| 1 | /* -*-c-*- |
| 2 | * |
| 3 | * The GCM authenticated encryption mode |
| 4 | * |
| 5 | * (c) 2017 Straylight/Edgeware |
| 6 | */ |
| 7 | |
| 8 | /*----- Licensing notice --------------------------------------------------* |
| 9 | * |
| 10 | * This file is part of Catacomb. |
| 11 | * |
| 12 | * Catacomb is free software: you can redistribute it and/or modify it |
| 13 | * under the terms of the GNU Library General Public License as published |
| 14 | * by the Free Software Foundation; either version 2 of the License, or |
| 15 | * (at your option) any later version. |
| 16 | * |
| 17 | * Catacomb is distributed in the hope that it will be useful, but |
| 18 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
| 19 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 20 | * Library General Public License for more details. |
| 21 | * |
| 22 | * You should have received a copy of the GNU Library General Public |
| 23 | * License along with Catacomb. If not, write to the Free Software |
| 24 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, |
| 25 | * USA. |
| 26 | */ |
| 27 | |
| 28 | /*----- Header files ------------------------------------------------------*/ |
| 29 | |
| 30 | #include "config.h" |
| 31 | |
| 32 | #include <stdio.h> |
| 33 | |
| 34 | #include <mLib/bits.h> |
| 35 | |
| 36 | #include "dispatch.h" |
| 37 | #include "gcm.h" |
| 38 | #include "gcm-def.h" |
| 39 | |
| 40 | /*----- Overall strategy --------------------------------------------------* |
| 41 | * |
| 42 | * GCM is pretty awful to implement in software. (This presentation is going |
| 43 | * to be somewhat different to that in the specification, but I think it |
| 44 | * makes more sense like this.) |
| 45 | * |
| 46 | * We're given a %$w$%-bit blockcipher %$E$% with a key %$K$%. |
| 47 | * |
| 48 | * The main part is arithmetic in the finite field %$k = \gf{2^w}$%, which we |
| 49 | * represent as the quotient ring %$\gf{2}[t]/(p_w(t))$% for some irreducible |
| 50 | * degree-%$w$% polynomial %$p(t)$%, whose precise value isn't very important |
| 51 | * right now. We choose a secret point %$x = E_K(0^w)$%. |
| 52 | * |
| 53 | * We choose a length size %$z$% as follows: if %$w < 96%$ then %$z = w$%; |
| 54 | * otherwise %$z = w/2$%. Format a message pair as follows: |
| 55 | * |
| 56 | * %$F(a, b) = P_w(a) \cat P_w(b) \cat [\ell(a)]_z \cat [\ell(b)]_z$% |
| 57 | * |
| 58 | * where %$P_w(x) = x \cat 0^n$% where $%0 \le n < w$% such that |
| 59 | * %$\ell(x) + n \equiv 0 \pmod{w}$%. |
| 60 | * |
| 61 | * Hash a (block-aligned) message %$u$% as follows. First, split %$u$% into |
| 62 | * %$w$%-bit blocks %$u_0$%, %$u_1$%, %%\ldots%%, %$u_{n-1}$%. Interpret |
| 63 | * these as elements of %$k$%. Then |
| 64 | * |
| 65 | * %$G_x(u) = u_0 t^n + u_1 t^{n-1} + \cdots + u_{n-1} t$% |
| 66 | * |
| 67 | * converted back to a %$w$%-bit string. |
| 68 | * |
| 69 | * We're ready to go now. Suppose we're to encrypt a message %$M$% with |
| 70 | * header %$H$% and nonce %$N$%. If %$\ell(N) + 32 = w$% then let |
| 71 | * %$N' = N$% and let %$i_0 = 1$%; otherwise, let %$U = G_t(F(\epsilon, N))$% |
| 72 | * and split this into %$N' = U[0 \bitsto w - 32]$% and |
| 73 | * %$[i_0]_{32} = U[w - 32 \bitsto w]$%. |
| 74 | * |
| 75 | * Let %$n = \lceil \ell(M)/w \rceil$%. Compute |
| 76 | * |
| 77 | * %$y_j = E_K(N' \cat [i_0 + j]_{32})$% |
| 78 | * |
| 79 | * for %$0 \le j \le n$%. Let |
| 80 | * |
| 81 | * %$s = (y_1 \cat y_2 \cat \cdots \cat y_n)[0 \bitsto \ell(M)$% |
| 82 | * |
| 83 | * Let %$C = M \xor s$% and let %$T = G_x(F(H, C)) \xor y_0$%. These are the |
| 84 | * ciphertext and tag respectively. |
| 85 | * |
| 86 | * So why is this awful? |
| 87 | * |
| 88 | * For one thing, the bits are in a completely terrible order. The bytes are |
| 89 | * arranged in little-endian order, so the unit coefficient is in the first |
| 90 | * byte, and the degree-127 coefficient is in the last byte. But within each |
| 91 | * byte, the lowest-degree coefficient is in the most significant bit. It's |
| 92 | * therefore better to think of GCM as using a big-endian byte-ordering |
| 93 | * convention, but with the bits backwards. |
| 94 | * |
| 95 | * But messing about with byte ordering is expensive, so let's not do that in |
| 96 | * the inner loop. But multiplication in %$k$% is not easy either. Some |
| 97 | * kind of precomputed table would be nice, but that will leak secrets |
| 98 | * through the cache. |
| 99 | * |
| 100 | * I choose a particularly simple table: given %$x$%, let %$X[i'] = x t^i$%. |
| 101 | * Then $%$x y = \sum_{0\le i<w} y_i X[i']$% which is just a bunch of |
| 102 | * bitmasking. But the natural order for examining bits of %$y$% is not |
| 103 | * necessarily the obvious one. We'll have already loaded %$y$% into |
| 104 | * internal form, as 32-bit words. The good order to process these is left |
| 105 | * to right, from high to low bits. But now the order of degrees depends on |
| 106 | * the endianness of our conversion of bytes to words. Oh, well. |
| 107 | * |
| 108 | * If we've adopted a big-endian convention, then we'll see the degrees in |
| 109 | * order, 0, 1, ..., all the way up to %$w - 1$% and everything is fine. If |
| 110 | * we've adopted a little-endian convention, though, we'll see an ordering |
| 111 | * like this: |
| 112 | * |
| 113 | * 24, 25, ..., 31, 16, 17, ..., 23, 8, 9, ..., 15, 0, 1, ..., 7, |
| 114 | * 56, 57, ..., 63, 48, 49, ..., 55, 40, 41, ..., 47, 32, 33, ..., 39, |
| 115 | * etc. |
| 116 | * |
| 117 | * which is the ordinary order with 0x18 = 24 XORed into the index. That is, |
| 118 | * %$i' = i$% if we've adopted a big-endian convention, and |
| 119 | * %$i' = i \xor 24$% if we've adopted a little-endian convention. |
| 120 | */ |
| 121 | |
| 122 | /*----- Low-level utilities -----------------------------------------------*/ |
| 123 | |
| 124 | /* --- @mult@ --- * |
| 125 | * |
| 126 | * Arguments: @const gcm_params *p@ = pointer to the parameters |
| 127 | * @uint32 *z@ = where to write the result |
| 128 | * @const uint32 *x@ = input field element |
| 129 | * |
| 130 | * Returns: --- |
| 131 | * |
| 132 | * Use: Multiply the input field element by %$t$%, and write the |
| 133 | * product to @z@. It's safe for @x@ and @z@ to be equal, but |
| 134 | * they should not otherwise overlap. Both input and output are |
| 135 | * in big-endian form, i.e., with the lowest-degree coefficients |
| 136 | * in the most significant bits. |
| 137 | */ |
| 138 | |
| 139 | static void mult(const gcm_params *p, uint32 *z, const uint32 *x) |
| 140 | { |
| 141 | uint32 m, c, t; |
| 142 | unsigned i; |
| 143 | |
| 144 | t = x[p->n - 1]; m = -(t&1u); c = m&p->poly; |
| 145 | for (i = 0; i < p->n; i++) { t = x[i]; z[i] = (t >> 1) ^ c; c = t << 31; } |
| 146 | } |
| 147 | |
| 148 | /* --- @mul@ --- * |
| 149 | * |
| 150 | * Arguments: @const gcm_params *p@ = pointer to the parameters |
| 151 | * @uint32 *z@ = where to write the result |
| 152 | * @const uint32 *x, *y@ = input field elements |
| 153 | * |
| 154 | * Returns: --- |
| 155 | * |
| 156 | * Use: Multiply the input field elements together, and write the |
| 157 | * product to @z@. It's safe for the operands to overlap. Both |
| 158 | * inputs and the output are in big-endian form, i.e., with the |
| 159 | * lowest-degree coefficients in the most significant bits. |
| 160 | */ |
| 161 | |
| 162 | static void mul(const gcm_params *p, uint32 *z, |
| 163 | const uint32 *x, const uint32 *y) |
| 164 | { |
| 165 | uint32 m, t, u[GCM_NMAX], v[GCM_NMAX]; |
| 166 | unsigned i, j, k; |
| 167 | |
| 168 | /* We can't do this in-place at all, so use temporary space. Make a copy |
| 169 | * of @x@ in @u@, where we can clobber it, and build the product in @v@. |
| 170 | */ |
| 171 | for (i = 0; i < p->n; i++) { u[i] = x[i]; v[i] = 0; } |
| 172 | |
| 173 | /* Repeatedly multiply @x@ (in @u@) by %$t$%, and add together those |
| 174 | * %$x t^i$% selected by the bits of @y@. This is basically what you get |
| 175 | * by streaming the result of @gcm_mktable@ into @gcm_mulk_...@. |
| 176 | */ |
| 177 | for (i = 0; i < p->n; i++) { |
| 178 | t = y[i]; |
| 179 | for (j = 0; j < 32; j++) { |
| 180 | m = -((t >> 31)&1u); |
| 181 | for (k = 0; k < p->n; k++) v[k] ^= u[k]&m; |
| 182 | mult(p, u, u); t <<= 1; |
| 183 | } |
| 184 | } |
| 185 | |
| 186 | /* Write out the result now that it's ready. */ |
| 187 | for (i = 0; i < p->n; i++) z[i] = v[i]; |
| 188 | } |
| 189 | |
| 190 | /*----- Table-based multiplication ----------------------------------------*/ |
| 191 | |
| 192 | /* --- @gcm_mktable@ --- * |
| 193 | * |
| 194 | * Arguments: @const gcm_params *p@ = pointer to the parameters |
| 195 | * @uint32 *ktab@ = where to write the table; there must be |
| 196 | * space for %$32 n$% $%n$%-word entries, i.e., |
| 197 | * %$32 n^2$% 32-bit words in total, where %$n$% is |
| 198 | * @p->n@, the block size in words |
| 199 | * @const uint32 *k@ = input field element |
| 200 | * |
| 201 | * Returns: --- |
| 202 | * |
| 203 | * Use: Construct a table for use by @gcm_mulk_...@ below, to |
| 204 | * multiply (vaguely) efficiently by @k@. |
| 205 | */ |
| 206 | |
| 207 | static void simple_mktable(const gcm_params *p, |
| 208 | uint32 *ktab, const uint32 *k) |
| 209 | { |
| 210 | unsigned m = (p->f&GCMF_SWAP ? 0x18 : 0); |
| 211 | unsigned i, j, o = m*p->n; |
| 212 | |
| 213 | /* As described above, the table stores entries %$K[i \xor m] = k t^i$%, |
| 214 | * where %$m = 0$% (big-endian cipher) or %$m = 24$% (little-endian). |
| 215 | * The first job is to store %$K[m] = k$%. |
| 216 | * |
| 217 | * We initially build the table with the entries in big-endian order, and |
| 218 | * then swap them if necessary. This makes the arithmetic functions more |
| 219 | * amenable for use by @gcm_concat@ below. |
| 220 | */ |
| 221 | if (!(p->f&GCMF_SWAP)) for (i = 0; i < p->n; i++) ktab[o + i] = k[i]; |
| 222 | else for (i = 0; i < p->n; i++) ktab[o + i] = ENDSWAP32(k[i]); |
| 223 | |
| 224 | /* Fill in the rest of the table by repeatedly multiplying the previous |
| 225 | * entry by %$t$%. |
| 226 | */ |
| 227 | for (i = 1; i < 32*p->n; i++) |
| 228 | { j = (i ^ m)*p->n; mult(p, ktab + j, ktab + o); o = j; } |
| 229 | |
| 230 | /* Finally, if the cipher uses a little-endian convention, then swap all of |
| 231 | * the individual words. |
| 232 | */ |
| 233 | if (p->f&GCMF_SWAP) |
| 234 | for (i = 0; i < 32*p->n*p->n; i++) ktab[i] = ENDSWAP32(ktab[i]); |
| 235 | } |
| 236 | |
| 237 | #if CPUFAM_X86 || CPUFAM_AMD64 |
| 238 | static void pclmul_mktable(const gcm_params *p, |
| 239 | uint32 *ktab, const uint32 *k) |
| 240 | { |
| 241 | unsigned n = p->n; |
| 242 | unsigned nz; |
| 243 | uint32 *t; |
| 244 | |
| 245 | /* We just need to store the value in a way which is convenient for the |
| 246 | * assembler code to read back. That involves reordering the words, and, |
| 247 | * in the case of 96-bit blocks, padding with zeroes to fill out a 128-bit |
| 248 | * chunk. |
| 249 | */ |
| 250 | |
| 251 | if (n == 3) nz = 1; |
| 252 | else nz = 0; |
| 253 | t = ktab + n + nz; |
| 254 | |
| 255 | if (p->f&GCMF_SWAP) while (n--) { *--t = ENDSWAP32(*k); k++; } |
| 256 | else while (n--) *--t = *k++; |
| 257 | while (nz--) *--t = 0; |
| 258 | } |
| 259 | #endif |
| 260 | |
| 261 | #if CPUFAM_ARMEL |
| 262 | static void arm_crypto_mktable(const gcm_params *p, |
| 263 | uint32 *ktab, const uint32 *k) |
| 264 | { |
| 265 | unsigned n = p->n; |
| 266 | uint32 *t; |
| 267 | |
| 268 | /* We just need to store the value in a way which is convenient for the |
| 269 | * assembler code to read back. That involves swapping the bytes in each |
| 270 | * 64-bit lane. |
| 271 | */ |
| 272 | |
| 273 | t = ktab; |
| 274 | if (p->f&GCMF_SWAP) { |
| 275 | while (n >= 2) { |
| 276 | t[1] = ENDSWAP32(k[0]); t[0] = ENDSWAP32(k[1]); |
| 277 | t += 2; k += 2; n -= 2; |
| 278 | } |
| 279 | if (n) { t[1] = ENDSWAP32(k[0]); t[0] = 0; } |
| 280 | } else { |
| 281 | while (n >= 2) { |
| 282 | t[1] = k[0]; t[0] = k[1]; |
| 283 | t += 2; k += 2; n -= 2; |
| 284 | } |
| 285 | if (n) { t[1] = k[0]; t[0] = 0; } |
| 286 | } |
| 287 | } |
| 288 | #endif |
| 289 | |
| 290 | #if CPUFAM_ARM64 |
| 291 | static uint32 rbit32(uint32 x) |
| 292 | { |
| 293 | uint32 z, t; |
| 294 | |
| 295 | #if GCC_VERSION_P(4, 3) |
| 296 | /* Two tricks here. Firstly, two separate steps, rather than a single |
| 297 | * block of assembler, to allow finer-grained instruction scheduling. |
| 298 | * Secondly, use `ENDSWAP32' so that the compiler can cancel it if the |
| 299 | * caller actually wants the bytes reordered. |
| 300 | */ |
| 301 | __asm__("rbit %w0, %w1" : "=r"(t) : "r"(x)); |
| 302 | z = ENDSWAP32(t); |
| 303 | #else |
| 304 | /* A generic but slightly clever implementation. */ |
| 305 | # define SWIZZLE(x, m, s) ((((x)&(m)) << (s)) | (((x)&~(m)) >> (s))) |
| 306 | /* 76543210 */ |
| 307 | t = SWIZZLE(x, 0x0f0f0f0f, 4); /* 32107654 -- swap nibbles */ |
| 308 | t = SWIZZLE(t, 0x33333333, 2); /* 10325476 -- swap bit pairs */ |
| 309 | z = SWIZZLE(t, 0x55555555, 1); /* 01234567 -- swap adjacent bits */ |
| 310 | # undef SWIZZLE |
| 311 | #endif |
| 312 | return (z); |
| 313 | } |
| 314 | |
| 315 | static void arm64_pmull_mktable(const gcm_params *p, |
| 316 | uint32 *ktab, const uint32 *k) |
| 317 | { |
| 318 | unsigned n = p->n; |
| 319 | uint32 *t; |
| 320 | |
| 321 | /* We just need to store the value in a way which is convenient for the |
| 322 | * assembler code to read back. That involves two transformations: |
| 323 | * |
| 324 | * * firstly, reversing the order of the bits in each byte; and, |
| 325 | * |
| 326 | * * secondly, storing two copies of each 64-bit chunk. |
| 327 | * |
| 328 | * Note that, in this case, we /want/ the little-endian byte order of GCM, |
| 329 | * so endianness-swapping happens in the big-endian case. |
| 330 | */ |
| 331 | |
| 332 | t = ktab; |
| 333 | if (p->f&GCMF_SWAP) { |
| 334 | while (n >= 2) { |
| 335 | t[0] = t[2] = rbit32(k[0]); |
| 336 | t[1] = t[3] = rbit32(k[1]); |
| 337 | t += 4; k += 2; n -= 2; |
| 338 | } |
| 339 | if (n) { t[0] = t[2] = rbit32(k[0]); t[1] = t[3] = 0; } |
| 340 | } else { |
| 341 | while (n >= 2) { |
| 342 | t[0] = t[2] = ENDSWAP32(rbit32(k[0])); |
| 343 | t[1] = t[3] = ENDSWAP32(rbit32(k[1])); |
| 344 | t += 4; k += 2; n -= 2; |
| 345 | } |
| 346 | if (n) { t[0] = t[2] = ENDSWAP32(rbit32(k[0])); t[1] = t[3] = 0; } |
| 347 | } |
| 348 | } |
| 349 | #endif |
| 350 | |
| 351 | CPU_DISPATCH(EMPTY, EMPTY, void, gcm_mktable, |
| 352 | (const gcm_params *p, uint32 *ktab, const uint32 *k), |
| 353 | (p, ktab, k), |
| 354 | pick_mktable, simple_mktable) |
| 355 | |
| 356 | static gcm_mktable__functype *pick_mktable(void) |
| 357 | { |
| 358 | #if CPUFAM_X86 || CPUFAM_AMD64 |
| 359 | DISPATCH_PICK_COND(gcm_mktable, pclmul_mktable, |
| 360 | cpu_feature_p(CPUFEAT_X86_SSSE3) && |
| 361 | cpu_feature_p(CPUFEAT_X86_PCLMUL)); |
| 362 | #endif |
| 363 | #if CPUFAM_ARMEL |
| 364 | DISPATCH_PICK_COND(gcm_mktable, arm_crypto_mktable, |
| 365 | cpu_feature_p(CPUFEAT_ARM_PMULL)); |
| 366 | #endif |
| 367 | #if CPUFAM_ARM64 |
| 368 | DISPATCH_PICK_COND(gcm_mktable, arm64_pmull_mktable, |
| 369 | cpu_feature_p(CPUFEAT_ARM_PMULL)); |
| 370 | #endif |
| 371 | DISPATCH_PICK_FALLBACK(gcm_mktable, simple_mktable); |
| 372 | } |
| 373 | |
| 374 | /* --- @recover_k@ --- * |
| 375 | * |
| 376 | * Arguments: @const gcm_params *p@ = pointer to the parameters |
| 377 | * @uint32 *k@ = block-sized vector in which to store %$k$% |
| 378 | * @const uint32 *ktab@ = the table encoding %$k$% |
| 379 | * |
| 380 | * Returns: --- |
| 381 | * |
| 382 | * Use: Recovers %$k$%, the secret from which @ktab@ was by |
| 383 | * @gcm_mktable@, from the table, and stores it in internal |
| 384 | * (big-endian) form in @k@. |
| 385 | */ |
| 386 | |
| 387 | static void simple_recover_k(const gcm_params *p, |
| 388 | uint32 *k, const uint32 *ktab) |
| 389 | { |
| 390 | unsigned i; |
| 391 | |
| 392 | /* If the blockcipher is big-endian, then the key is simply in the first |
| 393 | * table element, in the right format. If the blockcipher is little-endian |
| 394 | * then it's in element 24, and the bytes need swapping. |
| 395 | */ |
| 396 | |
| 397 | if (!(p->f&GCMF_SWAP)) for (i = 0; i < p->n; i++) k[i] = ktab[i]; |
| 398 | else for (i = 0; i < p->n; i++) k[i] = ENDSWAP32(ktab[24*p->n + i]); |
| 399 | } |
| 400 | |
| 401 | #if CPUFAM_X86 || CPUFAM_AMD64 |
| 402 | static void pclmul_recover_k(const gcm_params *p, |
| 403 | uint32 *k, const uint32 *ktab) |
| 404 | { |
| 405 | unsigned n = p->n; |
| 406 | unsigned nz; |
| 407 | const uint32 *t; |
| 408 | |
| 409 | /* The representation is already independent of the blockcipher endianness. |
| 410 | * We need to compensate for padding, and reorder the words. |
| 411 | */ |
| 412 | |
| 413 | if (n == 3) nz = 1; else nz = 0; |
| 414 | t = ktab + n + nz; |
| 415 | while (n--) *k++ = *--t; |
| 416 | } |
| 417 | #endif |
| 418 | |
| 419 | #if CPUFAM_ARMEL |
| 420 | static void arm_crypto_recover_k(const gcm_params *p, |
| 421 | uint32 *k, const uint32 *ktab) |
| 422 | { |
| 423 | unsigned n = p->n; |
| 424 | const uint32 *t; |
| 425 | |
| 426 | /* The representation is already independent of the blockcipher endianness. |
| 427 | * We only need to reorder the words. |
| 428 | */ |
| 429 | |
| 430 | t = ktab; |
| 431 | while (n >= 2) { k[1] = t[0]; k[0] = t[1]; t += 2; k += 2; n -= 2; } |
| 432 | if (n) k[0] = t[1]; |
| 433 | } |
| 434 | #endif |
| 435 | |
| 436 | #if CPUFAM_ARM64 |
| 437 | static void arm64_pmull_recover_k(const gcm_params *p, |
| 438 | uint32 *k, const uint32 *ktab) |
| 439 | { |
| 440 | unsigned n = p->n; |
| 441 | const uint32 *t; |
| 442 | |
| 443 | /* The representation is already independent of the blockcipher endianness. |
| 444 | * We need to skip the duplicate pieces, and unscramble the bytes. |
| 445 | */ |
| 446 | |
| 447 | t = ktab; |
| 448 | while (n >= 2) { |
| 449 | k[0] = ENDSWAP32(rbit32(t[0])); |
| 450 | k[1] = ENDSWAP32(rbit32(t[1])); |
| 451 | t += 4; k += 2; n -= 2; |
| 452 | } |
| 453 | if (n) k[0] = ENDSWAP32(rbit32(t[0])); |
| 454 | } |
| 455 | #endif |
| 456 | |
| 457 | CPU_DISPATCH(static, EMPTY, void, recover_k, |
| 458 | (const gcm_params *p, uint32 *k, const uint32 *ktab), |
| 459 | (p, k, ktab), |
| 460 | pick_recover_k, simple_recover_k) |
| 461 | |
| 462 | static recover_k__functype *pick_recover_k(void) |
| 463 | { |
| 464 | #if CPUFAM_X86 || CPUFAM_AMD64 |
| 465 | DISPATCH_PICK_COND(recover_k, pclmul_recover_k, |
| 466 | cpu_feature_p(CPUFEAT_X86_SSSE3) && |
| 467 | cpu_feature_p(CPUFEAT_X86_PCLMUL)); |
| 468 | #endif |
| 469 | #if CPUFAM_ARMEL |
| 470 | DISPATCH_PICK_COND(recover_k, arm_crypto_recover_k, |
| 471 | cpu_feature_p(CPUFEAT_ARM_PMULL)); |
| 472 | #endif |
| 473 | #if CPUFAM_ARM64 |
| 474 | DISPATCH_PICK_COND(recover_k, arm64_pmull_recover_k, |
| 475 | cpu_feature_p(CPUFEAT_ARM_PMULL)); |
| 476 | #endif |
| 477 | DISPATCH_PICK_FALLBACK(recover_k, simple_recover_k); |
| 478 | } |
| 479 | |
| 480 | /* --- @gcm_mulk_N{b,l}@ --- * |
| 481 | * |
| 482 | * Arguments: @uint32 *a@ = accumulator to multiply |
| 483 | * @const uint32 *ktab@ = table constructed by @gcm_mktable@ |
| 484 | * |
| 485 | * Returns: --- |
| 486 | * |
| 487 | * Use: Multiply @a@ by @k@ (implicitly represented in @ktab@), |
| 488 | * updating @a@ in-place. There are separate functions for each |
| 489 | * supported block size and endianness because this is the |
| 490 | * function whose performance actually matters. |
| 491 | */ |
| 492 | |
| 493 | #if CPUFAM_X86 || CPUFAM_AMD64 |
| 494 | # define DECL_MULK_X86ISH(var) extern gcm_mulk_##var##__functype \ |
| 495 | gcm_mulk_##var##_x86ish_pclmul_avx, \ |
| 496 | gcm_mulk_##var##_x86ish_pclmul; |
| 497 | # define PICK_MULK_X86ISH(var) do { \ |
| 498 | DISPATCH_PICK_COND(gcm_mulk_##var, gcm_mulk_##var##_x86ish_pclmul_avx, \ |
| 499 | cpu_feature_p(CPUFEAT_X86_AVX) && \ |
| 500 | cpu_feature_p(CPUFEAT_X86_PCLMUL) && \ |
| 501 | cpu_feature_p(CPUFEAT_X86_SSSE3)); \ |
| 502 | DISPATCH_PICK_COND(gcm_mulk_##var, gcm_mulk_##var##_x86ish_pclmul, \ |
| 503 | cpu_feature_p(CPUFEAT_X86_PCLMUL) && \ |
| 504 | cpu_feature_p(CPUFEAT_X86_SSSE3)); \ |
| 505 | } while (0) |
| 506 | #else |
| 507 | # define DECL_MULK_X86ISH(var) |
| 508 | # define PICK_MULK_X86ISH(var) do ; while (0) |
| 509 | #endif |
| 510 | |
| 511 | #if CPUFAM_ARMEL |
| 512 | # define DECL_MULK_ARM(var) \ |
| 513 | extern gcm_mulk_##var##__functype gcm_mulk_##var##_arm_crypto; |
| 514 | # define PICK_MULK_ARM(var) do { \ |
| 515 | DISPATCH_PICK_COND(gcm_mulk_##var, gcm_mulk_##var##_arm_crypto, \ |
| 516 | cpu_feature_p(CPUFEAT_ARM_PMULL)); \ |
| 517 | } while (0) |
| 518 | #else |
| 519 | # define DECL_MULK_ARM(var) |
| 520 | # define PICK_MULK_ARM(var) do ; while (0) |
| 521 | #endif |
| 522 | |
| 523 | #if CPUFAM_ARM64 |
| 524 | # define DECL_MULK_ARM64(var) \ |
| 525 | extern gcm_mulk_##var##__functype gcm_mulk_##var##_arm64_pmull; |
| 526 | # define PICK_MULK_ARM64(var) do { \ |
| 527 | DISPATCH_PICK_COND(gcm_mulk_##var, gcm_mulk_##var##_arm64_pmull, \ |
| 528 | cpu_feature_p(CPUFEAT_ARM_PMULL)); \ |
| 529 | } while (0) |
| 530 | #else |
| 531 | # define DECL_MULK_ARM64(var) |
| 532 | # define PICK_MULK_ARM64(var) do ; while (0) |
| 533 | #endif |
| 534 | |
| 535 | #define DEF_MULK(nbits) \ |
| 536 | \ |
| 537 | CPU_DISPATCH(EMPTY, EMPTY, void, gcm_mulk_##nbits##b, \ |
| 538 | (uint32 *a, const uint32 *ktab), (a, ktab), \ |
| 539 | pick_mulk_##nbits##b, simple_mulk_##nbits) \ |
| 540 | CPU_DISPATCH(EMPTY, EMPTY, void, gcm_mulk_##nbits##l, \ |
| 541 | (uint32 *a, const uint32 *ktab), (a, ktab), \ |
| 542 | pick_mulk_##nbits##l, simple_mulk_##nbits) \ |
| 543 | \ |
| 544 | static void simple_mulk_##nbits(uint32 *a, const uint32 *ktab) \ |
| 545 | { \ |
| 546 | uint32 m, t; \ |
| 547 | uint32 z[nbits/32]; \ |
| 548 | unsigned i, j, k; \ |
| 549 | \ |
| 550 | for (i = 0; i < nbits/32; i++) z[i] = 0; \ |
| 551 | \ |
| 552 | for (i = 0; i < nbits/32; i++) { \ |
| 553 | t = a[i]; \ |
| 554 | for (j = 0; j < 32; j++) { \ |
| 555 | m = -((t >> 31)&1u); \ |
| 556 | for (k = 0; k < nbits/32; k++) z[k] ^= *ktab++&m; \ |
| 557 | t <<= 1; \ |
| 558 | } \ |
| 559 | } \ |
| 560 | \ |
| 561 | for (i = 0; i < nbits/32; i++) a[i] = z[i]; \ |
| 562 | } \ |
| 563 | \ |
| 564 | DECL_MULK_X86ISH(nbits##b) \ |
| 565 | DECL_MULK_ARM(nbits##b) \ |
| 566 | DECL_MULK_ARM64(nbits##b) \ |
| 567 | static gcm_mulk_##nbits##b##__functype *pick_mulk_##nbits##b(void) \ |
| 568 | { \ |
| 569 | PICK_MULK_X86ISH(nbits##b); \ |
| 570 | PICK_MULK_ARM(nbits##b); \ |
| 571 | PICK_MULK_ARM64(nbits##b); \ |
| 572 | DISPATCH_PICK_FALLBACK(gcm_mulk_##nbits##b, simple_mulk_##nbits); \ |
| 573 | } \ |
| 574 | \ |
| 575 | DECL_MULK_X86ISH(nbits##l) \ |
| 576 | DECL_MULK_ARM(nbits##l) \ |
| 577 | DECL_MULK_ARM64(nbits##l) \ |
| 578 | static gcm_mulk_##nbits##l##__functype *pick_mulk_##nbits##l(void) \ |
| 579 | { \ |
| 580 | PICK_MULK_X86ISH(nbits##l); \ |
| 581 | PICK_MULK_ARM(nbits##l); \ |
| 582 | PICK_MULK_ARM64(nbits##l); \ |
| 583 | DISPATCH_PICK_FALLBACK(gcm_mulk_##nbits##l, simple_mulk_##nbits); \ |
| 584 | } |
| 585 | |
| 586 | GCM_WIDTHS(DEF_MULK) |
| 587 | |
| 588 | #define GCM_MULK_CASE(nbits) \ |
| 589 | case nbits/32: \ |
| 590 | if (_f&GCMF_SWAP) gcm_mulk_##nbits##l(_a, _ktab); \ |
| 591 | else gcm_mulk_##nbits##b(_a, _ktab); \ |
| 592 | break; |
| 593 | #define MULK(n, f, a, ktab) do { \ |
| 594 | uint32 *_a = (a); const uint32 *_ktab = (ktab); \ |
| 595 | unsigned _f = (f); \ |
| 596 | switch (n) { \ |
| 597 | GCM_WIDTHS(GCM_MULK_CASE) \ |
| 598 | default: abort(); \ |
| 599 | } \ |
| 600 | } while (0) |
| 601 | |
| 602 | /*----- Other utilities ---------------------------------------------------*/ |
| 603 | |
| 604 | /* --- @putlen@ --- * |
| 605 | * |
| 606 | * Arguments: @octet *p@ = pointer to output buffer |
| 607 | * @unsigned w@ = size of output buffer |
| 608 | * @unsigned blksz@ = block size (assumed fairly small) |
| 609 | * @unsigned long nblocks@ = number of blocks |
| 610 | * @unsigned nbytes@ = tail size in bytes (assumed small) |
| 611 | * |
| 612 | * Returns: --- |
| 613 | * |
| 614 | * Use: Store the overall length in %$\emph{bits}$% (i.e., |
| 615 | * @3*(nblocks*blksz + nbytes)@ in big-endian form in the |
| 616 | * buffer @p@. |
| 617 | */ |
| 618 | |
| 619 | static void putlen(octet *p, unsigned w, unsigned blksz, |
| 620 | unsigned long nblocks, unsigned nbytes) |
| 621 | { |
| 622 | unsigned long nblo = nblocks&((1ul << (ULONG_BITS/2)) - 1), |
| 623 | nbhi = nblocks >> ULONG_BITS/2; |
| 624 | unsigned long nlo = nblo*blksz + nbytes, nhi = nbhi*blksz; |
| 625 | |
| 626 | /* This is fiddly. Split @nblocks@, which is the big number, into high and |
| 627 | * low halves, multiply those separately by @blksz@, propagate carries, and |
| 628 | * then multiply by eight. |
| 629 | */ |
| 630 | nhi += nlo >> ULONG_BITS/2; |
| 631 | nlo &= (1ul << (ULONG_BITS/2)) - 1; |
| 632 | nlo <<= 3; |
| 633 | |
| 634 | /* Now write out the size, feeding bits in from @nhi@ as necessary. */ |
| 635 | p += w; |
| 636 | while (w--) { |
| 637 | *--p = U8(nlo); |
| 638 | nlo = (nlo >> 8) | ((nhi&0xff) << (ULONG_BITS/2 - 5)); |
| 639 | nhi >>= 8; |
| 640 | } |
| 641 | } |
| 642 | |
| 643 | /* --- @mix@ --- * |
| 644 | * |
| 645 | * Arguments: @const gcm_params *p@ = pointer to the parameters |
| 646 | * @uint32 *a@ = GHASH accumulator |
| 647 | * @const octet *q@ = pointer to an input block |
| 648 | * @const uint32 *ktab@ = multiplication table, built by |
| 649 | * @gcm_mktable@ |
| 650 | * |
| 651 | * Returns: --- |
| 652 | * |
| 653 | * Use: Fold the block @q@ into the GHASH accumulator. The |
| 654 | * calculation is %$a' = k (a + q)$%. |
| 655 | */ |
| 656 | |
| 657 | static void mix(const gcm_params *p, uint32 *a, |
| 658 | const octet *q, const uint32 *ktab) |
| 659 | { |
| 660 | unsigned i; |
| 661 | |
| 662 | if (p->f&GCMF_SWAP) |
| 663 | for (i = 0; i < p->n; i++) { a[i] ^= LOAD32_L(q); q += 4; } |
| 664 | else |
| 665 | for (i = 0; i < p->n; i++) { a[i] ^= LOAD32_B(q); q += 4; } |
| 666 | MULK(p->n, p->f, a, ktab); |
| 667 | } |
| 668 | |
| 669 | /* --- @gcm_ghashdone@ --- * |
| 670 | * |
| 671 | * Arguments: @const gcm_params *p@ = pointer to the parameters |
| 672 | * @uint32 *a@ = GHASH accumulator |
| 673 | * @const uint32 *ktab@ = multiplication table, built by |
| 674 | * @gcm_mktable@ |
| 675 | * @unsigned long xblocks, yblocks@ = number of whole blocks in |
| 676 | * the two inputs |
| 677 | * @unsigned xbytes, ybytes@ = number of trailing bytes in the |
| 678 | * two inputs |
| 679 | * |
| 680 | * Returns: --- |
| 681 | * |
| 682 | * Use: Finishes a GHASH operation by appending the appropriately |
| 683 | * encoded lengths of the two constituent messages. |
| 684 | */ |
| 685 | |
| 686 | void gcm_ghashdone(const gcm_params *p, uint32 *a, const uint32 *ktab, |
| 687 | unsigned long xblocks, unsigned xbytes, |
| 688 | unsigned long yblocks, unsigned ybytes) |
| 689 | { |
| 690 | octet b[4*GCM_NMAX]; |
| 691 | unsigned w = p->n < 3 ? 4*p->n : 2*p->n; |
| 692 | |
| 693 | /* Construct the encoded lengths. Note that smaller-block versions of GCM |
| 694 | * encode the lengths in separate blocks. GCM is only officially defined |
| 695 | * for 64- and 128-bit blocks; I've placed the cutoff somewhat arbitrarily |
| 696 | * at 96 bits. |
| 697 | */ |
| 698 | putlen(b, w, 4*p->n, xblocks, xbytes); |
| 699 | putlen(b + w, w, 4*p->n, yblocks, ybytes); |
| 700 | |
| 701 | /* Feed the lengths into the accumulator. */ |
| 702 | mix(p, a, b, ktab); |
| 703 | if (p->n < 3) mix(p, a, b + w, ktab); |
| 704 | } |
| 705 | |
| 706 | /* --- @gcm_concat@ --- * |
| 707 | * |
| 708 | * Arguments: @const gcm_params *p@ = pointer to the parameters |
| 709 | * @uint32 *z@ = GHASH accumulator for suffix, updated |
| 710 | * @const uint32 *x@ = GHASH accumulator for prefix |
| 711 | * @const uint32 *ktab@ = multiplication table, built by |
| 712 | * @gcm_mktable@ |
| 713 | * @unsigned long n@ = length of suffix in whole blocks |
| 714 | * |
| 715 | * Returns: --- |
| 716 | * |
| 717 | * Use: On entry, @x@ and @z@ are the results of hashing two strings |
| 718 | * %$a$% and %$b$%, each a whole number of blocks long; in |
| 719 | * particular, %$b$% is @n@ blocks long. On exit, @z@ is |
| 720 | * updated to be the hash of %$a \cat b$%. |
| 721 | */ |
| 722 | |
| 723 | void gcm_concat(const gcm_params *p, uint32 *z, const uint32 *x, |
| 724 | const uint32 *ktab, unsigned long n) |
| 725 | { |
| 726 | uint32 t[GCM_NMAX], u[GCM_NMAX]; |
| 727 | unsigned i, j; |
| 728 | |
| 729 | if (!n) { |
| 730 | /* If @n@ is zero, then there's not much to do. The mathematics |
| 731 | * (explained below) still works, but the code takes a shortcut which |
| 732 | * doesn't handle this case: so set %$z' = z + x k^n = z + x$%. |
| 733 | */ |
| 734 | |
| 735 | for (j = 0; j < p->n; j++) z[j] ^= x[j]; |
| 736 | } else { |
| 737 | /* We have %$x = a_0 t^m + \cdots + a_{m-2} t^2 + a_{m-1} t$% and |
| 738 | * %$z = b_0 t^n + \cdots + b_{n-2} t^2 + b_{n-1} t$%. What we'd like is |
| 739 | * the hash of %$a \cat b$%, which is %$z + x k^n$%. |
| 740 | * |
| 741 | * The first job, then, is to calculate %$k^n$%, and for this we use a |
| 742 | * simple left-to-right square-and-multiply algorithm. There's no need |
| 743 | * to keep %$n$% secret here. |
| 744 | */ |
| 745 | |
| 746 | /* Start by retrieving %$k$% from the table, and convert it to big-endian |
| 747 | * form. |
| 748 | */ |
| 749 | recover_k(p, u, ktab); |
| 750 | |
| 751 | /* Now calculate %$k^n$%. */ |
| 752 | i = ULONG_BITS; |
| 753 | #define BIT (1ul << (ULONG_BITS - 1)) |
| 754 | while (!(n&BIT)) { n <<= 1; i--; } |
| 755 | n <<= 1; i--; for (j = 0; j < p->n; j++) t[j] = u[j]; |
| 756 | while (i--) { mul(p, t, t, t); if (n&BIT) mul(p, t, t, u); n <<= 1; } |
| 757 | #undef BIT |
| 758 | |
| 759 | /* Next, calculate %$x k^n$%. If we're using a little-endian convention |
| 760 | * then we must convert %$x$%; otherwise we can just use it in place. |
| 761 | */ |
| 762 | if (!(p->f&GCMF_SWAP)) |
| 763 | mul(p, t, t, x); |
| 764 | else { |
| 765 | for (j = 0; j < p->n; j++) u[j] = ENDSWAP32(x[j]); |
| 766 | mul(p, t, t, u); |
| 767 | } |
| 768 | |
| 769 | /* Finally, add %$x k^n$% onto %$z$%, converting back to little-endian if |
| 770 | * necessary. |
| 771 | */ |
| 772 | if (!(p->f&GCMF_SWAP)) for (j = 0; j < p->n; j++) z[j] ^= t[j]; |
| 773 | else for (j = 0; j < p->n; j++) z[j] ^= ENDSWAP32(t[j]); |
| 774 | } |
| 775 | } |
| 776 | |
| 777 | /*----- Test rig ----------------------------------------------------------*/ |
| 778 | |
| 779 | #ifdef TEST_RIG |
| 780 | |
| 781 | #include <mLib/macros.h> |
| 782 | #include <mLib/quis.h> |
| 783 | #include <mLib/testrig.h> |
| 784 | |
| 785 | #ifdef ENABLE_ASM_DEBUG |
| 786 | # include "regdump.h" |
| 787 | #endif |
| 788 | |
| 789 | static void report_failure(const char *test, unsigned nbits, |
| 790 | const char *ref, dstr v[], dstr *d) |
| 791 | { |
| 792 | printf("test %s failed (nbits = %u)", test, nbits); |
| 793 | printf("\n\tx = "); type_hex.dump(&v[0], stdout); |
| 794 | printf("\n\ty = "); type_hex.dump(&v[1], stdout); |
| 795 | printf("\n\tz = "); type_hex.dump(&v[2], stdout); |
| 796 | printf("\n\t%s' = ", ref); type_hex.dump(d, stdout); |
| 797 | putchar('\n'); |
| 798 | } |
| 799 | |
| 800 | static void mulk(unsigned nbits, unsigned f, uint32 *x, const uint32 *ktab) |
| 801 | { MULK(nbits/32, f, x, ktab); } |
| 802 | |
| 803 | static int test_mul(uint32 poly, dstr v[]) |
| 804 | { |
| 805 | uint32 x[GCM_NMAX], y[GCM_NMAX], z[GCM_NMAX], ktab[32*GCM_NMAX*GCM_NMAX]; |
| 806 | gcm_params p; |
| 807 | dstr d = DSTR_INIT; |
| 808 | unsigned i, nbits; |
| 809 | int ok = 1; |
| 810 | enum { I_x, I_y, I_z }; |
| 811 | |
| 812 | nbits = 8*v[0].len; p.f = 0; p.n = nbits/32; p.poly = poly; |
| 813 | dstr_ensure(&d, nbits/8); d.len = nbits/8; |
| 814 | |
| 815 | #define LOADXY(E) do { \ |
| 816 | for (i = 0; i < nbits/32; i++) { \ |
| 817 | x[i] = LOAD32_##E(v[I_x].buf + 4*i); \ |
| 818 | y[i] = LOAD32_##E(v[I_y].buf + 4*i); \ |
| 819 | } \ |
| 820 | } while (0) |
| 821 | |
| 822 | #define INITZ(x) do { \ |
| 823 | for (i = 0; i < nbits/32; i++) z[i] = (x)[i]; \ |
| 824 | } while (0) |
| 825 | |
| 826 | #define CHECK(E, what, ref) do { \ |
| 827 | for (i = 0; i < nbits/32; i++) STORE32_##E(d.buf + 4*i, z[i]); \ |
| 828 | if (MEMCMP(d.buf, !=, v[I_##ref].buf, nbits/8)) \ |
| 829 | { ok = 0; report_failure(what, nbits, #ref, v, &d); } \ |
| 830 | } while (0) |
| 831 | |
| 832 | #define TEST_PREP_1(E, x, y, what) do { \ |
| 833 | gcm_mktable(&p, ktab, y); \ |
| 834 | recover_k(&p, z, ktab); CHECK(B, "mktable/recover_k (" #y ")", y); \ |
| 835 | INITZ(x); mulk(nbits, p.f, z, ktab); CHECK(E, what " (k = " #y ")", z); \ |
| 836 | } while (0) |
| 837 | |
| 838 | #define TEST_PREP(E, what) do { \ |
| 839 | TEST_PREP_1(E, x, y, what); \ |
| 840 | TEST_PREP_1(E, y, x, what); \ |
| 841 | } while (0) |
| 842 | |
| 843 | /* First, test plain multiply. */ |
| 844 | LOADXY(B); mul(&p, z, x, y); CHECK(B, "gcm_mul", z); |
| 845 | |
| 846 | /* Next, test big-endian prepared key. */ |
| 847 | LOADXY(B); TEST_PREP(B, "gcm_kmul_b"); |
| 848 | |
| 849 | /* Finally, test little-endian prepared key. */ |
| 850 | p.f = GCMF_SWAP; LOADXY(L); |
| 851 | TEST_PREP(L, "gcm_kmul_l"); |
| 852 | |
| 853 | #undef LOADXY |
| 854 | #undef INITZ |
| 855 | #undef CHECK |
| 856 | #undef TEST_PREP_1 |
| 857 | #undef TEST_PREP |
| 858 | |
| 859 | /* All done. */ |
| 860 | return (ok); |
| 861 | } |
| 862 | |
| 863 | #define TEST(nbits) \ |
| 864 | static int test_mul_##nbits(dstr v[]) \ |
| 865 | { return (test_mul(GCM_POLY_##nbits, v)); } |
| 866 | GCM_WIDTHS(TEST) |
| 867 | #undef TEST |
| 868 | |
| 869 | static test_chunk defs[] = { |
| 870 | #define TEST(nbits) \ |
| 871 | { "gcm-mul" #nbits, test_mul_##nbits, \ |
| 872 | { &type_hex, &type_hex, &type_hex, 0 } }, |
| 873 | GCM_WIDTHS(TEST) |
| 874 | #undef TEST |
| 875 | { 0, 0, { 0 } } |
| 876 | }; |
| 877 | |
| 878 | int main(int argc, char *argv[]) |
| 879 | { |
| 880 | ego(argv[0]); |
| 881 | #ifdef ENABLE_ASM_DEBUG |
| 882 | regdump_init(); |
| 883 | #endif |
| 884 | test_run(argc, argv, defs, SRCDIR"/t/gcm"); |
| 885 | return (0); |
| 886 | } |
| 887 | |
| 888 | #endif |
| 889 | |
| 890 | /*----- That's all, folks -------------------------------------------------*/ |