->g_pre_comp[i][12][1], pre->g_pre_comp[i][12][2],
1624                         0, pre->g_pre_comp[i][2][0], pre->g_pre_comp[i][2][1],
1625                         pre->g_pre_comp[i][2][2]);
1626                 for (j = 1; j < 8; ++j)
1627                         {
1628                         /* odd multiples: add G resp. 2^28*G */
1629                         point_add(
1630                                 pre->g_pre_comp[i][2*j+1][0], pre->g_pre_comp[i][2*j+1][1],
1631                                 pre->g_pre_comp[i][2*j+1][2], pre->g_pre_comp[i][2*j][0],
1632                                 pre->g_pre_comp[i][2*j][1], pre->g_pre_comp[i][2*j][2],
1633                                 0, pre->g_pre_comp[i][1][0], pre->g_pre_comp[i][1][1],
1634                                 pre->g_pre_comp[i][1][2]);
1635                         }
1636                 }
1637         make_points_affine(31, &(pre->g_pre_comp[0][1]), tmp_felems);
1638
1639         if (!EC_EX_DATA_set_data(&group->extra_data, pre, nistp224_pre_comp_dup,
1640                         nistp224_pre_comp_free, nistp224_pre_comp_clear_free))
1641                 goto err;
1642         ret = 1;
1643         pre = NULL;
1644  err:
1645         BN_CTX_end(ctx);
1646         if (generator != NULL)
1647                 EC_POINT_free(generator);
1648         if (new_ctx != NULL)
1649                 BN_CTX_free(new_ctx);
1650         if (pre)
1651                 nistp224_pre_comp_free(pre);
1652         return ret;
1653         }
1654
1655 int ec_GFp_nistp224_have_precompute_mult(const EC_GROUP *group)
1656         {
1657         if (EC_EX_DATA_get_data(group->extra_data, nistp224_pre_comp_dup,
1658                         nistp224_pre_comp_free, nistp224_pre_comp_clear_free)
1659                 != NULL)
1660                 return 1;
1661         else
1662                 return 0;
1663         }
1664
1665 #else
1666 static void *dummy=&dummy;
1667 #endif