andre@0: /* This Source Code Form is subject to the terms of the Mozilla Public andre@0: * License, v. 2.0. If a copy of the MPL was not distributed with this andre@0: * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ andre@0: andre@0: /* A 32-bit implementation of the NIST P-256 elliptic curve. */ andre@0: andre@0: #include andre@0: andre@0: #include "prtypes.h" andre@0: #include "mpi.h" andre@0: #include "mpi-priv.h" andre@0: #include "ecp.h" andre@0: andre@0: typedef PRUint8 u8; andre@0: typedef PRUint32 u32; andre@0: typedef PRUint64 u64; andre@0: andre@0: /* Our field elements are represented as nine, unsigned 32-bit words. Freebl's andre@0: * MPI library calls them digits, but here they are called limbs, which is andre@0: * GMP's terminology. andre@0: * andre@0: * The value of an felem (field element) is: andre@0: * x[0] + (x[1] * 2**29) + (x[2] * 2**57) + ... + (x[8] * 2**228) andre@0: * andre@0: * That is, each limb is alternately 29 or 28-bits wide in little-endian andre@0: * order. andre@0: * andre@0: * This means that an felem hits 2**257, rather than 2**256 as we would like. A andre@0: * 28, 29, ... pattern would cause us to hit 2**256, but that causes problems andre@0: * when multiplying as terms end up one bit short of a limb which would require andre@0: * much bit-shifting to correct. andre@0: * andre@0: * Finally, the values stored in an felem are in Montgomery form. So the value andre@0: * |y| is stored as (y*R) mod p, where p is the P-256 prime and R is 2**257. andre@0: */ andre@0: typedef u32 limb; andre@0: #define NLIMBS 9 andre@0: typedef limb felem[NLIMBS]; andre@0: andre@0: static const limb kBottom28Bits = 0xfffffff; andre@0: static const limb kBottom29Bits = 0x1fffffff; andre@0: andre@0: /* kOne is the number 1 as an felem. It's 2**257 mod p split up into 29 and andre@0: * 28-bit words. andre@0: */ andre@0: static const felem kOne = { andre@0: 2, 0, 0, 0xffff800, andre@0: 0x1fffffff, 0xfffffff, 0x1fbfffff, 0x1ffffff, andre@0: 0 andre@0: }; andre@0: static const felem kZero = {0}; andre@0: static const felem kP = { andre@0: 0x1fffffff, 0xfffffff, 0x1fffffff, 0x3ff, andre@0: 0, 0, 0x200000, 0xf000000, andre@0: 0xfffffff andre@0: }; andre@0: static const felem k2P = { andre@0: 0x1ffffffe, 0xfffffff, 0x1fffffff, 0x7ff, andre@0: 0, 0, 0x400000, 0xe000000, andre@0: 0x1fffffff andre@0: }; andre@0: andre@0: /* kPrecomputed contains precomputed values to aid the calculation of scalar andre@0: * multiples of the base point, G. It's actually two, equal length, tables andre@0: * concatenated. andre@0: * andre@0: * The first table contains (x,y) felem pairs for 16 multiples of the base andre@0: * point, G. andre@0: * andre@0: * Index | Index (binary) | Value andre@0: * 0 | 0000 | 0G (all zeros, omitted) andre@0: * 1 | 0001 | G andre@0: * 2 | 0010 | 2**64G andre@0: * 3 | 0011 | 2**64G + G andre@0: * 4 | 0100 | 2**128G andre@0: * 5 | 0101 | 2**128G + G andre@0: * 6 | 0110 | 2**128G + 2**64G andre@0: * 7 | 0111 | 2**128G + 2**64G + G andre@0: * 8 | 1000 | 2**192G andre@0: * 9 | 1001 | 2**192G + G andre@0: * 10 | 1010 | 2**192G + 2**64G andre@0: * 11 | 1011 | 2**192G + 2**64G + G andre@0: * 12 | 1100 | 2**192G + 2**128G andre@0: * 13 | 1101 | 2**192G + 2**128G + G andre@0: * 14 | 1110 | 2**192G + 2**128G + 2**64G andre@0: * 15 | 1111 | 2**192G + 2**128G + 2**64G + G andre@0: * andre@0: * The second table follows the same style, but the terms are 2**32G, andre@0: * 2**96G, 2**160G, 2**224G. andre@0: * andre@0: * This is ~2KB of data. andre@0: */ andre@0: static const limb kPrecomputed[NLIMBS * 2 * 15 * 2] = { andre@0: 0x11522878, 0xe730d41, 0xdb60179, 0x4afe2ff, 0x12883add, 0xcaddd88, 0x119e7edc, 0xd4a6eab, 0x3120bee, andre@0: 0x1d2aac15, 0xf25357c, 0x19e45cdd, 0x5c721d0, 0x1992c5a5, 0xa237487, 0x154ba21, 0x14b10bb, 0xae3fe3, andre@0: 0xd41a576, 0x922fc51, 0x234994f, 0x60b60d3, 0x164586ae, 0xce95f18, 0x1fe49073, 0x3fa36cc, 0x5ebcd2c, andre@0: 0xb402f2f, 0x15c70bf, 0x1561925c, 0x5a26704, 0xda91e90, 0xcdc1c7f, 0x1ea12446, 0xe1ade1e, 0xec91f22, andre@0: 0x26f7778, 0x566847e, 0xa0bec9e, 0x234f453, 0x1a31f21a, 0xd85e75c, 0x56c7109, 0xa267a00, 0xb57c050, andre@0: 0x98fb57, 0xaa837cc, 0x60c0792, 0xcfa5e19, 0x61bab9e, 0x589e39b, 0xa324c5, 0x7d6dee7, 0x2976e4b, andre@0: 0x1fc4124a, 0xa8c244b, 0x1ce86762, 0xcd61c7e, 0x1831c8e0, 0x75774e1, 0x1d96a5a9, 0x843a649, 0xc3ab0fa, andre@0: 0x6e2e7d5, 0x7673a2a, 0x178b65e8, 0x4003e9b, 0x1a1f11c2, 0x7816ea, 0xf643e11, 0x58c43df, 0xf423fc2, andre@0: 0x19633ffa, 0x891f2b2, 0x123c231c, 0x46add8c, 0x54700dd, 0x59e2b17, 0x172db40f, 0x83e277d, 0xb0dd609, andre@0: 0xfd1da12, 0x35c6e52, 0x19ede20c, 0xd19e0c0, 0x97d0f40, 0xb015b19, 0x449e3f5, 0xe10c9e, 0x33ab581, andre@0: 0x56a67ab, 0x577734d, 0x1dddc062, 0xc57b10d, 0x149b39d, 0x26a9e7b, 0xc35df9f, 0x48764cd, 0x76dbcca, andre@0: 0xca4b366, 0xe9303ab, 0x1a7480e7, 0x57e9e81, 0x1e13eb50, 0xf466cf3, 0x6f16b20, 0x4ba3173, 0xc168c33, andre@0: 0x15cb5439, 0x6a38e11, 0x73658bd, 0xb29564f, 0x3f6dc5b, 0x53b97e, 0x1322c4c0, 0x65dd7ff, 0x3a1e4f6, andre@0: 0x14e614aa, 0x9246317, 0x1bc83aca, 0xad97eed, 0xd38ce4a, 0xf82b006, 0x341f077, 0xa6add89, 0x4894acd, andre@0: 0x9f162d5, 0xf8410ef, 0x1b266a56, 0xd7f223, 0x3e0cb92, 0xe39b672, 0x6a2901a, 0x69a8556, 0x7e7c0, andre@0: 0x9b7d8d3, 0x309a80, 0x1ad05f7f, 0xc2fb5dd, 0xcbfd41d, 0x9ceb638, 0x1051825c, 0xda0cf5b, 0x812e881, andre@0: 0x6f35669, 0x6a56f2c, 0x1df8d184, 0x345820, 0x1477d477, 0x1645db1, 0xbe80c51, 0xc22be3e, 0xe35e65a, andre@0: 0x1aeb7aa0, 0xc375315, 0xf67bc99, 0x7fdd7b9, 0x191fc1be, 0x61235d, 0x2c184e9, 0x1c5a839, 0x47a1e26, andre@0: 0xb7cb456, 0x93e225d, 0x14f3c6ed, 0xccc1ac9, 0x17fe37f3, 0x4988989, 0x1a90c502, 0x2f32042, 0xa17769b, andre@0: 0xafd8c7c, 0x8191c6e, 0x1dcdb237, 0x16200c0, 0x107b32a1, 0x66c08db, 0x10d06a02, 0x3fc93, 0x5620023, andre@0: 0x16722b27, 0x68b5c59, 0x270fcfc, 0xfad0ecc, 0xe5de1c2, 0xeab466b, 0x2fc513c, 0x407f75c, 0xbaab133, andre@0: 0x9705fe9, 0xb88b8e7, 0x734c993, 0x1e1ff8f, 0x19156970, 0xabd0f00, 0x10469ea7, 0x3293ac0, 0xcdc98aa, andre@0: 0x1d843fd, 0xe14bfe8, 0x15be825f, 0x8b5212, 0xeb3fb67, 0x81cbd29, 0xbc62f16, 0x2b6fcc7, 0xf5a4e29, andre@0: 0x13560b66, 0xc0b6ac2, 0x51ae690, 0xd41e271, 0xf3e9bd4, 0x1d70aab, 0x1029f72, 0x73e1c35, 0xee70fbc, andre@0: 0xad81baf, 0x9ecc49a, 0x86c741e, 0xfe6be30, 0x176752e7, 0x23d416, 0x1f83de85, 0x27de188, 0x66f70b8, andre@0: 0x181cd51f, 0x96b6e4c, 0x188f2335, 0xa5df759, 0x17a77eb6, 0xfeb0e73, 0x154ae914, 0x2f3ec51, 0x3826b59, andre@0: 0xb91f17d, 0x1c72949, 0x1362bf0a, 0xe23fddf, 0xa5614b0, 0xf7d8f, 0x79061, 0x823d9d2, 0x8213f39, andre@0: 0x1128ae0b, 0xd095d05, 0xb85c0c2, 0x1ecb2ef, 0x24ddc84, 0xe35e901, 0x18411a4a, 0xf5ddc3d, 0x3786689, andre@0: 0x52260e8, 0x5ae3564, 0x542b10d, 0x8d93a45, 0x19952aa4, 0x996cc41, 0x1051a729, 0x4be3499, 0x52b23aa, andre@0: 0x109f307e, 0x6f5b6bb, 0x1f84e1e7, 0x77a0cfa, 0x10c4df3f, 0x25a02ea, 0xb048035, 0xe31de66, 0xc6ecaa3, andre@0: 0x28ea335, 0x2886024, 0x1372f020, 0xf55d35, 0x15e4684c, 0xf2a9e17, 0x1a4a7529, 0xcb7beb1, 0xb2a78a1, andre@0: 0x1ab21f1f, 0x6361ccf, 0x6c9179d, 0xb135627, 0x1267b974, 0x4408bad, 0x1cbff658, 0xe3d6511, 0xc7d76f, andre@0: 0x1cc7a69, 0xe7ee31b, 0x54fab4f, 0x2b914f, 0x1ad27a30, 0xcd3579e, 0xc50124c, 0x50daa90, 0xb13f72, andre@0: 0xb06aa75, 0x70f5cc6, 0x1649e5aa, 0x84a5312, 0x329043c, 0x41c4011, 0x13d32411, 0xb04a838, 0xd760d2d, andre@0: 0x1713b532, 0xbaa0c03, 0x84022ab, 0x6bcf5c1, 0x2f45379, 0x18ae070, 0x18c9e11e, 0x20bca9a, 0x66f496b, andre@0: 0x3eef294, 0x67500d2, 0xd7f613c, 0x2dbbeb, 0xb741038, 0xe04133f, 0x1582968d, 0xbe985f7, 0x1acbc1a, andre@0: 0x1a6a939f, 0x33e50f6, 0xd665ed4, 0xb4b7bd6, 0x1e5a3799, 0x6b33847, 0x17fa56ff, 0x65ef930, 0x21dc4a, andre@0: 0x2b37659, 0x450fe17, 0xb357b65, 0xdf5efac, 0x15397bef, 0x9d35a7f, 0x112ac15f, 0x624e62e, 0xa90ae2f, andre@0: 0x107eecd2, 0x1f69bbe, 0x77d6bce, 0x5741394, 0x13c684fc, 0x950c910, 0x725522b, 0xdc78583, 0x40eeabb, andre@0: 0x1fde328a, 0xbd61d96, 0xd28c387, 0x9e77d89, 0x12550c40, 0x759cb7d, 0x367ef34, 0xae2a960, 0x91b8bdc, andre@0: 0x93462a9, 0xf469ef, 0xb2e9aef, 0xd2ca771, 0x54e1f42, 0x7aaa49, 0x6316abb, 0x2413c8e, 0x5425bf9, andre@0: 0x1bed3e3a, 0xf272274, 0x1f5e7326, 0x6416517, 0xea27072, 0x9cedea7, 0x6e7633, 0x7c91952, 0xd806dce, andre@0: 0x8e2a7e1, 0xe421e1a, 0x418c9e1, 0x1dbc890, 0x1b395c36, 0xa1dc175, 0x1dc4ef73, 0x8956f34, 0xe4b5cf2, andre@0: 0x1b0d3a18, 0x3194a36, 0x6c2641f, 0xe44124c, 0xa2f4eaa, 0xa8c25ba, 0xf927ed7, 0x627b614, 0x7371cca, andre@0: 0xba16694, 0x417bc03, 0x7c0a7e3, 0x9c35c19, 0x1168a205, 0x8b6b00d, 0x10e3edc9, 0x9c19bf2, 0x5882229, andre@0: 0x1b2b4162, 0xa5cef1a, 0x1543622b, 0x9bd433e, 0x364e04d, 0x7480792, 0x5c9b5b3, 0xe85ff25, 0x408ef57, andre@0: 0x1814cfa4, 0x121b41b, 0xd248a0f, 0x3b05222, 0x39bb16a, 0xc75966d, 0xa038113, 0xa4a1769, 0x11fbc6c, andre@0: 0x917e50e, 0xeec3da8, 0x169d6eac, 0x10c1699, 0xa416153, 0xf724912, 0x15cd60b7, 0x4acbad9, 0x5efc5fa, andre@0: 0xf150ed7, 0x122b51, 0x1104b40a, 0xcb7f442, 0xfbb28ff, 0x6ac53ca, 0x196142cc, 0x7bf0fa9, 0x957651, andre@0: 0x4e0f215, 0xed439f8, 0x3f46bd5, 0x5ace82f, 0x110916b6, 0x6db078, 0xffd7d57, 0xf2ecaac, 0xca86dec, andre@0: 0x15d6b2da, 0x965ecc9, 0x1c92b4c2, 0x1f3811, 0x1cb080f5, 0x2d8b804, 0x19d1c12d, 0xf20bd46, 0x1951fa7, andre@0: 0xa3656c3, 0x523a425, 0xfcd0692, 0xd44ddc8, 0x131f0f5b, 0xaf80e4a, 0xcd9fc74, 0x99bb618, 0x2db944c, andre@0: 0xa673090, 0x1c210e1, 0x178c8d23, 0x1474383, 0x10b8743d, 0x985a55b, 0x2e74779, 0x576138, 0x9587927, andre@0: 0x133130fa, 0xbe05516, 0x9f4d619, 0xbb62570, 0x99ec591, 0xd9468fe, 0x1d07782d, 0xfc72e0b, 0x701b298, andre@0: 0x1863863b, 0x85954b8, 0x121a0c36, 0x9e7fedf, 0xf64b429, 0x9b9d71e, 0x14e2f5d8, 0xf858d3a, 0x942eea8, andre@0: 0xda5b765, 0x6edafff, 0xa9d18cc, 0xc65e4ba, 0x1c747e86, 0xe4ea915, 0x1981d7a1, 0x8395659, 0x52ed4e2, andre@0: 0x87d43b7, 0x37ab11b, 0x19d292ce, 0xf8d4692, 0x18c3053f, 0x8863e13, 0x4c146c0, 0x6bdf55a, 0x4e4457d, andre@0: 0x16152289, 0xac78ec2, 0x1a59c5a2, 0x2028b97, 0x71c2d01, 0x295851f, 0x404747b, 0x878558d, 0x7d29aa4, andre@0: 0x13d8341f, 0x8daefd7, 0x139c972d, 0x6b7ea75, 0xd4a9dde, 0xff163d8, 0x81d55d7, 0xa5bef68, 0xb7b30d8, andre@0: 0xbe73d6f, 0xaa88141, 0xd976c81, 0x7e7a9cc, 0x18beb771, 0xd773cbd, 0x13f51951, 0x9d0c177, 0x1c49a78, andre@0: }; andre@0: andre@0: /* Field element operations: andre@0: */ andre@0: andre@0: /* NON_ZERO_TO_ALL_ONES returns: andre@0: * 0xffffffff for 0 < x <= 2**31 andre@0: * 0 for x == 0 or x > 2**31. andre@0: * andre@0: * x must be a u32 or an equivalent type such as limb. andre@0: */ andre@0: #define NON_ZERO_TO_ALL_ONES(x) ((((u32)(x) - 1) >> 31) - 1) andre@0: andre@0: /* felem_reduce_carry adds a multiple of p in order to cancel |carry|, andre@0: * which is a term at 2**257. andre@0: * andre@0: * On entry: carry < 2**3, inout[0,2,...] < 2**29, inout[1,3,...] < 2**28. andre@0: * On exit: inout[0,2,..] < 2**30, inout[1,3,...] < 2**29. andre@0: */ andre@0: static void felem_reduce_carry(felem inout, limb carry) andre@0: { andre@0: const u32 carry_mask = NON_ZERO_TO_ALL_ONES(carry); andre@0: andre@0: inout[0] += carry << 1; andre@0: inout[3] += 0x10000000 & carry_mask; andre@0: /* carry < 2**3 thus (carry << 11) < 2**14 and we added 2**28 in the andre@0: * previous line therefore this doesn't underflow. andre@0: */ andre@0: inout[3] -= carry << 11; andre@0: inout[4] += (0x20000000 - 1) & carry_mask; andre@0: inout[5] += (0x10000000 - 1) & carry_mask; andre@0: inout[6] += (0x20000000 - 1) & carry_mask; andre@0: inout[6] -= carry << 22; andre@0: /* This may underflow if carry is non-zero but, if so, we'll fix it in the andre@0: * next line. andre@0: */ andre@0: inout[7] -= 1 & carry_mask; andre@0: inout[7] += carry << 25; andre@0: } andre@0: andre@0: /* felem_sum sets out = in+in2. andre@0: * andre@0: * On entry, in[i]+in2[i] must not overflow a 32-bit word. andre@0: * On exit: out[0,2,...] < 2**30, out[1,3,...] < 2**29 andre@0: */ andre@0: static void felem_sum(felem out, const felem in, const felem in2) andre@0: { andre@0: limb carry = 0; andre@0: unsigned int i; andre@0: for (i = 0;; i++) { andre@0: out[i] = in[i] + in2[i]; andre@0: out[i] += carry; andre@0: carry = out[i] >> 29; andre@0: out[i] &= kBottom29Bits; andre@0: andre@0: i++; andre@0: if (i == NLIMBS) andre@0: break; andre@0: andre@0: out[i] = in[i] + in2[i]; andre@0: out[i] += carry; andre@0: carry = out[i] >> 28; andre@0: out[i] &= kBottom28Bits; andre@0: } andre@0: andre@0: felem_reduce_carry(out, carry); andre@0: } andre@0: andre@0: #define two31m3 (((limb)1) << 31) - (((limb)1) << 3) andre@0: #define two30m2 (((limb)1) << 30) - (((limb)1) << 2) andre@0: #define two30p13m2 (((limb)1) << 30) + (((limb)1) << 13) - (((limb)1) << 2) andre@0: #define two31m2 (((limb)1) << 31) - (((limb)1) << 2) andre@0: #define two31p24m2 (((limb)1) << 31) + (((limb)1) << 24) - (((limb)1) << 2) andre@0: #define two30m27m2 (((limb)1) << 30) - (((limb)1) << 27) - (((limb)1) << 2) andre@0: andre@0: /* zero31 is 0 mod p. andre@0: */ andre@0: static const felem zero31 = { andre@0: two31m3, two30m2, two31m2, two30p13m2, andre@0: two31m2, two30m2, two31p24m2, two30m27m2, andre@0: two31m2 andre@0: }; andre@0: andre@0: /* felem_diff sets out = in-in2. andre@0: * andre@0: * On entry: in[0,2,...] < 2**30, in[1,3,...] < 2**29 and andre@0: * in2[0,2,...] < 2**30, in2[1,3,...] < 2**29. andre@0: * On exit: out[0,2,...] < 2**30, out[1,3,...] < 2**29. andre@0: */ andre@0: static void felem_diff(felem out, const felem in, const felem in2) andre@0: { andre@0: limb carry = 0; andre@0: unsigned int i; andre@0: andre@0: for (i = 0;; i++) { andre@0: out[i] = in[i] - in2[i]; andre@0: out[i] += zero31[i]; andre@0: out[i] += carry; andre@0: carry = out[i] >> 29; andre@0: out[i] &= kBottom29Bits; andre@0: andre@0: i++; andre@0: if (i == NLIMBS) andre@0: break; andre@0: andre@0: out[i] = in[i] - in2[i]; andre@0: out[i] += zero31[i]; andre@0: out[i] += carry; andre@0: carry = out[i] >> 28; andre@0: out[i] &= kBottom28Bits; andre@0: } andre@0: andre@0: felem_reduce_carry(out, carry); andre@0: } andre@0: andre@0: /* felem_reduce_degree sets out = tmp/R mod p where tmp contains 64-bit words andre@0: * with the same 29,28,... bit positions as an felem. andre@0: * andre@0: * The values in felems are in Montgomery form: x*R mod p where R = 2**257. andre@0: * Since we just multiplied two Montgomery values together, the result is andre@0: * x*y*R*R mod p. We wish to divide by R in order for the result also to be andre@0: * in Montgomery form. andre@0: * andre@0: * On entry: tmp[i] < 2**64 andre@0: * On exit: out[0,2,...] < 2**30, out[1,3,...] < 2**29 andre@0: */ andre@0: static void felem_reduce_degree(felem out, u64 tmp[17]) andre@0: { andre@0: /* The following table may be helpful when reading this code: andre@0: * andre@0: * Limb number: 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10... andre@0: * Width (bits): 29| 28| 29| 28| 29| 28| 29| 28| 29| 28| 29 andre@0: * Start bit: 0 | 29| 57| 86|114|143|171|200|228|257|285 andre@0: * (odd phase): 0 | 28| 57| 85|114|142|171|199|228|256|285 andre@0: */ andre@0: limb tmp2[18], carry, x, xMask; andre@0: unsigned int i; andre@0: andre@0: /* tmp contains 64-bit words with the same 29,28,29-bit positions as an andre@0: * felem. So the top of an element of tmp might overlap with another andre@0: * element two positions down. The following loop eliminates this andre@0: * overlap. andre@0: */ andre@0: tmp2[0] = tmp[0] & kBottom29Bits; andre@0: andre@0: /* In the following we use "(limb) tmp[x]" and "(limb) (tmp[x]>>32)" to try andre@0: * and hint to the compiler that it can do a single-word shift by selecting andre@0: * the right register rather than doing a double-word shift and truncating andre@0: * afterwards. andre@0: */ andre@0: tmp2[1] = ((limb) tmp[0]) >> 29; andre@0: tmp2[1] |= (((limb) (tmp[0] >> 32)) << 3) & kBottom28Bits; andre@0: tmp2[1] += ((limb) tmp[1]) & kBottom28Bits; andre@0: carry = tmp2[1] >> 28; andre@0: tmp2[1] &= kBottom28Bits; andre@0: andre@0: for (i = 2; i < 17; i++) { andre@0: tmp2[i] = ((limb) (tmp[i - 2] >> 32)) >> 25; andre@0: tmp2[i] += ((limb) (tmp[i - 1])) >> 28; andre@0: tmp2[i] += (((limb) (tmp[i - 1] >> 32)) << 4) & kBottom29Bits; andre@0: tmp2[i] += ((limb) tmp[i]) & kBottom29Bits; andre@0: tmp2[i] += carry; andre@0: carry = tmp2[i] >> 29; andre@0: tmp2[i] &= kBottom29Bits; andre@0: andre@0: i++; andre@0: if (i == 17) andre@0: break; andre@0: tmp2[i] = ((limb) (tmp[i - 2] >> 32)) >> 25; andre@0: tmp2[i] += ((limb) (tmp[i - 1])) >> 29; andre@0: tmp2[i] += (((limb) (tmp[i - 1] >> 32)) << 3) & kBottom28Bits; andre@0: tmp2[i] += ((limb) tmp[i]) & kBottom28Bits; andre@0: tmp2[i] += carry; andre@0: carry = tmp2[i] >> 28; andre@0: tmp2[i] &= kBottom28Bits; andre@0: } andre@0: andre@0: tmp2[17] = ((limb) (tmp[15] >> 32)) >> 25; andre@0: tmp2[17] += ((limb) (tmp[16])) >> 29; andre@0: tmp2[17] += (((limb) (tmp[16] >> 32)) << 3); andre@0: tmp2[17] += carry; andre@0: andre@0: /* Montgomery elimination of terms: andre@0: * andre@0: * Since R is 2**257, we can divide by R with a bitwise shift if we can andre@0: * ensure that the right-most 257 bits are all zero. We can make that true andre@0: * by adding multiplies of p without affecting the value. andre@0: * andre@0: * So we eliminate limbs from right to left. Since the bottom 29 bits of p andre@0: * are all ones, then by adding tmp2[0]*p to tmp2 we'll make tmp2[0] == 0. andre@0: * We can do that for 8 further limbs and then right shift to eliminate the andre@0: * extra factor of R. andre@0: */ andre@0: for (i = 0;; i += 2) { andre@0: tmp2[i + 1] += tmp2[i] >> 29; andre@0: x = tmp2[i] & kBottom29Bits; andre@0: xMask = NON_ZERO_TO_ALL_ONES(x); andre@0: tmp2[i] = 0; andre@0: andre@0: /* The bounds calculations for this loop are tricky. Each iteration of andre@0: * the loop eliminates two words by adding values to words to their andre@0: * right. andre@0: * andre@0: * The following table contains the amounts added to each word (as an andre@0: * offset from the value of i at the top of the loop). The amounts are andre@0: * accounted for from the first and second half of the loop separately andre@0: * and are written as, for example, 28 to mean a value <2**28. andre@0: * andre@0: * Word: 3 4 5 6 7 8 9 10 andre@0: * Added in top half: 28 11 29 21 29 28 andre@0: * 28 29 andre@0: * 29 andre@0: * Added in bottom half: 29 10 28 21 28 28 andre@0: * 29 andre@0: * andre@0: * The value that is currently offset 7 will be offset 5 for the next andre@0: * iteration and then offset 3 for the iteration after that. Therefore andre@0: * the total value added will be the values added at 7, 5 and 3. andre@0: * andre@0: * The following table accumulates these values. The sums at the bottom andre@0: * are written as, for example, 29+28, to mean a value < 2**29+2**28. andre@0: * andre@0: * Word: 3 4 5 6 7 8 9 10 11 12 13 andre@0: * 28 11 10 29 21 29 28 28 28 28 28 andre@0: * 29 28 11 28 29 28 29 28 29 28 andre@0: * 29 28 21 21 29 21 29 21 andre@0: * 10 29 28 21 28 21 28 andre@0: * 28 29 28 29 28 29 28 andre@0: * 11 10 29 10 29 10 andre@0: * 29 28 11 28 11 andre@0: * 29 29 andre@0: * -------------------------------------------- andre@0: * 30+ 31+ 30+ 31+ 30+ andre@0: * 28+ 29+ 28+ 29+ 21+ andre@0: * 21+ 28+ 21+ 28+ 10 andre@0: * 10 21+ 10 21+ andre@0: * 11 11 andre@0: * andre@0: * So the greatest amount is added to tmp2[10] and tmp2[12]. If andre@0: * tmp2[10/12] has an initial value of <2**29, then the maximum value andre@0: * will be < 2**31 + 2**30 + 2**28 + 2**21 + 2**11, which is < 2**32, andre@0: * as required. andre@0: */ andre@0: tmp2[i + 3] += (x << 10) & kBottom28Bits; andre@0: tmp2[i + 4] += (x >> 18); andre@0: andre@0: tmp2[i + 6] += (x << 21) & kBottom29Bits; andre@0: tmp2[i + 7] += x >> 8; andre@0: andre@0: /* At position 200, which is the starting bit position for word 7, we andre@0: * have a factor of 0xf000000 = 2**28 - 2**24. andre@0: */ andre@0: tmp2[i + 7] += 0x10000000 & xMask; andre@0: /* Word 7 is 28 bits wide, so the 2**28 term exactly hits word 8. */ andre@0: tmp2[i + 8] += (x - 1) & xMask; andre@0: tmp2[i + 7] -= (x << 24) & kBottom28Bits; andre@0: tmp2[i + 8] -= x >> 4; andre@0: andre@0: tmp2[i + 8] += 0x20000000 & xMask; andre@0: tmp2[i + 8] -= x; andre@0: tmp2[i + 8] += (x << 28) & kBottom29Bits; andre@0: tmp2[i + 9] += ((x >> 1) - 1) & xMask; andre@0: andre@0: if (i+1 == NLIMBS) andre@0: break; andre@0: tmp2[i + 2] += tmp2[i + 1] >> 28; andre@0: x = tmp2[i + 1] & kBottom28Bits; andre@0: xMask = NON_ZERO_TO_ALL_ONES(x); andre@0: tmp2[i + 1] = 0; andre@0: andre@0: tmp2[i + 4] += (x << 11) & kBottom29Bits; andre@0: tmp2[i + 5] += (x >> 18); andre@0: andre@0: tmp2[i + 7] += (x << 21) & kBottom28Bits; andre@0: tmp2[i + 8] += x >> 7; andre@0: andre@0: /* At position 199, which is the starting bit of the 8th word when andre@0: * dealing with a context starting on an odd word, we have a factor of andre@0: * 0x1e000000 = 2**29 - 2**25. Since we have not updated i, the 8th andre@0: * word from i+1 is i+8. andre@0: */ andre@0: tmp2[i + 8] += 0x20000000 & xMask; andre@0: tmp2[i + 9] += (x - 1) & xMask; andre@0: tmp2[i + 8] -= (x << 25) & kBottom29Bits; andre@0: tmp2[i + 9] -= x >> 4; andre@0: andre@0: tmp2[i + 9] += 0x10000000 & xMask; andre@0: tmp2[i + 9] -= x; andre@0: tmp2[i + 10] += (x - 1) & xMask; andre@0: } andre@0: andre@0: /* We merge the right shift with a carry chain. The words above 2**257 have andre@0: * widths of 28,29,... which we need to correct when copying them down. andre@0: */ andre@0: carry = 0; andre@0: for (i = 0; i < 8; i++) { andre@0: /* The maximum value of tmp2[i + 9] occurs on the first iteration and andre@0: * is < 2**30+2**29+2**28. Adding 2**29 (from tmp2[i + 10]) is andre@0: * therefore safe. andre@0: */ andre@0: out[i] = tmp2[i + 9]; andre@0: out[i] += carry; andre@0: out[i] += (tmp2[i + 10] << 28) & kBottom29Bits; andre@0: carry = out[i] >> 29; andre@0: out[i] &= kBottom29Bits; andre@0: andre@0: i++; andre@0: out[i] = tmp2[i + 9] >> 1; andre@0: out[i] += carry; andre@0: carry = out[i] >> 28; andre@0: out[i] &= kBottom28Bits; andre@0: } andre@0: andre@0: out[8] = tmp2[17]; andre@0: out[8] += carry; andre@0: carry = out[8] >> 29; andre@0: out[8] &= kBottom29Bits; andre@0: andre@0: felem_reduce_carry(out, carry); andre@0: } andre@0: andre@0: /* felem_square sets out=in*in. andre@0: * andre@0: * On entry: in[0,2,...] < 2**30, in[1,3,...] < 2**29. andre@0: * On exit: out[0,2,...] < 2**30, out[1,3,...] < 2**29. andre@0: */ andre@0: static void felem_square(felem out, const felem in) andre@0: { andre@0: u64 tmp[17]; andre@0: andre@0: tmp[0] = ((u64) in[0]) * in[0]; andre@0: tmp[1] = ((u64) in[0]) * (in[1] << 1); andre@0: tmp[2] = ((u64) in[0]) * (in[2] << 1) + andre@0: ((u64) in[1]) * (in[1] << 1); andre@0: tmp[3] = ((u64) in[0]) * (in[3] << 1) + andre@0: ((u64) in[1]) * (in[2] << 1); andre@0: tmp[4] = ((u64) in[0]) * (in[4] << 1) + andre@0: ((u64) in[1]) * (in[3] << 2) + andre@0: ((u64) in[2]) * in[2]; andre@0: tmp[5] = ((u64) in[0]) * (in[5] << 1) + andre@0: ((u64) in[1]) * (in[4] << 1) + andre@0: ((u64) in[2]) * (in[3] << 1); andre@0: tmp[6] = ((u64) in[0]) * (in[6] << 1) + andre@0: ((u64) in[1]) * (in[5] << 2) + andre@0: ((u64) in[2]) * (in[4] << 1) + andre@0: ((u64) in[3]) * (in[3] << 1); andre@0: tmp[7] = ((u64) in[0]) * (in[7] << 1) + andre@0: ((u64) in[1]) * (in[6] << 1) + andre@0: ((u64) in[2]) * (in[5] << 1) + andre@0: ((u64) in[3]) * (in[4] << 1); andre@0: /* tmp[8] has the greatest value of 2**61 + 2**60 + 2**61 + 2**60 + 2**60, andre@0: * which is < 2**64 as required. andre@0: */ andre@0: tmp[8] = ((u64) in[0]) * (in[8] << 1) + andre@0: ((u64) in[1]) * (in[7] << 2) + andre@0: ((u64) in[2]) * (in[6] << 1) + andre@0: ((u64) in[3]) * (in[5] << 2) + andre@0: ((u64) in[4]) * in[4]; andre@0: tmp[9] = ((u64) in[1]) * (in[8] << 1) + andre@0: ((u64) in[2]) * (in[7] << 1) + andre@0: ((u64) in[3]) * (in[6] << 1) + andre@0: ((u64) in[4]) * (in[5] << 1); andre@0: tmp[10] = ((u64) in[2]) * (in[8] << 1) + andre@0: ((u64) in[3]) * (in[7] << 2) + andre@0: ((u64) in[4]) * (in[6] << 1) + andre@0: ((u64) in[5]) * (in[5] << 1); andre@0: tmp[11] = ((u64) in[3]) * (in[8] << 1) + andre@0: ((u64) in[4]) * (in[7] << 1) + andre@0: ((u64) in[5]) * (in[6] << 1); andre@0: tmp[12] = ((u64) in[4]) * (in[8] << 1) + andre@0: ((u64) in[5]) * (in[7] << 2) + andre@0: ((u64) in[6]) * in[6]; andre@0: tmp[13] = ((u64) in[5]) * (in[8] << 1) + andre@0: ((u64) in[6]) * (in[7] << 1); andre@0: tmp[14] = ((u64) in[6]) * (in[8] << 1) + andre@0: ((u64) in[7]) * (in[7] << 1); andre@0: tmp[15] = ((u64) in[7]) * (in[8] << 1); andre@0: tmp[16] = ((u64) in[8]) * in[8]; andre@0: andre@0: felem_reduce_degree(out, tmp); andre@0: } andre@0: andre@0: /* felem_mul sets out=in*in2. andre@0: * andre@0: * On entry: in[0,2,...] < 2**30, in[1,3,...] < 2**29 and andre@0: * in2[0,2,...] < 2**30, in2[1,3,...] < 2**29. andre@0: * On exit: out[0,2,...] < 2**30, out[1,3,...] < 2**29. andre@0: */ andre@0: static void felem_mul(felem out, const felem in, const felem in2) andre@0: { andre@0: u64 tmp[17]; andre@0: andre@0: tmp[0] = ((u64) in[0]) * in2[0]; andre@0: tmp[1] = ((u64) in[0]) * (in2[1] << 0) + andre@0: ((u64) in[1]) * (in2[0] << 0); andre@0: tmp[2] = ((u64) in[0]) * (in2[2] << 0) + andre@0: ((u64) in[1]) * (in2[1] << 1) + andre@0: ((u64) in[2]) * (in2[0] << 0); andre@0: tmp[3] = ((u64) in[0]) * (in2[3] << 0) + andre@0: ((u64) in[1]) * (in2[2] << 0) + andre@0: ((u64) in[2]) * (in2[1] << 0) + andre@0: ((u64) in[3]) * (in2[0] << 0); andre@0: tmp[4] = ((u64) in[0]) * (in2[4] << 0) + andre@0: ((u64) in[1]) * (in2[3] << 1) + andre@0: ((u64) in[2]) * (in2[2] << 0) + andre@0: ((u64) in[3]) * (in2[1] << 1) + andre@0: ((u64) in[4]) * (in2[0] << 0); andre@0: tmp[5] = ((u64) in[0]) * (in2[5] << 0) + andre@0: ((u64) in[1]) * (in2[4] << 0) + andre@0: ((u64) in[2]) * (in2[3] << 0) + andre@0: ((u64) in[3]) * (in2[2] << 0) + andre@0: ((u64) in[4]) * (in2[1] << 0) + andre@0: ((u64) in[5]) * (in2[0] << 0); andre@0: tmp[6] = ((u64) in[0]) * (in2[6] << 0) + andre@0: ((u64) in[1]) * (in2[5] << 1) + andre@0: ((u64) in[2]) * (in2[4] << 0) + andre@0: ((u64) in[3]) * (in2[3] << 1) + andre@0: ((u64) in[4]) * (in2[2] << 0) + andre@0: ((u64) in[5]) * (in2[1] << 1) + andre@0: ((u64) in[6]) * (in2[0] << 0); andre@0: tmp[7] = ((u64) in[0]) * (in2[7] << 0) + andre@0: ((u64) in[1]) * (in2[6] << 0) + andre@0: ((u64) in[2]) * (in2[5] << 0) + andre@0: ((u64) in[3]) * (in2[4] << 0) + andre@0: ((u64) in[4]) * (in2[3] << 0) + andre@0: ((u64) in[5]) * (in2[2] << 0) + andre@0: ((u64) in[6]) * (in2[1] << 0) + andre@0: ((u64) in[7]) * (in2[0] << 0); andre@0: /* tmp[8] has the greatest value but doesn't overflow. See logic in andre@0: * felem_square. andre@0: */ andre@0: tmp[8] = ((u64) in[0]) * (in2[8] << 0) + andre@0: ((u64) in[1]) * (in2[7] << 1) + andre@0: ((u64) in[2]) * (in2[6] << 0) + andre@0: ((u64) in[3]) * (in2[5] << 1) + andre@0: ((u64) in[4]) * (in2[4] << 0) + andre@0: ((u64) in[5]) * (in2[3] << 1) + andre@0: ((u64) in[6]) * (in2[2] << 0) + andre@0: ((u64) in[7]) * (in2[1] << 1) + andre@0: ((u64) in[8]) * (in2[0] << 0); andre@0: tmp[9] = ((u64) in[1]) * (in2[8] << 0) + andre@0: ((u64) in[2]) * (in2[7] << 0) + andre@0: ((u64) in[3]) * (in2[6] << 0) + andre@0: ((u64) in[4]) * (in2[5] << 0) + andre@0: ((u64) in[5]) * (in2[4] << 0) + andre@0: ((u64) in[6]) * (in2[3] << 0) + andre@0: ((u64) in[7]) * (in2[2] << 0) + andre@0: ((u64) in[8]) * (in2[1] << 0); andre@0: tmp[10] = ((u64) in[2]) * (in2[8] << 0) + andre@0: ((u64) in[3]) * (in2[7] << 1) + andre@0: ((u64) in[4]) * (in2[6] << 0) + andre@0: ((u64) in[5]) * (in2[5] << 1) + andre@0: ((u64) in[6]) * (in2[4] << 0) + andre@0: ((u64) in[7]) * (in2[3] << 1) + andre@0: ((u64) in[8]) * (in2[2] << 0); andre@0: tmp[11] = ((u64) in[3]) * (in2[8] << 0) + andre@0: ((u64) in[4]) * (in2[7] << 0) + andre@0: ((u64) in[5]) * (in2[6] << 0) + andre@0: ((u64) in[6]) * (in2[5] << 0) + andre@0: ((u64) in[7]) * (in2[4] << 0) + andre@0: ((u64) in[8]) * (in2[3] << 0); andre@0: tmp[12] = ((u64) in[4]) * (in2[8] << 0) + andre@0: ((u64) in[5]) * (in2[7] << 1) + andre@0: ((u64) in[6]) * (in2[6] << 0) + andre@0: ((u64) in[7]) * (in2[5] << 1) + andre@0: ((u64) in[8]) * (in2[4] << 0); andre@0: tmp[13] = ((u64) in[5]) * (in2[8] << 0) + andre@0: ((u64) in[6]) * (in2[7] << 0) + andre@0: ((u64) in[7]) * (in2[6] << 0) + andre@0: ((u64) in[8]) * (in2[5] << 0); andre@0: tmp[14] = ((u64) in[6]) * (in2[8] << 0) + andre@0: ((u64) in[7]) * (in2[7] << 1) + andre@0: ((u64) in[8]) * (in2[6] << 0); andre@0: tmp[15] = ((u64) in[7]) * (in2[8] << 0) + andre@0: ((u64) in[8]) * (in2[7] << 0); andre@0: tmp[16] = ((u64) in[8]) * (in2[8] << 0); andre@0: andre@0: felem_reduce_degree(out, tmp); andre@0: } andre@0: andre@0: static void felem_assign(felem out, const felem in) andre@0: { andre@0: memcpy(out, in, sizeof(felem)); andre@0: } andre@0: andre@0: /* felem_inv calculates |out| = |in|^{-1} andre@0: * andre@0: * Based on Fermat's Little Theorem: andre@0: * a^p = a (mod p) andre@0: * a^{p-1} = 1 (mod p) andre@0: * a^{p-2} = a^{-1} (mod p) andre@0: */ andre@0: static void felem_inv(felem out, const felem in) andre@0: { andre@0: felem ftmp, ftmp2; andre@0: /* each e_I will hold |in|^{2^I - 1} */ andre@0: felem e2, e4, e8, e16, e32, e64; andre@0: unsigned int i; andre@0: andre@0: felem_square(ftmp, in); /* 2^1 */ andre@0: felem_mul(ftmp, in, ftmp); /* 2^2 - 2^0 */ andre@0: felem_assign(e2, ftmp); andre@0: felem_square(ftmp, ftmp); /* 2^3 - 2^1 */ andre@0: felem_square(ftmp, ftmp); /* 2^4 - 2^2 */ andre@0: felem_mul(ftmp, ftmp, e2); /* 2^4 - 2^0 */ andre@0: felem_assign(e4, ftmp); andre@0: felem_square(ftmp, ftmp); /* 2^5 - 2^1 */ andre@0: felem_square(ftmp, ftmp); /* 2^6 - 2^2 */ andre@0: felem_square(ftmp, ftmp); /* 2^7 - 2^3 */ andre@0: felem_square(ftmp, ftmp); /* 2^8 - 2^4 */ andre@0: felem_mul(ftmp, ftmp, e4); /* 2^8 - 2^0 */ andre@0: felem_assign(e8, ftmp); andre@0: for (i = 0; i < 8; i++) { andre@0: felem_square(ftmp, ftmp); andre@0: } /* 2^16 - 2^8 */ andre@0: felem_mul(ftmp, ftmp, e8); /* 2^16 - 2^0 */ andre@0: felem_assign(e16, ftmp); andre@0: for (i = 0; i < 16; i++) { andre@0: felem_square(ftmp, ftmp); andre@0: } /* 2^32 - 2^16 */ andre@0: felem_mul(ftmp, ftmp, e16); /* 2^32 - 2^0 */ andre@0: felem_assign(e32, ftmp); andre@0: for (i = 0; i < 32; i++) { andre@0: felem_square(ftmp, ftmp); andre@0: } /* 2^64 - 2^32 */ andre@0: felem_assign(e64, ftmp); andre@0: felem_mul(ftmp, ftmp, in); /* 2^64 - 2^32 + 2^0 */ andre@0: for (i = 0; i < 192; i++) { andre@0: felem_square(ftmp, ftmp); andre@0: } /* 2^256 - 2^224 + 2^192 */ andre@0: andre@0: felem_mul(ftmp2, e64, e32); /* 2^64 - 2^0 */ andre@0: for (i = 0; i < 16; i++) { andre@0: felem_square(ftmp2, ftmp2); andre@0: } /* 2^80 - 2^16 */ andre@0: felem_mul(ftmp2, ftmp2, e16); /* 2^80 - 2^0 */ andre@0: for (i = 0; i < 8; i++) { andre@0: felem_square(ftmp2, ftmp2); andre@0: } /* 2^88 - 2^8 */ andre@0: felem_mul(ftmp2, ftmp2, e8); /* 2^88 - 2^0 */ andre@0: for (i = 0; i < 4; i++) { andre@0: felem_square(ftmp2, ftmp2); andre@0: } /* 2^92 - 2^4 */ andre@0: felem_mul(ftmp2, ftmp2, e4); /* 2^92 - 2^0 */ andre@0: felem_square(ftmp2, ftmp2); /* 2^93 - 2^1 */ andre@0: felem_square(ftmp2, ftmp2); /* 2^94 - 2^2 */ andre@0: felem_mul(ftmp2, ftmp2, e2); /* 2^94 - 2^0 */ andre@0: felem_square(ftmp2, ftmp2); /* 2^95 - 2^1 */ andre@0: felem_square(ftmp2, ftmp2); /* 2^96 - 2^2 */ andre@0: felem_mul(ftmp2, ftmp2, in); /* 2^96 - 3 */ andre@0: andre@0: felem_mul(out, ftmp2, ftmp); /* 2^256 - 2^224 + 2^192 + 2^96 - 3 */ andre@0: } andre@0: andre@0: /* felem_scalar_3 sets out=3*out. andre@0: * andre@0: * On entry: out[0,2,...] < 2**30, out[1,3,...] < 2**29. andre@0: * On exit: out[0,2,...] < 2**30, out[1,3,...] < 2**29. andre@0: */ andre@0: static void felem_scalar_3(felem out) andre@0: { andre@0: limb carry = 0; andre@0: unsigned int i; andre@0: andre@0: for (i = 0;; i++) { andre@0: out[i] *= 3; andre@0: out[i] += carry; andre@0: carry = out[i] >> 29; andre@0: out[i] &= kBottom29Bits; andre@0: andre@0: i++; andre@0: if (i == NLIMBS) andre@0: break; andre@0: andre@0: out[i] *= 3; andre@0: out[i] += carry; andre@0: carry = out[i] >> 28; andre@0: out[i] &= kBottom28Bits; andre@0: } andre@0: andre@0: felem_reduce_carry(out, carry); andre@0: } andre@0: andre@0: /* felem_scalar_4 sets out=4*out. andre@0: * andre@0: * On entry: out[0,2,...] < 2**30, out[1,3,...] < 2**29. andre@0: * On exit: out[0,2,...] < 2**30, out[1,3,...] < 2**29. andre@0: */ andre@0: static void felem_scalar_4(felem out) andre@0: { andre@0: limb carry = 0, next_carry; andre@0: unsigned int i; andre@0: andre@0: for (i = 0;; i++) { andre@0: next_carry = out[i] >> 27; andre@0: out[i] <<= 2; andre@0: out[i] &= kBottom29Bits; andre@0: out[i] += carry; andre@0: carry = next_carry + (out[i] >> 29); andre@0: out[i] &= kBottom29Bits; andre@0: andre@0: i++; andre@0: if (i == NLIMBS) andre@0: break; andre@0: next_carry = out[i] >> 26; andre@0: out[i] <<= 2; andre@0: out[i] &= kBottom28Bits; andre@0: out[i] += carry; andre@0: carry = next_carry + (out[i] >> 28); andre@0: out[i] &= kBottom28Bits; andre@0: } andre@0: andre@0: felem_reduce_carry(out, carry); andre@0: } andre@0: andre@0: /* felem_scalar_8 sets out=8*out. andre@0: * andre@0: * On entry: out[0,2,...] < 2**30, out[1,3,...] < 2**29. andre@0: * On exit: out[0,2,...] < 2**30, out[1,3,...] < 2**29. andre@0: */ andre@0: static void felem_scalar_8(felem out) andre@0: { andre@0: limb carry = 0, next_carry; andre@0: unsigned int i; andre@0: andre@0: for (i = 0;; i++) { andre@0: next_carry = out[i] >> 26; andre@0: out[i] <<= 3; andre@0: out[i] &= kBottom29Bits; andre@0: out[i] += carry; andre@0: carry = next_carry + (out[i] >> 29); andre@0: out[i] &= kBottom29Bits; andre@0: andre@0: i++; andre@0: if (i == NLIMBS) andre@0: break; andre@0: next_carry = out[i] >> 25; andre@0: out[i] <<= 3; andre@0: out[i] &= kBottom28Bits; andre@0: out[i] += carry; andre@0: carry = next_carry + (out[i] >> 28); andre@0: out[i] &= kBottom28Bits; andre@0: } andre@0: andre@0: felem_reduce_carry(out, carry); andre@0: } andre@0: andre@0: /* felem_is_zero_vartime returns 1 iff |in| == 0. It takes a variable amount of andre@0: * time depending on the value of |in|. andre@0: */ andre@0: static char felem_is_zero_vartime(const felem in) andre@0: { andre@0: limb carry; andre@0: int i; andre@0: limb tmp[NLIMBS]; andre@0: felem_assign(tmp, in); andre@0: andre@0: /* First, reduce tmp to a minimal form. andre@0: */ andre@0: do { andre@0: carry = 0; andre@0: for (i = 0;; i++) { andre@0: tmp[i] += carry; andre@0: carry = tmp[i] >> 29; andre@0: tmp[i] &= kBottom29Bits; andre@0: andre@0: i++; andre@0: if (i == NLIMBS) andre@0: break; andre@0: andre@0: tmp[i] += carry; andre@0: carry = tmp[i] >> 28; andre@0: tmp[i] &= kBottom28Bits; andre@0: } andre@0: andre@0: felem_reduce_carry(tmp, carry); andre@0: } while (carry); andre@0: andre@0: /* tmp < 2**257, so the only possible zero values are 0, p and 2p. andre@0: */ andre@0: return memcmp(tmp, kZero, sizeof(tmp)) == 0 || andre@0: memcmp(tmp, kP, sizeof(tmp)) == 0 || andre@0: memcmp(tmp, k2P, sizeof(tmp)) == 0; andre@0: } andre@0: andre@0: /* Group operations: andre@0: * andre@0: * Elements of the elliptic curve group are represented in Jacobian andre@0: * coordinates: (x, y, z). An affine point (x', y') is x'=x/z**2, y'=y/z**3 in andre@0: * Jacobian form. andre@0: */ andre@0: andre@0: /* point_double sets {x_out,y_out,z_out} = 2*{x,y,z}. andre@0: * andre@0: * See http://www.hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-0.html#doubling-dbl-2009-l andre@0: */ andre@0: static void point_double(felem x_out, felem y_out, felem z_out, andre@0: const felem x, const felem y, const felem z) andre@0: { andre@0: felem delta, gamma, alpha, beta, tmp, tmp2; andre@0: andre@0: felem_square(delta, z); andre@0: felem_square(gamma, y); andre@0: felem_mul(beta, x, gamma); andre@0: andre@0: felem_sum(tmp, x, delta); andre@0: felem_diff(tmp2, x, delta); andre@0: felem_mul(alpha, tmp, tmp2); andre@0: felem_scalar_3(alpha); andre@0: andre@0: felem_sum(tmp, y, z); andre@0: felem_square(tmp, tmp); andre@0: felem_diff(tmp, tmp, gamma); andre@0: felem_diff(z_out, tmp, delta); andre@0: andre@0: felem_scalar_4(beta); andre@0: felem_square(x_out, alpha); andre@0: felem_diff(x_out, x_out, beta); andre@0: felem_diff(x_out, x_out, beta); andre@0: andre@0: felem_diff(tmp, beta, x_out); andre@0: felem_mul(tmp, alpha, tmp); andre@0: felem_square(tmp2, gamma); andre@0: felem_scalar_8(tmp2); andre@0: felem_diff(y_out, tmp, tmp2); andre@0: } andre@0: andre@0: /* point_add_mixed sets {x_out,y_out,z_out} = {x1,y1,z1} + {x2,y2,1}. andre@0: * (i.e. the second point is affine.) andre@0: * andre@0: * See http://www.hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-0.html#addition-add-2007-bl andre@0: * andre@0: * Note that this function does not handle P+P, infinity+P nor P+infinity andre@0: * correctly. andre@0: */ andre@0: static void point_add_mixed(felem x_out, felem y_out, felem z_out, andre@0: const felem x1, const felem y1, const felem z1, andre@0: const felem x2, const felem y2) andre@0: { andre@0: felem z1z1, z1z1z1, s2, u2, h, i, j, r, rr, v, tmp; andre@0: andre@0: felem_square(z1z1, z1); andre@0: felem_sum(tmp, z1, z1); andre@0: andre@0: felem_mul(u2, x2, z1z1); andre@0: felem_mul(z1z1z1, z1, z1z1); andre@0: felem_mul(s2, y2, z1z1z1); andre@0: felem_diff(h, u2, x1); andre@0: felem_sum(i, h, h); andre@0: felem_square(i, i); andre@0: felem_mul(j, h, i); andre@0: felem_diff(r, s2, y1); andre@0: felem_sum(r, r, r); andre@0: felem_mul(v, x1, i); andre@0: andre@0: felem_mul(z_out, tmp, h); andre@0: felem_square(rr, r); andre@0: felem_diff(x_out, rr, j); andre@0: felem_diff(x_out, x_out, v); andre@0: felem_diff(x_out, x_out, v); andre@0: andre@0: felem_diff(tmp, v, x_out); andre@0: felem_mul(y_out, tmp, r); andre@0: felem_mul(tmp, y1, j); andre@0: felem_diff(y_out, y_out, tmp); andre@0: felem_diff(y_out, y_out, tmp); andre@0: } andre@0: andre@0: /* point_add sets {x_out,y_out,z_out} = {x1,y1,z1} + {x2,y2,z2}. andre@0: * andre@0: * See http://www.hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-0.html#addition-add-2007-bl andre@0: * andre@0: * Note that this function does not handle P+P, infinity+P nor P+infinity andre@0: * correctly. andre@0: */ andre@0: static void point_add(felem x_out, felem y_out, felem z_out, andre@0: const felem x1, const felem y1, const felem z1, andre@0: const felem x2, const felem y2, const felem z2) andre@0: { andre@0: felem z1z1, z1z1z1, z2z2, z2z2z2, s1, s2, u1, u2, h, i, j, r, rr, v, tmp; andre@0: andre@0: felem_square(z1z1, z1); andre@0: felem_square(z2z2, z2); andre@0: felem_mul(u1, x1, z2z2); andre@0: andre@0: felem_sum(tmp, z1, z2); andre@0: felem_square(tmp, tmp); andre@0: felem_diff(tmp, tmp, z1z1); andre@0: felem_diff(tmp, tmp, z2z2); andre@0: andre@0: felem_mul(z2z2z2, z2, z2z2); andre@0: felem_mul(s1, y1, z2z2z2); andre@0: andre@0: felem_mul(u2, x2, z1z1); andre@0: felem_mul(z1z1z1, z1, z1z1); andre@0: felem_mul(s2, y2, z1z1z1); andre@0: felem_diff(h, u2, u1); andre@0: felem_sum(i, h, h); andre@0: felem_square(i, i); andre@0: felem_mul(j, h, i); andre@0: felem_diff(r, s2, s1); andre@0: felem_sum(r, r, r); andre@0: felem_mul(v, u1, i); andre@0: andre@0: felem_mul(z_out, tmp, h); andre@0: felem_square(rr, r); andre@0: felem_diff(x_out, rr, j); andre@0: felem_diff(x_out, x_out, v); andre@0: felem_diff(x_out, x_out, v); andre@0: andre@0: felem_diff(tmp, v, x_out); andre@0: felem_mul(y_out, tmp, r); andre@0: felem_mul(tmp, s1, j); andre@0: felem_diff(y_out, y_out, tmp); andre@0: felem_diff(y_out, y_out, tmp); andre@0: } andre@0: andre@0: /* point_add_or_double_vartime sets {x_out,y_out,z_out} = {x1,y1,z1} + andre@0: * {x2,y2,z2}. andre@0: * andre@0: * See http://www.hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-0.html#addition-add-2007-bl andre@0: * andre@0: * This function handles the case where {x1,y1,z1}={x2,y2,z2}. andre@0: */ andre@0: static void point_add_or_double_vartime( andre@0: felem x_out, felem y_out, felem z_out, andre@0: const felem x1, const felem y1, const felem z1, andre@0: const felem x2, const felem y2, const felem z2) andre@0: { andre@0: felem z1z1, z1z1z1, z2z2, z2z2z2, s1, s2, u1, u2, h, i, j, r, rr, v, tmp; andre@0: char x_equal, y_equal; andre@0: andre@0: felem_square(z1z1, z1); andre@0: felem_square(z2z2, z2); andre@0: felem_mul(u1, x1, z2z2); andre@0: andre@0: felem_sum(tmp, z1, z2); andre@0: felem_square(tmp, tmp); andre@0: felem_diff(tmp, tmp, z1z1); andre@0: felem_diff(tmp, tmp, z2z2); andre@0: andre@0: felem_mul(z2z2z2, z2, z2z2); andre@0: felem_mul(s1, y1, z2z2z2); andre@0: andre@0: felem_mul(u2, x2, z1z1); andre@0: felem_mul(z1z1z1, z1, z1z1); andre@0: felem_mul(s2, y2, z1z1z1); andre@0: felem_diff(h, u2, u1); andre@0: x_equal = felem_is_zero_vartime(h); andre@0: felem_sum(i, h, h); andre@0: felem_square(i, i); andre@0: felem_mul(j, h, i); andre@0: felem_diff(r, s2, s1); andre@0: y_equal = felem_is_zero_vartime(r); andre@0: if (x_equal && y_equal) { andre@0: point_double(x_out, y_out, z_out, x1, y1, z1); andre@0: return; andre@0: } andre@0: felem_sum(r, r, r); andre@0: felem_mul(v, u1, i); andre@0: andre@0: felem_mul(z_out, tmp, h); andre@0: felem_square(rr, r); andre@0: felem_diff(x_out, rr, j); andre@0: felem_diff(x_out, x_out, v); andre@0: felem_diff(x_out, x_out, v); andre@0: andre@0: felem_diff(tmp, v, x_out); andre@0: felem_mul(y_out, tmp, r); andre@0: felem_mul(tmp, s1, j); andre@0: felem_diff(y_out, y_out, tmp); andre@0: felem_diff(y_out, y_out, tmp); andre@0: } andre@0: andre@0: /* copy_conditional sets out=in if mask = 0xffffffff in constant time. andre@0: * andre@0: * On entry: mask is either 0 or 0xffffffff. andre@0: */ andre@0: static void copy_conditional(felem out, const felem in, limb mask) andre@0: { andre@0: int i; andre@0: andre@0: for (i = 0; i < NLIMBS; i++) { andre@0: const limb tmp = mask & (in[i] ^ out[i]); andre@0: out[i] ^= tmp; andre@0: } andre@0: } andre@0: andre@0: /* select_affine_point sets {out_x,out_y} to the index'th entry of table. andre@0: * On entry: index < 16, table[0] must be zero. andre@0: */ andre@0: static void select_affine_point(felem out_x, felem out_y, andre@0: const limb *table, limb index) andre@0: { andre@0: limb i, j; andre@0: andre@0: memset(out_x, 0, sizeof(felem)); andre@0: memset(out_y, 0, sizeof(felem)); andre@0: andre@0: for (i = 1; i < 16; i++) { andre@0: limb mask = i ^ index; andre@0: mask |= mask >> 2; andre@0: mask |= mask >> 1; andre@0: mask &= 1; andre@0: mask--; andre@0: for (j = 0; j < NLIMBS; j++, table++) { andre@0: out_x[j] |= *table & mask; andre@0: } andre@0: for (j = 0; j < NLIMBS; j++, table++) { andre@0: out_y[j] |= *table & mask; andre@0: } andre@0: } andre@0: } andre@0: andre@0: /* select_jacobian_point sets {out_x,out_y,out_z} to the index'th entry of andre@0: * table. On entry: index < 16, table[0] must be zero. andre@0: */ andre@0: static void select_jacobian_point(felem out_x, felem out_y, felem out_z, andre@0: const limb *table, limb index) andre@0: { andre@0: limb i, j; andre@0: andre@0: memset(out_x, 0, sizeof(felem)); andre@0: memset(out_y, 0, sizeof(felem)); andre@0: memset(out_z, 0, sizeof(felem)); andre@0: andre@0: /* The implicit value at index 0 is all zero. We don't need to perform that andre@0: * iteration of the loop because we already set out_* to zero. andre@0: */ andre@0: table += 3*NLIMBS; andre@0: andre@0: for (i = 1; i < 16; i++) { andre@0: limb mask = i ^ index; andre@0: mask |= mask >> 2; andre@0: mask |= mask >> 1; andre@0: mask &= 1; andre@0: mask--; andre@0: for (j = 0; j < NLIMBS; j++, table++) { andre@0: out_x[j] |= *table & mask; andre@0: } andre@0: for (j = 0; j < NLIMBS; j++, table++) { andre@0: out_y[j] |= *table & mask; andre@0: } andre@0: for (j = 0; j < NLIMBS; j++, table++) { andre@0: out_z[j] |= *table & mask; andre@0: } andre@0: } andre@0: } andre@0: andre@0: /* get_bit returns the bit'th bit of scalar. */ andre@0: static char get_bit(const u8 scalar[32], int bit) andre@0: { andre@0: return ((scalar[bit >> 3]) >> (bit & 7)) & 1; andre@0: } andre@0: andre@0: /* scalar_base_mult sets {nx,ny,nz} = scalar*G where scalar is a little-endian andre@0: * number. Note that the value of scalar must be less than the order of the andre@0: * group. andre@0: */ andre@0: static void scalar_base_mult(felem nx, felem ny, felem nz, const u8 scalar[32]) andre@0: { andre@0: int i, j; andre@0: limb n_is_infinity_mask = -1, p_is_noninfinite_mask, mask; andre@0: u32 table_offset; andre@0: andre@0: felem px, py; andre@0: felem tx, ty, tz; andre@0: andre@0: memset(nx, 0, sizeof(felem)); andre@0: memset(ny, 0, sizeof(felem)); andre@0: memset(nz, 0, sizeof(felem)); andre@0: andre@0: /* The loop adds bits at positions 0, 64, 128 and 192, followed by andre@0: * positions 32,96,160 and 224 and does this 32 times. andre@0: */ andre@0: for (i = 0; i < 32; i++) { andre@0: if (i) { andre@0: point_double(nx, ny, nz, nx, ny, nz); andre@0: } andre@0: table_offset = 0; andre@0: for (j = 0; j <= 32; j += 32) { andre@0: char bit0 = get_bit(scalar, 31 - i + j); andre@0: char bit1 = get_bit(scalar, 95 - i + j); andre@0: char bit2 = get_bit(scalar, 159 - i + j); andre@0: char bit3 = get_bit(scalar, 223 - i + j); andre@0: limb index = bit0 | (bit1 << 1) | (bit2 << 2) | (bit3 << 3); andre@0: andre@0: select_affine_point(px, py, kPrecomputed + table_offset, index); andre@0: table_offset += 30 * NLIMBS; andre@0: andre@0: /* Since scalar is less than the order of the group, we know that andre@0: * {nx,ny,nz} != {px,py,1}, unless both are zero, which we handle andre@0: * below. andre@0: */ andre@0: point_add_mixed(tx, ty, tz, nx, ny, nz, px, py); andre@0: /* The result of point_add_mixed is incorrect if {nx,ny,nz} is zero andre@0: * (a.k.a. the point at infinity). We handle that situation by andre@0: * copying the point from the table. andre@0: */ andre@0: copy_conditional(nx, px, n_is_infinity_mask); andre@0: copy_conditional(ny, py, n_is_infinity_mask); andre@0: copy_conditional(nz, kOne, n_is_infinity_mask); andre@0: andre@0: /* Equally, the result is also wrong if the point from the table is andre@0: * zero, which happens when the index is zero. We handle that by andre@0: * only copying from {tx,ty,tz} to {nx,ny,nz} if index != 0. andre@0: */ andre@0: p_is_noninfinite_mask = NON_ZERO_TO_ALL_ONES(index); andre@0: mask = p_is_noninfinite_mask & ~n_is_infinity_mask; andre@0: copy_conditional(nx, tx, mask); andre@0: copy_conditional(ny, ty, mask); andre@0: copy_conditional(nz, tz, mask); andre@0: /* If p was not zero, then n is now non-zero. */ andre@0: n_is_infinity_mask &= ~p_is_noninfinite_mask; andre@0: } andre@0: } andre@0: } andre@0: andre@0: /* point_to_affine converts a Jacobian point to an affine point. If the input andre@0: * is the point at infinity then it returns (0, 0) in constant time. andre@0: */ andre@0: static void point_to_affine(felem x_out, felem y_out, andre@0: const felem nx, const felem ny, const felem nz) { andre@0: felem z_inv, z_inv_sq; andre@0: felem_inv(z_inv, nz); andre@0: felem_square(z_inv_sq, z_inv); andre@0: felem_mul(x_out, nx, z_inv_sq); andre@0: felem_mul(z_inv, z_inv, z_inv_sq); andre@0: felem_mul(y_out, ny, z_inv); andre@0: } andre@0: andre@0: /* scalar_mult sets {nx,ny,nz} = scalar*{x,y}. */ andre@0: static void scalar_mult(felem nx, felem ny, felem nz, andre@0: const felem x, const felem y, const u8 scalar[32]) andre@0: { andre@0: int i; andre@0: felem px, py, pz, tx, ty, tz; andre@0: felem precomp[16][3]; andre@0: limb n_is_infinity_mask, index, p_is_noninfinite_mask, mask; andre@0: andre@0: /* We precompute 0,1,2,... times {x,y}. */ andre@0: memset(precomp, 0, sizeof(felem) * 3); andre@0: memcpy(&precomp[1][0], x, sizeof(felem)); andre@0: memcpy(&precomp[1][1], y, sizeof(felem)); andre@0: memcpy(&precomp[1][2], kOne, sizeof(felem)); andre@0: andre@0: for (i = 2; i < 16; i += 2) { andre@0: point_double(precomp[i][0], precomp[i][1], precomp[i][2], andre@0: precomp[i / 2][0], precomp[i / 2][1], precomp[i / 2][2]); andre@0: andre@0: point_add_mixed(precomp[i + 1][0], precomp[i + 1][1], precomp[i + 1][2], andre@0: precomp[i][0], precomp[i][1], precomp[i][2], x, y); andre@0: } andre@0: andre@0: memset(nx, 0, sizeof(felem)); andre@0: memset(ny, 0, sizeof(felem)); andre@0: memset(nz, 0, sizeof(felem)); andre@0: n_is_infinity_mask = -1; andre@0: andre@0: /* We add in a window of four bits each iteration and do this 64 times. */ andre@0: for (i = 0; i < 64; i++) { andre@0: if (i) { andre@0: point_double(nx, ny, nz, nx, ny, nz); andre@0: point_double(nx, ny, nz, nx, ny, nz); andre@0: point_double(nx, ny, nz, nx, ny, nz); andre@0: point_double(nx, ny, nz, nx, ny, nz); andre@0: } andre@0: andre@0: index = scalar[31 - i / 2]; andre@0: if ((i & 1) == 1) { andre@0: index &= 15; andre@0: } else { andre@0: index >>= 4; andre@0: } andre@0: andre@0: /* See the comments in scalar_base_mult about handling infinities. */ andre@0: select_jacobian_point(px, py, pz, precomp[0][0], index); andre@0: point_add(tx, ty, tz, nx, ny, nz, px, py, pz); andre@0: copy_conditional(nx, px, n_is_infinity_mask); andre@0: copy_conditional(ny, py, n_is_infinity_mask); andre@0: copy_conditional(nz, pz, n_is_infinity_mask); andre@0: andre@0: p_is_noninfinite_mask = NON_ZERO_TO_ALL_ONES(index); andre@0: mask = p_is_noninfinite_mask & ~n_is_infinity_mask; andre@0: copy_conditional(nx, tx, mask); andre@0: copy_conditional(ny, ty, mask); andre@0: copy_conditional(nz, tz, mask); andre@0: n_is_infinity_mask &= ~p_is_noninfinite_mask; andre@0: } andre@0: } andre@0: andre@0: /* Interface with Freebl: */ andre@0: andre@0: /* BYTESWAP_MP_DIGIT_TO_LE swaps the bytes of a mp_digit to andre@0: * little-endian order. andre@0: */ andre@0: #ifdef IS_BIG_ENDIAN andre@0: #ifdef __APPLE__ andre@0: #include andre@0: #define BYTESWAP32(x) OSSwapInt32(x) andre@0: #define BYTESWAP64(x) OSSwapInt64(x) andre@0: #else andre@0: #define BYTESWAP32(x) \ andre@0: ((x) >> 24 | (x) >> 8 & 0xff00 | ((x) & 0xff00) << 8 | (x) << 24) andre@0: #define BYTESWAP64(x) \ andre@0: ((x) >> 56 | (x) >> 40 & 0xff00 | \ andre@0: (x) >> 24 & 0xff0000 | (x) >> 8 & 0xff000000 | \ andre@0: ((x) & 0xff000000) << 8 | ((x) & 0xff0000) << 24 | \ andre@0: ((x) & 0xff00) << 40 | (x) << 56) andre@0: #endif andre@0: andre@0: #ifdef MP_USE_UINT_DIGIT andre@0: #define BYTESWAP_MP_DIGIT_TO_LE(x) BYTESWAP32(x) andre@0: #else andre@0: #define BYTESWAP_MP_DIGIT_TO_LE(x) BYTESWAP64(x) andre@0: #endif andre@0: #endif /* IS_BIG_ENDIAN */ andre@0: andre@0: #ifdef MP_USE_UINT_DIGIT andre@0: static const mp_digit kRInvDigits[8] = { andre@0: 0x80000000, 1, 0xffffffff, 0, andre@0: 0x80000001, 0xfffffffe, 1, 0x7fffffff andre@0: }; andre@0: #else andre@0: static const mp_digit kRInvDigits[4] = { andre@0: PR_UINT64(0x180000000), 0xffffffff, andre@0: PR_UINT64(0xfffffffe80000001), PR_UINT64(0x7fffffff00000001) andre@0: }; andre@0: #endif andre@0: #define MP_DIGITS_IN_256_BITS (32/sizeof(mp_digit)) andre@0: static const mp_int kRInv = { andre@0: MP_ZPOS, andre@0: MP_DIGITS_IN_256_BITS, andre@0: MP_DIGITS_IN_256_BITS, andre@0: (mp_digit*) kRInvDigits andre@0: }; andre@0: andre@0: static const limb kTwo28 = 0x10000000; andre@0: static const limb kTwo29 = 0x20000000; andre@0: andre@0: /* to_montgomery sets out = R*in. */ andre@0: static mp_err to_montgomery(felem out, const mp_int *in, const ECGroup *group) andre@0: { andre@0: /* There are no MPI functions for bitshift operations and we wish to shift andre@0: * in 257 bits left so we move the digits 256-bits left and then multiply andre@0: * by two. andre@0: */ andre@0: mp_int in_shifted; andre@0: int i; andre@0: mp_err res; andre@0: andre@0: mp_init(&in_shifted); andre@0: s_mp_pad(&in_shifted, MP_USED(in) + MP_DIGITS_IN_256_BITS); andre@0: memcpy(&MP_DIGIT(&in_shifted, MP_DIGITS_IN_256_BITS), andre@0: MP_DIGITS(in), andre@0: MP_USED(in)*sizeof(mp_digit)); andre@0: mp_mul_2(&in_shifted, &in_shifted); andre@0: MP_CHECKOK(group->meth->field_mod(&in_shifted, &in_shifted, group->meth)); andre@0: andre@0: for (i = 0;; i++) { andre@0: out[i] = MP_DIGIT(&in_shifted, 0) & kBottom29Bits; andre@0: mp_div_d(&in_shifted, kTwo29, &in_shifted, NULL); andre@0: andre@0: i++; andre@0: if (i == NLIMBS) andre@0: break; andre@0: out[i] = MP_DIGIT(&in_shifted, 0) & kBottom28Bits; andre@0: mp_div_d(&in_shifted, kTwo28, &in_shifted, NULL); andre@0: } andre@0: andre@0: CLEANUP: andre@0: mp_clear(&in_shifted); andre@0: return res; andre@0: } andre@0: andre@0: /* from_montgomery sets out=in/R. */ andre@0: static mp_err from_montgomery(mp_int *out, const felem in, andre@0: const ECGroup *group) andre@0: { andre@0: mp_int result, tmp; andre@0: mp_err res; andre@0: int i; andre@0: andre@0: mp_init(&result); andre@0: mp_init(&tmp); andre@0: andre@0: MP_CHECKOK(mp_add_d(&tmp, in[NLIMBS-1], &result)); andre@0: for (i = NLIMBS-2; i >= 0; i--) { andre@0: if ((i & 1) == 0) { andre@0: MP_CHECKOK(mp_mul_d(&result, kTwo29, &tmp)); andre@0: } else { andre@0: MP_CHECKOK(mp_mul_d(&result, kTwo28, &tmp)); andre@0: } andre@0: MP_CHECKOK(mp_add_d(&tmp, in[i], &result)); andre@0: } andre@0: andre@0: MP_CHECKOK(mp_mul(&result, &kRInv, out)); andre@0: MP_CHECKOK(group->meth->field_mod(out, out, group->meth)); andre@0: andre@0: CLEANUP: andre@0: mp_clear(&result); andre@0: mp_clear(&tmp); andre@0: return res; andre@0: } andre@0: andre@0: /* scalar_from_mp_int sets out_scalar=n, where n < the group order. */ andre@0: static void scalar_from_mp_int(u8 out_scalar[32], const mp_int *n) andre@0: { andre@0: /* We require that |n| is less than the order of the group and therefore it andre@0: * will fit into |out_scalar|. However, these is a timing side-channel here andre@0: * that we cannot avoid: if |n| is sufficiently small it may be one or more andre@0: * words too short and we'll copy less data. andre@0: */ andre@0: memset(out_scalar, 0, 32); andre@0: #ifdef IS_LITTLE_ENDIAN andre@0: memcpy(out_scalar, MP_DIGITS(n), MP_USED(n) * sizeof(mp_digit)); andre@0: #else andre@0: { andre@0: mp_size i; andre@0: mp_digit swapped[MP_DIGITS_IN_256_BITS]; andre@0: for (i = 0; i < MP_USED(n); i++) { andre@0: swapped[i] = BYTESWAP_MP_DIGIT_TO_LE(MP_DIGIT(n, i)); andre@0: } andre@0: memcpy(out_scalar, swapped, MP_USED(n) * sizeof(mp_digit)); andre@0: } andre@0: #endif andre@0: } andre@0: andre@0: /* ec_GFp_nistp256_base_point_mul sets {out_x,out_y} = nG, where n is < the andre@0: * order of the group. andre@0: */ andre@0: static mp_err ec_GFp_nistp256_base_point_mul(const mp_int *n, andre@0: mp_int *out_x, mp_int *out_y, andre@0: const ECGroup *group) andre@0: { andre@0: u8 scalar[32]; andre@0: felem x, y, z, x_affine, y_affine; andre@0: mp_err res; andre@0: andre@0: /* FIXME(agl): test that n < order. */ andre@0: andre@0: scalar_from_mp_int(scalar, n); andre@0: scalar_base_mult(x, y, z, scalar); andre@0: point_to_affine(x_affine, y_affine, x, y, z); andre@0: MP_CHECKOK(from_montgomery(out_x, x_affine, group)); andre@0: MP_CHECKOK(from_montgomery(out_y, y_affine, group)); andre@0: andre@0: CLEANUP: andre@0: return res; andre@0: } andre@0: andre@0: /* ec_GFp_nistp256_point_mul sets {out_x,out_y} = n*{in_x,in_y}, where n is < andre@0: * the order of the group. andre@0: */ andre@0: static mp_err ec_GFp_nistp256_point_mul(const mp_int *n, andre@0: const mp_int *in_x, const mp_int *in_y, andre@0: mp_int *out_x, mp_int *out_y, andre@0: const ECGroup *group) andre@0: { andre@0: u8 scalar[32]; andre@0: felem x, y, z, x_affine, y_affine, px, py; andre@0: mp_err res; andre@0: andre@0: scalar_from_mp_int(scalar, n); andre@0: andre@0: MP_CHECKOK(to_montgomery(px, in_x, group)); andre@0: MP_CHECKOK(to_montgomery(py, in_y, group)); andre@0: andre@0: scalar_mult(x, y, z, px, py, scalar); andre@0: point_to_affine(x_affine, y_affine, x, y, z); andre@0: MP_CHECKOK(from_montgomery(out_x, x_affine, group)); andre@0: MP_CHECKOK(from_montgomery(out_y, y_affine, group)); andre@0: andre@0: CLEANUP: andre@0: return res; andre@0: } andre@0: andre@0: /* ec_GFp_nistp256_point_mul_vartime sets {out_x,out_y} = n1*G + andre@0: * n2*{in_x,in_y}, where n1 and n2 are < the order of the group. andre@0: * andre@0: * As indicated by the name, this function operates in variable time. This andre@0: * is safe because it's used for signature validation which doesn't deal andre@0: * with secrets. andre@0: */ andre@0: static mp_err ec_GFp_nistp256_points_mul_vartime( andre@0: const mp_int *n1, const mp_int *n2, andre@0: const mp_int *in_x, const mp_int *in_y, andre@0: mp_int *out_x, mp_int *out_y, andre@0: const ECGroup *group) andre@0: { andre@0: u8 scalar1[32], scalar2[32]; andre@0: felem x1, y1, z1, x2, y2, z2, x_affine, y_affine, px, py; andre@0: mp_err res = MP_OKAY; andre@0: andre@0: /* If n2 == NULL, this is just a base-point multiplication. */ andre@0: if (n2 == NULL) { andre@0: return ec_GFp_nistp256_base_point_mul(n1, out_x, out_y, group); andre@0: } andre@0: andre@0: /* If n1 == nULL, this is just an arbitary-point multiplication. */ andre@0: if (n1 == NULL) { andre@0: return ec_GFp_nistp256_point_mul(n2, in_x, in_y, out_x, out_y, group); andre@0: } andre@0: andre@0: /* If both scalars are zero, then the result is the point at infinity. */ andre@0: if (mp_cmp_z(n1) == 0 && mp_cmp_z(n2) == 0) { andre@0: mp_zero(out_x); andre@0: mp_zero(out_y); andre@0: return res; andre@0: } andre@0: andre@0: scalar_from_mp_int(scalar1, n1); andre@0: scalar_from_mp_int(scalar2, n2); andre@0: andre@0: MP_CHECKOK(to_montgomery(px, in_x, group)); andre@0: MP_CHECKOK(to_montgomery(py, in_y, group)); andre@0: scalar_base_mult(x1, y1, z1, scalar1); andre@0: scalar_mult(x2, y2, z2, px, py, scalar2); andre@0: andre@0: if (mp_cmp_z(n2) == 0) { andre@0: /* If n2 == 0, then {x2,y2,z2} is zero and the result is just andre@0: * {x1,y1,z1}. */ andre@0: } else if (mp_cmp_z(n1) == 0) { andre@0: /* If n1 == 0, then {x1,y1,z1} is zero and the result is just andre@0: * {x2,y2,z2}. */ andre@0: memcpy(x1, x2, sizeof(x2)); andre@0: memcpy(y1, y2, sizeof(y2)); andre@0: memcpy(z1, z2, sizeof(z2)); andre@0: } else { andre@0: /* This function handles the case where {x1,y1,z1} == {x2,y2,z2}. */ andre@0: point_add_or_double_vartime(x1, y1, z1, x1, y1, z1, x2, y2, z2); andre@0: } andre@0: andre@0: point_to_affine(x_affine, y_affine, x1, y1, z1); andre@0: MP_CHECKOK(from_montgomery(out_x, x_affine, group)); andre@0: MP_CHECKOK(from_montgomery(out_y, y_affine, group)); andre@0: andre@0: CLEANUP: andre@0: return res; andre@0: } andre@0: andre@0: /* Wire in fast point multiplication for named curves. */ andre@0: mp_err ec_group_set_gfp256_32(ECGroup *group, ECCurveName name) andre@0: { andre@0: if (name == ECCurve_NIST_P256) { andre@0: group->base_point_mul = &ec_GFp_nistp256_base_point_mul; andre@0: group->point_mul = &ec_GFp_nistp256_point_mul; andre@0: group->points_mul = &ec_GFp_nistp256_points_mul_vartime; andre@0: } andre@0: return MP_OKAY; andre@0: }