crypto/modes/gcm128.c

   1 /* ====================================================================
   2  * Copyright (c) 2010 The OpenSSL Project.  All rights reserved.
   3  *
   4  * Redistribution and use in source and binary forms, with or without
   5  * modification, are permitted provided that the following conditions
   6  * are met:
   7  *
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  *
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in
  13  *    the documentation and/or other materials provided with the
  14  *    distribution.
  15  *
  16  * 3. All advertising materials mentioning features or use of this
  17  *    software must display the following acknowledgment:
  18  *    "This product includes software developed by the OpenSSL Project
  19  *    for use in the OpenSSL Toolkit. (http://www.openssl.org/)"
  20  *
  21  * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
  22  *    endorse or promote products derived from this software without
  23  *    prior written permission. For written permission, please contact
  24  *    openssl-core@openssl.org.
  25  *
  26  * 5. Products derived from this software may not be called "OpenSSL"
  27  *    nor may "OpenSSL" appear in their names without prior written
  28  *    permission of the OpenSSL Project.
  29  *
  30  * 6. Redistributions of any form whatsoever must retain the following
  31  *    acknowledgment:
  32  *    "This product includes software developed by the OpenSSL Project
  33  *    for use in the OpenSSL Toolkit (http://www.openssl.org/)"
  34  *
  35  * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
  36  * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  37  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  38  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE OpenSSL PROJECT OR
  39  * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  40  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  41  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  42  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  43  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  44  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  45  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
  46  * OF THE POSSIBILITY OF SUCH DAMAGE.
  47  * ====================================================================
  48  */
  49
  50 #include <openssl/crypto.h>
  51 #include "modes_lcl.h"
  52 #include <string.h>
  53
  54 #ifndef MODES_DEBUG
  55 # ifndef NDEBUG
  56 #  define NDEBUG
  57 # endif
  58 #endif
  59 #include <assert.h>
  60
  61 typedef struct { u64 hi,lo; } u128;
  62
  63 #if defined(BSWAP4) && defined(STRICT_ALIGNMENT)
  64 /* redefine, because alignment is ensured */
  65 #undef  GETU32
  66 #define GETU32(p)       BSWAP4(*(const u32 *)(p))
  67 #undef  PUTU32
  68 #define PUTU32(p,v)     *(u32 *)(p) = BSWAP4(v)
  69 #endif
  70
  71 #define PACK(s)         ((size_t)(s)<<(sizeof(size_t)*8-16))
  72 #define REDUCE1BIT(V)   do { \
  73         if (sizeof(size_t)==8) { \
  74                 u64 T = U64(0xe100000000000000) & (0-(V.lo&1)); \
  75                 V.lo  = (V.hi<<63)|(V.lo>>1); \
  76                 V.hi  = (V.hi>>1 )^T; \
  77         } \
  78         else { \
  79                 u32 T = 0xe1000000U & (0-(u32)(V.lo&1)); \
  80                 V.lo  = (V.hi<<63)|(V.lo>>1); \
  81                 V.hi  = (V.hi>>1 )^((u64)T<<32); \
  82         } \
  83 } while(0)
  84
  85 #ifdef  TABLE_BITS
  86 #undef  TABLE_BITS
  87 #endif
  88 /*
  89  * Even though permitted values for TABLE_BITS are 8, 4 and 1, it should
  90  * never be set to 8. 8 is effectively reserved for testing purposes.
  91  * TABLE_BITS>1 are lookup-table-driven implementations referred to as
  92  * "Shoup's" in GCM specification. In other words OpenSSL does not cover
  93  * whole spectrum of possible table driven implementations. Why? In
  94  * non-"Shoup's" case memory access pattern is segmented in such manner,
  95  * that it's trivial to see that cache timing information can reveal
  96  * fair portion of intermediate hash value. Given that ciphertext is
  97  * always available to attacker, it's possible for him to attempt to
  98  * deduce secret parameter H and if successful, tamper with messages
  99  * [which is nothing but trivial in CTR mode]. In "Shoup's" case it's
 100  * not as trivial, but there is no reason to believe that it's resistant
 101  * to cache-timing attack. And the thing about "8-bit" implementation is
 102  * that it consumes 16 (sixteen) times more memory, 4KB per individual
 103  * key + 1KB shared. Well, on pros side it should be twice as fast as
 104  * "4-bit" version. And for gcc-generated x86[_64] code, "8-bit" version
 105  * was observed to run ~75% faster, closer to 100% for commercial
 106  * compilers... Yet "4-bit" procedure is preferred, because it's
 107  * believed to provide better security-performance balance and adequate
 108  * all-round performance. "All-round" refers to things like:
 109  *
 110  * - shorter setup time effectively improves overall timing for
 111  *   handling short messages;
 112  * - larger table allocation can become unbearable because of VM
 113  *   subsystem penalties (for example on Windows large enough free
 114  *   results in VM working set trimming, meaning that consequent
 115  *   malloc would immediately incur working set expansion);
 116  * - larger table has larger cache footprint, which can affect
 117  *   performance of other code paths (not necessarily even from same
 118  *   thread in Hyper-Threading world);
 119  */
 120 #define TABLE_BITS 4
 121
 122 #if     TABLE_BITS==8
 123
 124 static void gcm_init_8bit(u128 Htable[256], u64 H[2])
 125 {
 126         int  i, j;
 127         u128 V;
 128
 129         Htable[0].hi = 0;
 130         Htable[0].lo = 0;
 131         V.hi = H[0];
 132         V.lo = H[1];
 133
 134         for (Htable[128]=V, i=64; i>0; i>>=1) {
 135                 REDUCE1BIT(V);
 136                 Htable[i] = V;
 137         }
 138
 139         for (i=2; i<256; i<<=1) {
 140                 u128 *Hi = Htable+i, H0 = *Hi;
 141                 for (j=1; j<i; ++j) {
 142                         Hi[j].hi = H0.hi^Htable[j].hi;
 143                         Hi[j].lo = H0.lo^Htable[j].lo;
 144                 }
 145         }
 146 }
 147
 148 static void gcm_gmult_8bit(u64 Xi[2], u128 Htable[256])
 149 {
 150         u128 Z = { 0, 0};
 151         const u8 *xi = (const u8 *)Xi+15;
 152         size_t rem, n = *xi;
 153         const union { long one; char little; } is_endian = {1};
 154         static const size_t rem_8bit[256] = {
 155                 PACK(0x0000), PACK(0x01C2), PACK(0x0384), PACK(0x0246),
 156                 PACK(0x0708), PACK(0x06CA), PACK(0x048C), PACK(0x054E),
 157                 PACK(0x0E10), PACK(0x0FD2), PACK(0x0D94), PACK(0x0C56),
 158                 PACK(0x0918), PACK(0x08DA), PACK(0x0A9C), PACK(0x0B5E),
 159                 PACK(0x1C20), PACK(0x1DE2), PACK(0x1FA4), PACK(0x1E66),
 160                 PACK(0x1B28), PACK(0x1AEA), PACK(0x18AC), PACK(0x196E),
 161                 PACK(0x1230), PACK(0x13F2), PACK(0x11B4), PACK(0x1076),
 162                 PACK(0x1538), PACK(0x14FA), PACK(0x16BC), PACK(0x177E),
 163                 PACK(0x3840), PACK(0x3982), PACK(0x3BC4), PACK(0x3A06),
 164                 PACK(0x3F48), PACK(0x3E8A), PACK(0x3CCC), PACK(0x3D0E),
 165                 PACK(0x3650), PACK(0x3792), PACK(0x35D4), PACK(0x3416),
 166                 PACK(0x3158), PACK(0x309A), PACK(0x32DC), PACK(0x331E),
 167                 PACK(0x2460), PACK(0x25A2), PACK(0x27E4), PACK(0x2626),
 168                 PACK(0x2368), PACK(0x22AA), PACK(0x20EC), PACK(0x212E),
 169                 PACK(0x2A70), PACK(0x2BB2), PACK(0x29F4), PACK(0x2836),
 170                 PACK(0x2D78), PACK(0x2CBA), PACK(0x2EFC), PACK(0x2F3E),
 171                 PACK(0x7080), PACK(0x7142), PACK(0x7304), PACK(0x72C6),
 172                 PACK(0x7788), PACK(0x764A), PACK(0x740C), PACK(0x75CE),
 173                 PACK(0x7E90), PACK(0x7F52), PACK(0x7D14), PACK(0x7CD6),
 174                 PACK(0x7998), PACK(0x785A), PACK(0x7A1C), PACK(0x7BDE),
 175                 PACK(0x6CA0), PACK(0x6D62), PACK(0x6F24), PACK(0x6EE6),
 176                 PACK(0x6BA8), PACK(0x6A6A), PACK(0x682C), PACK(0x69EE),
 177                 PACK(0x62B0), PACK(0x6372), PACK(0x6134), PACK(0x60F6),
 178                 PACK(0x65B8), PACK(0x647A), PACK(0x663C), PACK(0x67FE),
 179                 PACK(0x48C0), PACK(0x4902), PACK(0x4B44), PACK(0x4A86),
 180                 PACK(0x4FC8), PACK(0x4E0A), PACK(0x4C4C), PACK(0x4D8E),
 181                 PACK(0x46D0), PACK(0x4712), PACK(0x4554), PACK(0x4496),
 182                 PACK(0x41D8), PACK(0x401A), PACK(0x425C), PACK(0x439E),
 183                 PACK(0x54E0), PACK(0x5522), PACK(0x5764), PACK(0x56A6),
 184                 PACK(0x53E8), PACK(0x522A), PACK(0x506C), PACK(0x51AE),
 185                 PACK(0x5AF0), PACK(0x5B32), PACK(0x5974), PACK(0x58B6),
 186                 PACK(0x5DF8), PACK(0x5C3A), PACK(0x5E7C), PACK(0x5FBE),
 187                 PACK(0xE100), PACK(0xE0C2), PACK(0xE284), PACK(0xE346),
 188                 PACK(0xE608), PACK(0xE7CA), PACK(0xE58C), PACK(0xE44E),
 189                 PACK(0xEF10), PACK(0xEED2), PACK(0xEC94), PACK(0xED56),
 190                 PACK(0xE818), PACK(0xE9DA), PACK(0xEB9C), PACK(0xEA5E),
 191                 PACK(0xFD20), PACK(0xFCE2), PACK(0xFEA4), PACK(0xFF66),
 192                 PACK(0xFA28), PACK(0xFBEA), PACK(0xF9AC), PACK(0xF86E),
 193                 PACK(0xF330), PACK(0xF2F2), PACK(0xF0B4), PACK(0xF176),
 194                 PACK(0xF438), PACK(0xF5FA), PACK(0xF7BC), PACK(0xF67E),
 195                 PACK(0xD940), PACK(0xD882), PACK(0xDAC4), PACK(0xDB06),
 196                 PACK(0xDE48), PACK(0xDF8A), PACK(0xDDCC), PACK(0xDC0E),
 197                 PACK(0xD750), PACK(0xD692), PACK(0xD4D4), PACK(0xD516),
 198                 PACK(0xD058), PACK(0xD19A), PACK(0xD3DC), PACK(0xD21E),
 199                 PACK(0xC560), PACK(0xC4A2), PACK(0xC6E4), PACK(0xC726),
 200                 PACK(0xC268), PACK(0xC3AA), PACK(0xC1EC), PACK(0xC02E),
 201                 PACK(0xCB70), PACK(0xCAB2), PACK(0xC8F4), PACK(0xC936),
 202                 PACK(0xCC78), PACK(0xCDBA), PACK(0xCFFC), PACK(0xCE3E),
 203                 PACK(0x9180), PACK(0x9042), PACK(0x9204), PACK(0x93C6),
 204                 PACK(0x9688), PACK(0x974A), PACK(0x950C), PACK(0x94CE),
 205                 PACK(0x9F90), PACK(0x9E52), PACK(0x9C14), PACK(0x9DD6),
 206                 PACK(0x9898), PACK(0x995A), PACK(0x9B1C), PACK(0x9ADE),
 207                 PACK(0x8DA0), PACK(0x8C62), PACK(0x8E24), PACK(0x8FE6),
 208                 PACK(0x8AA8), PACK(0x8B6A), PACK(0x892C), PACK(0x88EE),
 209                 PACK(0x83B0), PACK(0x8272), PACK(0x8034), PACK(0x81F6),
 210                 PACK(0x84B8), PACK(0x857A), PACK(0x873C), PACK(0x86FE),
 211                 PACK(0xA9C0), PACK(0xA802), PACK(0xAA44), PACK(0xAB86),
 212                 PACK(0xAEC8), PACK(0xAF0A), PACK(0xAD4C), PACK(0xAC8E),
 213                 PACK(0xA7D0), PACK(0xA612), PACK(0xA454), PACK(0xA596),
 214                 PACK(0xA0D8), PACK(0xA11A), PACK(0xA35C), PACK(0xA29E),
 215                 PACK(0xB5E0), PACK(0xB422), PACK(0xB664), PACK(0xB7A6),
 216                 PACK(0xB2E8), PACK(0xB32A), PACK(0xB16C), PACK(0xB0AE),
 217                 PACK(0xBBF0), PACK(0xBA32), PACK(0xB874), PACK(0xB9B6),
 218                 PACK(0xBCF8), PACK(0xBD3A), PACK(0xBF7C), PACK(0xBEBE) };
 219
 220         while (1) {
 221                 Z.hi ^= Htable[n].hi;
 222                 Z.lo ^= Htable[n].lo;
 223
 224                 if ((u8 *)Xi==xi)       break;
 225
 226                 n = *(--xi);
 227
 228                 rem  = (size_t)Z.lo&0xff;
 229                 Z.lo = (Z.hi<<56)|(Z.lo>>8);
 230                 Z.hi = (Z.hi>>8);
 231                 if (sizeof(size_t)==8)
 232                         Z.hi ^= rem_8bit[rem];
 233                 else
 234                         Z.hi ^= (u64)rem_8bit[rem]<<32;
 235         }
 236
 237         if (is_endian.little) {
 238 #ifdef BSWAP8
 239                 Xi[0] = BSWAP8(Z.hi);
 240                 Xi[1] = BSWAP8(Z.lo);
 241 #else
 242                 u8 *p = (u8 *)Xi;
 243                 u32 v;
 244                 v = (u32)(Z.hi>>32);    PUTU32(p,v);
 245                 v = (u32)(Z.hi);        PUTU32(p+4,v);
 246                 v = (u32)(Z.lo>>32);    PUTU32(p+8,v);
 247                 v = (u32)(Z.lo);        PUTU32(p+12,v);
 248 #endif
 249         }
 250         else {
 251                 Xi[0] = Z.hi;
 252                 Xi[1] = Z.lo;
 253         }
 254 }
 255 #define GCM_MUL(ctx,Xi)   gcm_gmult_8bit(ctx->Xi.u,ctx->Htable)
 256
 257 #elif   TABLE_BITS==4
 258
 259 static void gcm_init_4bit(u128 Htable[16], u64 H[2])
 260 {
 261         u128 V;
 262 #if defined(OPENSSL_SMALL_FOOTPRINT)
 263         int  i;
 264 #endif
 265
 266         Htable[0].hi = 0;
 267         Htable[0].lo = 0;
 268         V.hi = H[0];
 269         V.lo = H[1];
 270
 271 #if defined(OPENSSL_SMALL_FOOTPRINT)
 272         for (Htable[8]=V, i=4; i>0; i>>=1) {
 273                 REDUCE1BIT(V);
 274                 Htable[i] = V;
 275         }
 276
 277         for (i=2; i<16; i<<=1) {
 278                 u128 *Hi = Htable+i;
 279                 int   j;
 280                 for (V=*Hi, j=1; j<i; ++j) {
 281                         Hi[j].hi = V.hi^Htable[j].hi;
 282                         Hi[j].lo = V.lo^Htable[j].lo;
 283                 }
 284         }
 285 #else
 286         Htable[8] = V;
 287         REDUCE1BIT(V);
 288         Htable[4] = V;
 289         REDUCE1BIT(V);
 290         Htable[2] = V;
 291         REDUCE1BIT(V);
 292         Htable[1] = V;
 293         Htable[3].hi  = V.hi^Htable[2].hi, Htable[3].lo  = V.lo^Htable[2].lo;
 294         V=Htable[4];
 295         Htable[5].hi  = V.hi^Htable[1].hi, Htable[5].lo  = V.lo^Htable[1].lo;
 296         Htable[6].hi  = V.hi^Htable[2].hi, Htable[6].lo  = V.lo^Htable[2].lo;
 297         Htable[7].hi  = V.hi^Htable[3].hi, Htable[7].lo  = V.lo^Htable[3].lo;
 298         V=Htable[8];
 299         Htable[9].hi  = V.hi^Htable[1].hi, Htable[9].lo  = V.lo^Htable[1].lo;
 300         Htable[10].hi = V.hi^Htable[2].hi, Htable[10].lo = V.lo^Htable[2].lo;
 301         Htable[11].hi = V.hi^Htable[3].hi, Htable[11].lo = V.lo^Htable[3].lo;
 302         Htable[12].hi = V.hi^Htable[4].hi, Htable[12].lo = V.lo^Htable[4].lo;
 303         Htable[13].hi = V.hi^Htable[5].hi, Htable[13].lo = V.lo^Htable[5].lo;
 304         Htable[14].hi = V.hi^Htable[6].hi, Htable[14].lo = V.lo^Htable[6].lo;
 305         Htable[15].hi = V.hi^Htable[7].hi, Htable[15].lo = V.lo^Htable[7].lo;
 306 #endif
 307 #if defined(GHASH_ASM) && (defined(__arm__) || defined(__arm))
 308         /*
 309          * ARM assembler expects specific dword order in Htable.
 310          */
 311         {
 312         int j;
 313         const union { long one; char little; } is_endian = {1};
 314
 315         if (is_endian.little)
 316                 for (j=0;j<16;++j) {
 317                         V = Htable[j];
 318                         Htable[j].hi = V.lo;
 319                         Htable[j].lo = V.hi;
 320                 }
 321         else
 322                 for (j=0;j<16;++j) {
 323                         V = Htable[j];
 324                         Htable[j].hi = V.lo<<32|V.lo>>32;
 325                         Htable[j].lo = V.hi<<32|V.hi>>32;
 326                 }
 327         }
 328 #endif
 329 }
 330
 331 #ifndef GHASH_ASM
 332 static const size_t rem_4bit[16] = {
 333         PACK(0x0000), PACK(0x1C20), PACK(0x3840), PACK(0x2460),
 334         PACK(0x7080), PACK(0x6CA0), PACK(0x48C0), PACK(0x54E0),
 335         PACK(0xE100), PACK(0xFD20), PACK(0xD940), PACK(0xC560),
 336         PACK(0x9180), PACK(0x8DA0), PACK(0xA9C0), PACK(0xB5E0) };
 337
 338 static void gcm_gmult_4bit(u64 Xi[2], const u128 Htable[16])
 339 {
 340         u128 Z;
 341         int cnt = 15;
 342         size_t rem, nlo, nhi;
 343         const union { long one; char little; } is_endian = {1};
 344
 345         nlo  = ((const u8 *)Xi)[15];
 346         nhi  = nlo>>4;
 347         nlo &= 0xf;
 348
 349         Z.hi = Htable[nlo].hi;
 350         Z.lo = Htable[nlo].lo;
 351
 352         while (1) {
 353                 rem  = (size_t)Z.lo&0xf;
 354                 Z.lo = (Z.hi<<60)|(Z.lo>>4);
 355                 Z.hi = (Z.hi>>4);
 356                 if (sizeof(size_t)==8)
 357                         Z.hi ^= rem_4bit[rem];
 358                 else
 359                         Z.hi ^= (u64)rem_4bit[rem]<<32;
 360
 361                 Z.hi ^= Htable[nhi].hi;
 362                 Z.lo ^= Htable[nhi].lo;
 363
 364                 if (--cnt<0)            break;
 365
 366                 nlo  = ((const u8 *)Xi)[cnt];
 367                 nhi  = nlo>>4;
 368                 nlo &= 0xf;
 369
 370                 rem  = (size_t)Z.lo&0xf;
 371                 Z.lo = (Z.hi<<60)|(Z.lo>>4);
 372                 Z.hi = (Z.hi>>4);
 373                 if (sizeof(size_t)==8)
 374                         Z.hi ^= rem_4bit[rem];
 375                 else
 376                         Z.hi ^= (u64)rem_4bit[rem]<<32;
 377
 378                 Z.hi ^= Htable[nlo].hi;
 379                 Z.lo ^= Htable[nlo].lo;
 380         }
 381
 382         if (is_endian.little) {
 383 #ifdef BSWAP8
 384                 Xi[0] = BSWAP8(Z.hi);
 385                 Xi[1] = BSWAP8(Z.lo);
 386 #else
 387                 u8 *p = (u8 *)Xi;
 388                 u32 v;
 389                 v = (u32)(Z.hi>>32);    PUTU32(p,v);
 390                 v = (u32)(Z.hi);        PUTU32(p+4,v);
 391                 v = (u32)(Z.lo>>32);    PUTU32(p+8,v);
 392                 v = (u32)(Z.lo);        PUTU32(p+12,v);
 393 #endif
 394         }
 395         else {
 396                 Xi[0] = Z.hi;
 397                 Xi[1] = Z.lo;
 398         }
 399 }
 400
 401 #if !defined(OPENSSL_SMALL_FOOTPRINT)
 402 /*
 403  * Streamed gcm_mult_4bit, see CRYPTO_gcm128_[en|de]crypt for
 404  * details... Compiler-generated code doesn't seem to give any
 405  * performance improvement, at least not on x86[_64]. It's here
 406  * mostly as reference and a placeholder for possible future
 407  * non-trivial optimization[s]...
 408  */
 409 static void gcm_ghash_4bit(u64 Xi[2],const u128 Htable[16],
 410                                 const u8 *inp,size_t len)
 411 {
 412     u128 Z;
 413     int cnt;
 414     size_t rem, nlo, nhi;
 415     const union { long one; char little; } is_endian = {1};
 416
 417 #if 1
 418     do {
 419         cnt  = 15;
 420         nlo  = ((const u8 *)Xi)[15];
 421         nlo ^= inp[15];
 422         nhi  = nlo>>4;
 423         nlo &= 0xf;
 424
 425         Z.hi = Htable[nlo].hi;
 426         Z.lo = Htable[nlo].lo;
 427
 428         while (1) {
 429                 rem  = (size_t)Z.lo&0xf;
 430                 Z.lo = (Z.hi<<60)|(Z.lo>>4);
 431                 Z.hi = (Z.hi>>4);
 432                 if (sizeof(size_t)==8)
 433                         Z.hi ^= rem_4bit[rem];
 434                 else
 435                         Z.hi ^= (u64)rem_4bit[rem]<<32;
 436
 437                 Z.hi ^= Htable[nhi].hi;
 438                 Z.lo ^= Htable[nhi].lo;
 439
 440                 if (--cnt<0)            break;
 441
 442                 nlo  = ((const u8 *)Xi)[cnt];
 443                 nlo ^= inp[cnt];
 444                 nhi  = nlo>>4;
 445                 nlo &= 0xf;
 446
 447                 rem  = (size_t)Z.lo&0xf;
 448                 Z.lo = (Z.hi<<60)|(Z.lo>>4);
 449                 Z.hi = (Z.hi>>4);
 450                 if (sizeof(size_t)==8)
 451                         Z.hi ^= rem_4bit[rem];
 452                 else
 453                         Z.hi ^= (u64)rem_4bit[rem]<<32;
 454
 455                 Z.hi ^= Htable[nlo].hi;
 456                 Z.lo ^= Htable[nlo].lo;
 457         }
 458 #else
 459     /*
 460      * Extra 256+16 bytes per-key plus 512 bytes shared tables
 461      * [should] give ~50% improvement... One could have PACK()-ed
 462      * the rem_8bit even here, but the priority is to minimize
 463      * cache footprint...
 464      */
 465     u128 Hshr4[16];     /* Htable shifted right by 4 bits */
 466     u8   Hshl4[16];     /* Htable shifted left  by 4 bits */
 467     static const unsigned short rem_8bit[256] = {
 468         0x0000, 0x01C2, 0x0384, 0x0246, 0x0708, 0x06CA, 0x048C, 0x054E,
 469         0x0E10, 0x0FD2, 0x0D94, 0x0C56, 0x0918, 0x08DA, 0x0A9C, 0x0B5E,
 470         0x1C20, 0x1DE2, 0x1FA4, 0x1E66, 0x1B28, 0x1AEA, 0x18AC, 0x196E,
 471         0x1230, 0x13F2, 0x11B4, 0x1076, 0x1538, 0x14FA, 0x16BC, 0x177E,
 472         0x3840, 0x3982, 0x3BC4, 0x3A06, 0x3F48, 0x3E8A, 0x3CCC, 0x3D0E,
 473         0x3650, 0x3792, 0x35D4, 0x3416, 0x3158, 0x309A, 0x32DC, 0x331E,
 474         0x2460, 0x25A2, 0x27E4, 0x2626, 0x2368, 0x22AA, 0x20EC, 0x212E,
 475         0x2A70, 0x2BB2, 0x29F4, 0x2836, 0x2D78, 0x2CBA, 0x2EFC, 0x2F3E,
 476         0x7080, 0x7142, 0x7304, 0x72C6, 0x7788, 0x764A, 0x740C, 0x75CE,
 477         0x7E90, 0x7F52, 0x7D14, 0x7CD6, 0x7998, 0x785A, 0x7A1C, 0x7BDE,
 478         0x6CA0, 0x6D62, 0x6F24, 0x6EE6, 0x6BA8, 0x6A6A, 0x682C, 0x69EE,
 479         0x62B0, 0x6372, 0x6134, 0x60F6, 0x65B8, 0x647A, 0x663C, 0x67FE,
 480         0x48C0, 0x4902, 0x4B44, 0x4A86, 0x4FC8, 0x4E0A, 0x4C4C, 0x4D8E,
 481         0x46D0, 0x4712, 0x4554, 0x4496, 0x41D8, 0x401A, 0x425C, 0x439E,
 482         0x54E0, 0x5522, 0x5764, 0x56A6, 0x53E8, 0x522A, 0x506C, 0x51AE,
 483         0x5AF0, 0x5B32, 0x5974, 0x58B6, 0x5DF8, 0x5C3A, 0x5E7C, 0x5FBE,
 484         0xE100, 0xE0C2, 0xE284, 0xE346, 0xE608, 0xE7CA, 0xE58C, 0xE44E,
 485         0xEF10, 0xEED2, 0xEC94, 0xED56, 0xE818, 0xE9DA, 0xEB9C, 0xEA5E,
 486         0xFD20, 0xFCE2, 0xFEA4, 0xFF66, 0xFA28, 0xFBEA, 0xF9AC, 0xF86E,
 487         0xF330, 0xF2F2, 0xF0B4, 0xF176, 0xF438, 0xF5FA, 0xF7BC, 0xF67E,
 488         0xD940, 0xD882, 0xDAC4, 0xDB06, 0xDE48, 0xDF8A, 0xDDCC, 0xDC0E,
 489         0xD750, 0xD692, 0xD4D4, 0xD516, 0xD058, 0xD19A, 0xD3DC, 0xD21E,
 490         0xC560, 0xC4A2, 0xC6E4, 0xC726, 0xC268, 0xC3AA, 0xC1EC, 0xC02E,
 491         0xCB70, 0xCAB2, 0xC8F4, 0xC936, 0xCC78, 0xCDBA, 0xCFFC, 0xCE3E,
 492         0x9180, 0x9042, 0x9204, 0x93C6, 0x9688, 0x974A, 0x950C, 0x94CE,
 493         0x9F90, 0x9E52, 0x9C14, 0x9DD6, 0x9898, 0x995A, 0x9B1C, 0x9ADE,
 494         0x8DA0, 0x8C62, 0x8E24, 0x8FE6, 0x8AA8, 0x8B6A, 0x892C, 0x88EE,
 495         0x83B0, 0x8272, 0x8034, 0x81F6, 0x84B8, 0x857A, 0x873C, 0x86FE,
 496         0xA9C0, 0xA802, 0xAA44, 0xAB86, 0xAEC8, 0xAF0A, 0xAD4C, 0xAC8E,
 497         0xA7D0, 0xA612, 0xA454, 0xA596, 0xA0D8, 0xA11A, 0xA35C, 0xA29E,
 498         0xB5E0, 0xB422, 0xB664, 0xB7A6, 0xB2E8, 0xB32A, 0xB16C, 0xB0AE,
 499         0xBBF0, 0xBA32, 0xB874, 0xB9B6, 0xBCF8, 0xBD3A, 0xBF7C, 0xBEBE };
 500     /*
 501      * This pre-processing phase slows down procedure by approximately
 502      * same time as it makes each loop spin faster. In other words
 503      * single block performance is approximately same as straightforward
 504      * "4-bit" implementation, and then it goes only faster...
 505      */
 506     for (cnt=0; cnt<16; ++cnt) {
 507         Z.hi = Htable[cnt].hi;
 508         Z.lo = Htable[cnt].lo;
 509         Hshr4[cnt].lo = (Z.hi<<60)|(Z.lo>>4);
 510         Hshr4[cnt].hi = (Z.hi>>4);
 511         Hshl4[cnt]    = (u8)(Z.lo<<4);
 512     }
 513
 514     do {
 515         for (Z.lo=0, Z.hi=0, cnt=15; cnt; --cnt) {
 516                 nlo  = ((const u8 *)Xi)[cnt];
 517                 nlo ^= inp[cnt];
 518                 nhi  = nlo>>4;
 519                 nlo &= 0xf;
 520
 521                 Z.hi ^= Htable[nlo].hi;
 522                 Z.lo ^= Htable[nlo].lo;
 523
 524                 rem = (size_t)Z.lo&0xff;
 525
 526                 Z.lo = (Z.hi<<56)|(Z.lo>>8);
 527                 Z.hi = (Z.hi>>8);
 528
 529                 Z.hi ^= Hshr4[nhi].hi;
 530                 Z.lo ^= Hshr4[nhi].lo;
 531                 Z.hi ^= (u64)rem_8bit[rem^Hshl4[nhi]]<<48;
 532         }
 533
 534         nlo  = ((const u8 *)Xi)[0];
 535         nlo ^= inp[0];
 536         nhi  = nlo>>4;
 537         nlo &= 0xf;
 538
 539         Z.hi ^= Htable[nlo].hi;
 540         Z.lo ^= Htable[nlo].lo;
 541
 542         rem = (size_t)Z.lo&0xf;
 543
 544         Z.lo = (Z.hi<<60)|(Z.lo>>4);
 545         Z.hi = (Z.hi>>4);
 546
 547         Z.hi ^= Htable[nhi].hi;
 548         Z.lo ^= Htable[nhi].lo;
 549         Z.hi ^= ((u64)rem_8bit[rem<<4])<<48;
 550 #endif
 551
 552         if (is_endian.little) {
 553 #ifdef BSWAP8
 554                 Xi[0] = BSWAP8(Z.hi);
 555                 Xi[1] = BSWAP8(Z.lo);
 556 #else
 557                 u8 *p = (u8 *)Xi;
 558                 u32 v;
 559                 v = (u32)(Z.hi>>32);    PUTU32(p,v);
 560                 v = (u32)(Z.hi);        PUTU32(p+4,v);
 561                 v = (u32)(Z.lo>>32);    PUTU32(p+8,v);
 562                 v = (u32)(Z.lo);        PUTU32(p+12,v);
 563 #endif
 564         }
 565         else {
 566                 Xi[0] = Z.hi;
 567                 Xi[1] = Z.lo;
 568         }
 569     } while (inp+=16, len-=16);
 570 }
 571 #endif
 572 #else
 573 void gcm_gmult_4bit(u64 Xi[2],const u128 Htable[16]);
 574 void gcm_ghash_4bit(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
 575 #endif
 576
 577 #define GCM_MUL(ctx,Xi)   gcm_gmult_4bit(ctx->Xi.u,ctx->Htable)
 578 #if defined(GHASH_ASM) || !defined(OPENSSL_SMALL_FOOTPRINT)
 579 #define GHASH(ctx,in,len) gcm_ghash_4bit((ctx)->Xi.u,(ctx)->Htable,in,len)
 580 /* GHASH_CHUNK is "stride parameter" missioned to mitigate cache
 581  * trashing effect. In other words idea is to hash data while it's
 582  * still in L1 cache after encryption pass... */
 583 #define GHASH_CHUNK       (3*1024)
 584 #endif
 585
 586 #else   /* TABLE_BITS */
 587
 588 static void gcm_gmult_1bit(u64 Xi[2],const u64 H[2])
 589 {
 590         u128 V,Z = { 0,0 };
 591         long X;
 592         int  i,j;
 593         const long *xi = (const long *)Xi;
 594         const union { long one; char little; } is_endian = {1};
 595
 596         V.hi = H[0];    /* H is in host byte order, no byte swapping */
 597         V.lo = H[1];
 598
 599         for (j=0; j<16/sizeof(long); ++j) {
 600                 if (is_endian.little) {
 601                         if (sizeof(long)==8) {
 602 #ifdef BSWAP8
 603                                 X = (long)(BSWAP8(xi[j]));
 604 #else
 605                                 const u8 *p = (const u8 *)(xi+j);
 606                                 X = (long)((u64)GETU32(p)<<32|GETU32(p+4));
 607 #endif
 608                         }
 609                         else {
 610                                 const u8 *p = (const u8 *)(xi+j);
 611                                 X = (long)GETU32(p);
 612                         }
 613                 }
 614                 else
 615                         X = xi[j];
 616
 617                 for (i=0; i<8*sizeof(long); ++i, X<<=1) {
 618                         u64 M = (u64)(X>>(8*sizeof(long)-1));
 619                         Z.hi ^= V.hi&M;
 620                         Z.lo ^= V.lo&M;
 621
 622                         REDUCE1BIT(V);
 623                 }
 624         }
 625
 626         if (is_endian.little) {
 627 #ifdef BSWAP8
 628                 Xi[0] = BSWAP8(Z.hi);
 629                 Xi[1] = BSWAP8(Z.lo);
 630 #else
 631                 u8 *p = (u8 *)Xi;
 632                 u32 v;
 633                 v = (u32)(Z.hi>>32);    PUTU32(p,v);
 634                 v = (u32)(Z.hi);        PUTU32(p+4,v);
 635                 v = (u32)(Z.lo>>32);    PUTU32(p+8,v);
 636                 v = (u32)(Z.lo);        PUTU32(p+12,v);
 637 #endif
 638         }
 639         else {
 640                 Xi[0] = Z.hi;
 641                 Xi[1] = Z.lo;
 642         }
 643 }
 644 #define GCM_MUL(ctx,Xi)   gcm_gmult_1bit(ctx->Xi.u,ctx->H.u)
 645
 646 #endif
 647
 648 struct gcm128_context {
 649         /* Following 6 names follow names in GCM specification */
 650         union { u64 u[2]; u32 d[4]; u8 c[16]; } Yi,EKi,EK0,
 651                                                 Xi,H,len;
 652         /* Pre-computed table used by gcm_gmult_* */
 653 #if TABLE_BITS==8
 654         u128 Htable[256];
 655 #else
 656         u128 Htable[16];
 657         void (*gmult)(u64 Xi[2],const u128 Htable[16]);
 658         void (*ghash)(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
 659 #endif
 660         unsigned int mres, ares;
 661         block128_f block;
 662         void *key;
 663 };
 664
 665 #if     TABLE_BITS==4 && defined(GHASH_ASM) && !defined(I386_ONLY) && \
 666         (defined(__i386)        || defined(__i386__)    || \
 667          defined(__x86_64)      || defined(__x86_64__)  || \
 668          defined(_M_IX86)       || defined(_M_AMD64)    || defined(_M_X64))
 669 # define GHASH_ASM_IAX
 670 extern unsigned int OPENSSL_ia32cap_P[2];
 671
 672 void gcm_init_clmul(u128 Htable[16],const u64 Xi[2]);
 673 void gcm_gmult_clmul(u64 Xi[2],const u128 Htable[16]);
 674 void gcm_ghash_clmul(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
 675
 676 # if    defined(__i386) || defined(__i386__) || defined(_M_IX86)
 677 #  define GHASH_ASM_X86
 678 void gcm_gmult_4bit_mmx(u64 Xi[2],const u128 Htable[16]);
 679 void gcm_ghash_4bit_mmx(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
 680
 681 void gcm_gmult_4bit_x86(u64 Xi[2],const u128 Htable[16]);
 682 void gcm_ghash_4bit_x86(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
 683 # endif
 684
 685 # undef  GCM_MUL
 686 # define GCM_MUL(ctx,Xi)   (*((ctx)->gmult))(ctx->Xi.u,ctx->Htable)
 687 # undef  GHASH
 688 # define GHASH(ctx,in,len) (*((ctx)->ghash))((ctx)->Xi.u,(ctx)->Htable,in,len)
 689 #endif
 690
 691 void CRYPTO_gcm128_init(GCM128_CONTEXT *ctx,void *key,block128_f block)
 692 {
 693         const union { long one; char little; } is_endian = {1};
 694
 695         memset(ctx,0,sizeof(*ctx));
 696         ctx->block = block;
 697         ctx->key   = key;
 698
 699         (*block)(ctx->H.c,ctx->H.c,key);
 700
 701         if (is_endian.little) {
 702                 /* H is stored in host byte order */
 703 #ifdef BSWAP8
 704                 ctx->H.u[0] = BSWAP8(ctx->H.u[0]);
 705                 ctx->H.u[1] = BSWAP8(ctx->H.u[1]);
 706 #else
 707                 u8 *p = ctx->H.c;
 708                 u64 hi,lo;
 709                 hi = (u64)GETU32(p)  <<32|GETU32(p+4);
 710                 lo = (u64)GETU32(p+8)<<32|GETU32(p+12);
 711                 ctx->H.u[0] = hi;
 712                 ctx->H.u[1] = lo;
 713 #endif
 714         }
 715
 716 #if     TABLE_BITS==8
 717         gcm_init_8bit(ctx->Htable,ctx->H.u);
 718 #elif   TABLE_BITS==4
 719 # if    defined(GHASH_ASM_IAX)                  /* both x86 and x86_64 */
 720         if (OPENSSL_ia32cap_P[1]&(1<<1)) {
 721                 gcm_init_clmul(ctx->Htable,ctx->H.u);
 722                 ctx->gmult = gcm_gmult_clmul;
 723                 ctx->ghash = gcm_ghash_clmul;
 724                 return;
 725         }
 726         gcm_init_4bit(ctx->Htable,ctx->H.u);
 727 #  if   defined(GHASH_ASM_X86)                  /* x86 only */
 728         if (OPENSSL_ia32cap_P[0]&(1<<23)) {
 729                 ctx->gmult = gcm_gmult_4bit_mmx;
 730                 ctx->ghash = gcm_ghash_4bit_mmx;
 731         } else {
 732                 ctx->gmult = gcm_gmult_4bit_x86;
 733                 ctx->ghash = gcm_ghash_4bit_x86;
 734         }
 735 #  else
 736         ctx->gmult = gcm_gmult_4bit;
 737         ctx->ghash = gcm_ghash_4bit;
 738 #  endif
 739 # else
 740         gcm_init_4bit(ctx->Htable,ctx->H.u);
 741 # endif
 742 #endif
 743 }
 744
 745 void CRYPTO_gcm128_setiv(GCM128_CONTEXT *ctx,const unsigned char *iv,size_t len)
 746 {
 747         const union { long one; char little; } is_endian = {1};
 748         unsigned int ctr;
 749
 750         ctx->Yi.u[0]  = 0;
 751         ctx->Yi.u[1]  = 0;
 752         ctx->Xi.u[0]  = 0;
 753         ctx->Xi.u[1]  = 0;
 754         ctx->len.u[0] = 0;      /* AAD length */
 755         ctx->len.u[1] = 0;      /* message length */
 756         ctx->ares = 0;
 757         ctx->mres = 0;
 758
 759         if (len==12) {
 760                 memcpy(ctx->Yi.c,iv,12);
 761                 ctx->Yi.c[15]=1;
 762                 ctr=1;
 763         }
 764         else {
 765                 size_t i;
 766                 u64 len0 = len;
 767
 768                 while (len>=16) {
 769                         for (i=0; i<16; ++i) ctx->Yi.c[i] ^= iv[i];
 770                         GCM_MUL(ctx,Yi);
 771                         iv += 16;
 772                         len -= 16;
 773                 }
 774                 if (len) {
 775                         for (i=0; i<len; ++i) ctx->Yi.c[i] ^= iv[i];
 776                         GCM_MUL(ctx,Yi);
 777                 }
 778                 len0 <<= 3;
 779                 if (is_endian.little) {
 780 #ifdef BSWAP8
 781                         ctx->Yi.u[1]  ^= BSWAP8(len0);
 782 #else
 783                         ctx->Yi.c[8]  ^= (u8)(len0>>56);
 784                         ctx->Yi.c[9]  ^= (u8)(len0>>48);
 785                         ctx->Yi.c[10] ^= (u8)(len0>>40);
 786                         ctx->Yi.c[11] ^= (u8)(len0>>32);
 787                         ctx->Yi.c[12] ^= (u8)(len0>>24);
 788                         ctx->Yi.c[13] ^= (u8)(len0>>16);
 789                         ctx->Yi.c[14] ^= (u8)(len0>>8);
 790                         ctx->Yi.c[15] ^= (u8)(len0);
 791 #endif
 792                 }
 793                 else
 794                         ctx->Yi.u[1]  ^= len0;
 795
 796                 GCM_MUL(ctx,Yi);
 797
 798                 if (is_endian.little)
 799                         ctr = GETU32(ctx->Yi.c+12);
 800                 else
 801                         ctr = ctx->Yi.d[3];
 802         }
 803
 804         (*ctx->block)(ctx->Yi.c,ctx->EK0.c,ctx->key);
 805         ++ctr;
 806         if (is_endian.little)
 807                 PUTU32(ctx->Yi.c+12,ctr);
 808         else
 809                 ctx->Yi.d[3] = ctr;
 810 }
 811
 812 void CRYPTO_gcm128_aad(GCM128_CONTEXT *ctx,const unsigned char *aad,size_t len)
 813 {
 814         size_t i;
 815         int n;
 816
 817         ctx->len.u[0] += len;
 818         n = ctx->ares;
 819
 820         if (n) {
 821                 while (n && len) {
 822                         ctx->Xi.c[n] ^= *(aad++);
 823                         --len;
 824                         n = (n+1)%16;
 825                 }
 826                 if (n==0) GCM_MUL(ctx,Xi);
 827                 else {
 828                         ctx->ares = n;
 829                         return;
 830                 }
 831         }
 832
 833 #ifdef GHASH
 834         if ((i = (len&(size_t)-16))) {
 835                 GHASH(ctx,aad,i);
 836                 aad += i;
 837                 len -= i;
 838         }
 839 #else
 840         while (len>=16) {
 841                 for (i=0; i<16; ++i) ctx->Xi.c[i] ^= aad[i];
 842                 GCM_MUL(ctx,Xi);
 843                 aad += 16;
 844                 len -= 16;
 845         }
 846 #endif
 847         if (len) {
 848                 n = (int)len;
 849                 for (i=0; i<len; ++i) ctx->Xi.c[i] ^= aad[i];
 850         }
 851
 852         ctx->ares = n;
 853 }
 854
 855 void CRYPTO_gcm128_encrypt(GCM128_CONTEXT *ctx,
 856                 const unsigned char *in, unsigned char *out,
 857                 size_t len)
 858 {
 859         const union { long one; char little; } is_endian = {1};
 860         unsigned int n, ctr;
 861         size_t i;
 862
 863         if (ctx->ares) {
 864                 /* First call to encrypt finalizes GHASH(AAD) */
 865                 GCM_MUL(ctx,Xi);
 866                 ctx->ares = 0;
 867         }
 868
 869         ctx->len.u[1] += len;
 870         n   = ctx->mres;
 871         if (is_endian.little)
 872                 ctr = GETU32(ctx->Yi.c+12);
 873         else
 874                 ctr = ctx->Yi.d[3];
 875
 876 #if !defined(OPENSSL_SMALL_FOOTPRINT)
 877         if (16%sizeof(size_t) == 0) do {        /* always true actually */
 878                 if (n) {
 879                         while (n && len) {
 880                                 ctx->Xi.c[n] ^= *(out++) = *(in++)^ctx->EKi.c[n];
 881                                 --len;
 882                                 n = (n+1)%16;
 883                         }
 884                         if (n==0) GCM_MUL(ctx,Xi);
 885                         else {
 886                                 ctx->mres = n;
 887                                 return;
 888                         }
 889                 }
 890 #if defined(STRICT_ALIGNMENT)
 891                 if (((size_t)in|(size_t)out)%sizeof(size_t) != 0)
 892                         break;
 893 #endif
 894 #if defined(GHASH) && defined(GHASH_CHUNK)
 895                 while (len>=GHASH_CHUNK) {
 896                     size_t j=GHASH_CHUNK;
 897
 898                     while (j) {
 899                         (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
 900                         ++ctr;
 901                         if (is_endian.little)
 902                                 PUTU32(ctx->Yi.c+12,ctr);
 903                         else
 904                                 ctx->Yi.d[3] = ctr;
 905                         for (i=0; i<16; i+=sizeof(size_t))
 906                                 *(size_t *)(out+i) =
 907                                 *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
 908                         out += 16;
 909                         in  += 16;
 910                         j   -= 16;
 911                     }
 912                     GHASH(ctx,out-GHASH_CHUNK,GHASH_CHUNK);
 913                     len -= GHASH_CHUNK;
 914                 }
 915                 if ((i = (len&(size_t)-16))) {
 916                     size_t j=i;
 917
 918                     while (len>=16) {
 919                         (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
 920                         ++ctr;
 921                         if (is_endian.little)
 922                                 PUTU32(ctx->Yi.c+12,ctr);
 923                         else
 924                                 ctx->Yi.d[3] = ctr;
 925                         for (i=0; i<16; i+=sizeof(size_t))
 926                                 *(size_t *)(out+i) =
 927                                 *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
 928                         out += 16;
 929                         in  += 16;
 930                         len -= 16;
 931                     }
 932                     GHASH(ctx,out-j,j);
 933                 }
 934 #else
 935                 while (len>=16) {
 936                         (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
 937                         ++ctr;
 938                         if (is_endian.little)
 939                                 PUTU32(ctx->Yi.c+12,ctr);
 940                         else
 941                                 ctx->Yi.d[3] = ctr;
 942                         for (i=0; i<16; i+=sizeof(size_t))
 943                                 *(size_t *)(ctx->Xi.c+i) ^=
 944                                 *(size_t *)(out+i) =
 945                                 *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
 946                         GCM_MUL(ctx,Xi);
 947                         out += 16;
 948                         in  += 16;
 949                         len -= 16;
 950                 }
 951 #endif
 952                 if (len) {
 953                         (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
 954                         ++ctr;
 955                         if (is_endian.little)
 956                                 PUTU32(ctx->Yi.c+12,ctr);
 957                         else
 958                                 ctx->Yi.d[3] = ctr;
 959                         while (len--) {
 960                                 ctx->Xi.c[n] ^= out[n] = in[n]^ctx->EKi.c[n];
 961                                 ++n;
 962                         }
 963                 }
 964
 965                 ctx->mres = n;
 966                 return;
 967         } while(0);
 968 #endif
 969         for (i=0;i<len;++i) {
 970                 if (n==0) {
 971                         (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
 972                         ++ctr;
 973                         if (is_endian.little)
 974                                 PUTU32(ctx->Yi.c+12,ctr);
 975                         else
 976                                 ctx->Yi.d[3] = ctr;
 977                 }
 978                 ctx->Xi.c[n] ^= out[i] = in[i]^ctx->EKi.c[n];
 979                 n = (n+1)%16;
 980                 if (n==0)
 981                         GCM_MUL(ctx,Xi);
 982         }
 983
 984         ctx->mres = n;
 985 }
 986
 987 void CRYPTO_gcm128_decrypt(GCM128_CONTEXT *ctx,
 988                 const unsigned char *in, unsigned char *out,
 989                 size_t len)
 990 {
 991         const union { long one; char little; } is_endian = {1};
 992         unsigned int n, ctr;
 993         size_t i;
 994
 995         if (ctx->ares) {
 996                 /* First call to decrypt finalizes GHASH(AAD) */
 997                 GCM_MUL(ctx,Xi);
 998                 ctx->ares = 0;
 999         }
1000
1001         ctx->len.u[1] += len;
1002         n   = ctx->mres;
1003         if (is_endian.little)
1004                 ctr = GETU32(ctx->Yi.c+12);
1005         else
1006                 ctr = ctx->Yi.d[3];
1007
1008 #if !defined(OPENSSL_SMALL_FOOTPRINT)
1009         if (16%sizeof(size_t) == 0) do {        /* always true actually */
1010                 if (n) {
1011                         while (n && len) {
1012                                 u8 c = *(in++);
1013                                 *(out++) = c^ctx->EKi.c[n];
1014                                 ctx->Xi.c[n] ^= c;
1015                                 --len;
1016                                 n = (n+1)%16;
1017                         }
1018                         if (n==0) GCM_MUL (ctx,Xi);
1019                         else {
1020                                 ctx->mres = n;
1021                                 return;
1022                         }
1023                 }
1024 #if defined(STRICT_ALIGNMENT)
1025                 if (((size_t)in|(size_t)out)%sizeof(size_t) != 0)
1026                         break;
1027 #endif
1028 #if defined(GHASH) && defined(GHASH_CHUNK)
1029                 while (len>=GHASH_CHUNK) {
1030                     size_t j=GHASH_CHUNK;
1031
1032                     GHASH(ctx,in,GHASH_CHUNK);
1033                     while (j) {
1034                         (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
1035                         ++ctr;
1036                         if (is_endian.little)
1037                                 PUTU32(ctx->Yi.c+12,ctr);
1038                         else
1039                                 ctx->Yi.d[3] = ctr;
1040                         for (i=0; i<16; i+=sizeof(size_t))
1041                                 *(size_t *)(out+i) =
1042                                 *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
1043                         out += 16;
1044                         in  += 16;
1045                         j   -= 16;
1046                     }
1047                     len -= GHASH_CHUNK;
1048                 }
1049                 if ((i = (len&(size_t)-16))) {
1050                     GHASH(ctx,in,i);
1051                     while (len>=16) {
1052                         (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
1053                         ++ctr;
1054                         if (is_endian.little)
1055                                 PUTU32(ctx->Yi.c+12,ctr);
1056                         else
1057                                 ctx->Yi.d[3] = ctr;
1058                         for (i=0; i<16; i+=sizeof(size_t))
1059                                 *(size_t *)(out+i) =
1060                                 *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
1061                         out += 16;
1062                         in  += 16;
1063                         len -= 16;
1064                     }
1065                 }
1066 #else
1067                 while (len>=16) {
1068                         (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
1069                         ++ctr;
1070                         if (is_endian.little)
1071                                 PUTU32(ctx->Yi.c+12,ctr);
1072                         else
1073                                 ctx->Yi.d[3] = ctr;
1074                         for (i=0; i<16; i+=sizeof(size_t)) {
1075                                 size_t c = *(size_t *)(in+i);
1076                                 *(size_t *)(out+i) = c^*(size_t *)(ctx->EKi.c+i);
1077                                 *(size_t *)(ctx->Xi.c+i) ^= c;
1078                         }
1079                         GCM_MUL(ctx,Xi);
1080                         out += 16;
1081                         in  += 16;
1082                         len -= 16;
1083                 }
1084 #endif
1085                 if (len) {
1086                         (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
1087                         ++ctr;
1088                         if (is_endian.little)
1089                                 PUTU32(ctx->Yi.c+12,ctr);
1090                         else
1091                                 ctx->Yi.d[3] = ctr;
1092                         while (len--) {
1093                                 u8 c = in[n];
1094                                 ctx->Xi.c[n] ^= c;
1095                                 out[n] = c^ctx->EKi.c[n];
1096                                 ++n;
1097                         }
1098                 }
1099
1100                 ctx->mres = n;
1101                 return;
1102         } while(0);
1103 #endif
1104         for (i=0;i<len;++i) {
1105                 u8 c;
1106                 if (n==0) {
1107                         (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
1108                         ++ctr;
1109                         if (is_endian.little)
1110                                 PUTU32(ctx->Yi.c+12,ctr);
1111                         else
1112                                 ctx->Yi.d[3] = ctr;
1113                 }
1114                 c = in[i];
1115                 out[i] = c^ctx->EKi.c[n];
1116                 ctx->Xi.c[n] ^= c;
1117                 n = (n+1)%16;
1118                 if (n==0)
1119                         GCM_MUL(ctx,Xi);
1120         }
1121
1122         ctx->mres = n;
1123 }
1124
1125 void CRYPTO_gcm128_encrypt_ctr32(GCM128_CONTEXT *ctx,
1126                 const unsigned char *in, unsigned char *out,
1127                 size_t len, ctr128_f stream)
1128 {
1129         const union { long one; char little; } is_endian = {1};
1130         unsigned int n, ctr;
1131         size_t i;
1132
1133         if (ctx->ares) {
1134                 /* First call to encrypt finalizes GHASH(AAD) */
1135                 GCM_MUL(ctx,Xi);
1136                 ctx->ares = 0;
1137         }
1138
1139         ctx->len.u[1] += len;
1140         n   = ctx->mres;
1141         if (is_endian.little)
1142                 ctr = GETU32(ctx->Yi.c+12);
1143         else
1144                 ctr = ctx->Yi.d[3];
1145
1146         if (n) {
1147                 while (n && len) {
1148                         ctx->Xi.c[n] ^= *(out++) = *(in++)^ctx->EKi.c[n];
1149                         --len;
1150                         n = (n+1)%16;
1151                 }
1152                 if (n==0) GCM_MUL(ctx,Xi);
1153                 else {
1154                         ctx->mres = n;
1155                         return;
1156                 }
1157         }
1158 #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
1159         while (len>=GHASH_CHUNK) {
1160                 (*stream)(in,out,GHASH_CHUNK/16,ctx->key,ctx->Yi.c);
1161                 ctr += GHASH_CHUNK/16;
1162                 if (is_endian.little)
1163                         PUTU32(ctx->Yi.c+12,ctr);
1164                 else
1165                         ctx->Yi.d[3] = ctr;
1166                 GHASH(ctx,out,GHASH_CHUNK);
1167                 out += GHASH_CHUNK;
1168                 in  += GHASH_CHUNK;
1169                 len -= GHASH_CHUNK;
1170         }
1171 #endif
1172         if ((i = (len&(size_t)-16))) {
1173                 size_t j=i/16;
1174
1175                 (*stream)(in,out,j,ctx->key,ctx->Yi.c);
1176                 ctr += (unsigned int)j;
1177                 if (is_endian.little)
1178                         PUTU32(ctx->Yi.c+12,ctr);
1179                 else
1180                         ctx->Yi.d[3] = ctr;
1181                 in  += i;
1182                 len -= i;
1183 #if defined(GHASH)
1184                 GHASH(ctx,out,i);
1185                 out += i;
1186 #else
1187                 while (j--) {
1188                         for (i=0;i<16;++i) ctx->Xi.c[i] ^= out[i];
1189                         GCM_MUL(ctx,Xi);
1190                         out += 16;
1191                 }
1192 #endif
1193         }
1194         if (len) {
1195                 (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
1196                 ++ctr;
1197                 if (is_endian.little)
1198                         PUTU32(ctx->Yi.c+12,ctr);
1199                 else
1200                         ctx->Yi.d[3] = ctr;
1201                 while (len--) {
1202                         ctx->Xi.c[n] ^= out[n] = in[n]^ctx->EKi.c[n];
1203                         ++n;
1204                 }
1205         }
1206
1207         ctx->mres = n;
1208 }
1209
1210 void CRYPTO_gcm128_decrypt_ctr32(GCM128_CONTEXT *ctx,
1211                 const unsigned char *in, unsigned char *out,
1212                 size_t len,ctr128_f stream)
1213 {
1214         const union { long one; char little; } is_endian = {1};
1215         unsigned int n, ctr;
1216         size_t i;
1217
1218         if (ctx->ares) {
1219                 /* First call to decrypt finalizes GHASH(AAD) */
1220                 GCM_MUL(ctx,Xi);
1221                 ctx->ares = 0;
1222         }
1223
1224         ctx->len.u[1] += len;
1225         n   = ctx->mres;
1226         if (is_endian.little)
1227                 ctr = GETU32(ctx->Yi.c+12);
1228         else
1229                 ctr = ctx->Yi.d[3];
1230
1231         if (n) {
1232                 while (n && len) {
1233                         u8 c = *(in++);
1234                         *(out++) = c^ctx->EKi.c[n];
1235                         ctx->Xi.c[n] ^= c;
1236                         --len;
1237                         n = (n+1)%16;
1238                 }
1239                 if (n==0) GCM_MUL (ctx,Xi);
1240                 else {
1241                         ctx->mres = n;
1242                         return;
1243                 }
1244         }
1245 #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
1246         while (len>=GHASH_CHUNK) {
1247                 GHASH(ctx,in,GHASH_CHUNK);
1248                 (*stream)(in,out,GHASH_CHUNK/16,ctx->key,ctx->Yi.c);
1249                 ctr += GHASH_CHUNK/16;
1250                 if (is_endian.little)
1251                         PUTU32(ctx->Yi.c+12,ctr);
1252                 else
1253                         ctx->Yi.d[3] = ctr;
1254                 out += GHASH_CHUNK;
1255                 in  += GHASH_CHUNK;
1256                 len -= GHASH_CHUNK;
1257         }
1258 #endif
1259         if ((i = (len&(size_t)-16))) {
1260                 size_t j=i/16;
1261
1262 #if defined(GHASH)
1263                 GHASH(ctx,in,i);
1264 #else
1265                 while (j--) {
1266                         size_t k;
1267                         for (k=0;k<16;++k) ctx->Xi.c[k] ^= in[k];
1268                         GCM_MUL(ctx,Xi);
1269                         in += 16;
1270                 }
1271                 j   = i/16;
1272                 in -= i;
1273 #endif
1274                 (*stream)(in,out,j,ctx->key,ctx->Yi.c);
1275                 ctr += (unsigned int)j;
1276                 if (is_endian.little)
1277                         PUTU32(ctx->Yi.c+12,ctr);
1278                 else
1279                         ctx->Yi.d[3] = ctr;
1280                 out += i;
1281                 in  += i;
1282                 len -= i;
1283         }
1284         if (len) {
1285                 (*ctx->block)(ctx->Yi.c,ctx->EKi.c,ctx->key);
1286                 ++ctr;
1287                 if (is_endian.little)
1288                         PUTU32(ctx->Yi.c+12,ctr);
1289                 else
1290                         ctx->Yi.d[3] = ctr;
1291                 while (len--) {
1292                         u8 c = in[n];
1293                         ctx->Xi.c[n] ^= c;
1294                         out[n] = c^ctx->EKi.c[n];
1295                         ++n;
1296                 }
1297         }
1298
1299         ctx->mres = n;
1300 }
1301
1302 int CRYPTO_gcm128_finish(GCM128_CONTEXT *ctx,const unsigned char *tag,
1303                         size_t len)
1304 {
1305         const union { long one; char little; } is_endian = {1};
1306         u64 alen = ctx->len.u[0]<<3;
1307         u64 clen = ctx->len.u[1]<<3;
1308
1309         if (ctx->mres)
1310                 GCM_MUL(ctx,Xi);
1311
1312         if (is_endian.little) {
1313 #ifdef BSWAP8
1314                 alen = BSWAP8(alen);
1315                 clen = BSWAP8(clen);
1316 #else
1317                 u8 *p = ctx->len.c;
1318
1319                 ctx->len.u[0] = alen;
1320                 ctx->len.u[1] = clen;
1321
1322                 alen = (u64)GETU32(p)  <<32|GETU32(p+4);
1323                 clen = (u64)GETU32(p+8)<<32|GETU32(p+12);
1324 #endif
1325         }
1326
1327         ctx->Xi.u[0] ^= alen;
1328         ctx->Xi.u[1] ^= clen;
1329         GCM_MUL(ctx,Xi);
1330
1331         ctx->Xi.u[0] ^= ctx->EK0.u[0];
1332         ctx->Xi.u[1] ^= ctx->EK0.u[1];
1333
1334         if (tag && len<=sizeof(ctx->Xi))
1335                 return memcmp(ctx->Xi.c,tag,len);
1336         else
1337                 return -1;
1338 }
1339
1340 GCM128_CONTEXT *CRYPTO_gcm128_new(void *key, block128_f block)
1341 {
1342         GCM128_CONTEXT *ret;
1343
1344         if ((ret = (GCM128_CONTEXT *)OPENSSL_malloc(sizeof(GCM128_CONTEXT))))
1345                 CRYPTO_gcm128_init(ret,key,block);
1346
1347         return ret;
1348 }
1349
1350 void CRYPTO_gcm128_release(GCM128_CONTEXT *ctx)
1351 {
1352         if (ctx) {
1353                 OPENSSL_cleanse(ctx,sizeof(*ctx));
1354                 OPENSSL_free(ctx);
1355         }
1356 }
1357
1358 #if defined(SELFTEST)
1359 #include <stdio.h>
1360 #include <openssl/aes.h>
1361
1362 /* Test Case 1 */
1363 static const u8 K1[16],
1364                 *P1=NULL,
1365                 *A1=NULL,
1366                 IV1[12],
1367                 *C1=NULL,
1368                 T1[]=  {0x58,0xe2,0xfc,0xce,0xfa,0x7e,0x30,0x61,0x36,0x7f,0x1d,0x57,0xa4,0xe7,0x45,0x5a};
1369
1370 /* Test Case 2 */
1371 #define K2 K1
1372 #define A2 A1
1373 #define IV2 IV1
1374 static const u8 P2[16],
1375                 C2[]=  {0x03,0x88,0xda,0xce,0x60,0xb6,0xa3,0x92,0xf3,0x28,0xc2,0xb9,0x71,0xb2,0xfe,0x78},
1376                 T2[]=  {0xab,0x6e,0x47,0xd4,0x2c,0xec,0x13,0xbd,0xf5,0x3a,0x67,0xb2,0x12,0x57,0xbd,0xdf};
1377
1378 /* Test Case 3 */
1379 #define A3 A2
1380 static const u8 K3[]=  {0xfe,0xff,0xe9,0x92,0x86,0x65,0x73,0x1c,0x6d,0x6a,0x8f,0x94,0x67,0x30,0x83,0x08},
1381                 P3[]=  {0xd9,0x31,0x32,0x25,0xf8,0x84,0x06,0xe5,0xa5,0x59,0x09,0xc5,0xaf,0xf5,0x26,0x9a,
1382                         0x86,0xa7,0xa9,0x53,0x15,0x34,0xf7,0xda,0x2e,0x4c,0x30,0x3d,0x8a,0x31,0x8a,0x72,
1383                         0x1c,0x3c,0x0c,0x95,0x95,0x68,0x09,0x53,0x2f,0xcf,0x0e,0x24,0x49,0xa6,0xb5,0x25,
1384                         0xb1,0x6a,0xed,0xf5,0xaa,0x0d,0xe6,0x57,0xba,0x63,0x7b,0x39,0x1a,0xaf,0xd2,0x55},
1385                 IV3[]= {0xca,0xfe,0xba,0xbe,0xfa,0xce,0xdb,0xad,0xde,0xca,0xf8,0x88},
1386                 C3[]=  {0x42,0x83,0x1e,0xc2,0x21,0x77,0x74,0x24,0x4b,0x72,0x21,0xb7,0x84,0xd0,0xd4,0x9c,
1387                         0xe3,0xaa,0x21,0x2f,0x2c,0x02,0xa4,0xe0,0x35,0xc1,0x7e,0x23,0x29,0xac,0xa1,0x2e,
1388                         0x21,0xd5,0x14,0xb2,0x54,0x66,0x93,0x1c,0x7d,0x8f,0x6a,0x5a,0xac,0x84,0xaa,0x05,
1389                         0x1b,0xa3,0x0b,0x39,0x6a,0x0a,0xac,0x97,0x3d,0x58,0xe0,0x91,0x47,0x3f,0x59,0x85},
1390                 T3[]=  {0x4d,0x5c,0x2a,0xf3,0x27,0xcd,0x64,0xa6,0x2c,0xf3,0x5a,0xbd,0x2b,0xa6,0xfa,0xb4};
1391
1392 /* Test Case 4 */
1393 #define K4 K3
1394 #define IV4 IV3
1395 static const u8 P4[]=  {0xd9,0x31,0x32,0x25,0xf8,0x84,0x06,0xe5,0xa5,0x59,0x09,0xc5,0xaf,0xf5,0x26,0x9a,
1396                         0x86,0xa7,0xa9,0x53,0x15,0x34,0xf7,0xda,0x2e,0x4c,0x30,0x3d,0x8a,0x31,0x8a,0x72,
1397                         0x1c,0x3c,0x0c,0x95,0x95,0x68,0x09,0x53,0x2f,0xcf,0x0e,0x24,0x49,0xa6,0xb5,0x25,
1398                         0xb1,0x6a,0xed,0xf5,0xaa,0x0d,0xe6,0x57,0xba,0x63,0x7b,0x39},
1399                 A4[]=  {0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,
1400                         0xab,0xad,0xda,0xd2},
1401                 C4[]=  {0x42,0x83,0x1e,0xc2,0x21,0x77,0x74,0x24,0x4b,0x72,0x21,0xb7,0x84,0xd0,0xd4,0x9c,
1402                         0xe3,0xaa,0x21,0x2f,0x2c,0x02,0xa4,0xe0,0x35,0xc1,0x7e,0x23,0x29,0xac,0xa1,0x2e,
1403                         0x21,0xd5,0x14,0xb2,0x54,0x66,0x93,0x1c,0x7d,0x8f,0x6a,0x5a,0xac,0x84,0xaa,0x05,
1404                         0x1b,0xa3,0x0b,0x39,0x6a,0x0a,0xac,0x97,0x3d,0x58,0xe0,0x91},
1405                 T4[]=  {0x5b,0xc9,0x4f,0xbc,0x32,0x21,0xa5,0xdb,0x94,0xfa,0xe9,0x5a,0xe7,0x12,0x1a,0x47};
1406
1407 /* Test Case 5 */
1408 #define K5 K4
1409 #define P5 P4
1410 static const u8 A5[]=  {0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,
1411                         0xab,0xad,0xda,0xd2},
1412                 IV5[]= {0xca,0xfe,0xba,0xbe,0xfa,0xce,0xdb,0xad},
1413                 C5[]=  {0x61,0x35,0x3b,0x4c,0x28,0x06,0x93,0x4a,0x77,0x7f,0xf5,0x1f,0xa2,0x2a,0x47,0x55,
1414                         0x69,0x9b,0x2a,0x71,0x4f,0xcd,0xc6,0xf8,0x37,0x66,0xe5,0xf9,0x7b,0x6c,0x74,0x23,
1415                         0x73,0x80,0x69,0x00,0xe4,0x9f,0x24,0xb2,0x2b,0x09,0x75,0x44,0xd4,0x89,0x6b,0x42,
1416                         0x49,0x89,0xb5,0xe1,0xeb,0xac,0x0f,0x07,0xc2,0x3f,0x45,0x98},
1417                 T5[]=  {0x36,0x12,0xd2,0xe7,0x9e,0x3b,0x07,0x85,0x56,0x1b,0xe1,0x4a,0xac,0xa2,0xfc,0xcb};
1418
1419 /* Test Case 6 */
1420 #define K6 K5
1421 #define P6 P5
1422 #define A6 A5
1423 static const u8 IV6[]= {0x93,0x13,0x22,0x5d,0xf8,0x84,0x06,0xe5,0x55,0x90,0x9c,0x5a,0xff,0x52,0x69,0xaa,
1424                         0x6a,0x7a,0x95,0x38,0x53,0x4f,0x7d,0xa1,0xe4,0xc3,0x03,0xd2,0xa3,0x18,0xa7,0x28,
1425                         0xc3,0xc0,0xc9,0x51,0x56,0x80,0x95,0x39,0xfc,0xf0,0xe2,0x42,0x9a,0x6b,0x52,0x54,
1426                         0x16,0xae,0xdb,0xf5,0xa0,0xde,0x6a,0x57,0xa6,0x37,0xb3,0x9b},
1427                 C6[]=  {0x8c,0xe2,0x49,0x98,0x62,0x56,0x15,0xb6,0x03,0xa0,0x33,0xac,0xa1,0x3f,0xb8,0x94,
1428                         0xbe,0x91,0x12,0xa5,0xc3,0xa2,0x11,0xa8,0xba,0x26,0x2a,0x3c,0xca,0x7e,0x2c,0xa7,
1429                         0x01,0xe4,0xa9,0xa4,0xfb,0xa4,0x3c,0x90,0xcc,0xdc,0xb2,0x81,0xd4,0x8c,0x7c,0x6f,
1430                         0xd6,0x28,0x75,0xd2,0xac,0xa4,0x17,0x03,0x4c,0x34,0xae,0xe5},
1431                 T6[]=  {0x61,0x9c,0xc5,0xae,0xff,0xfe,0x0b,0xfa,0x46,0x2a,0xf4,0x3c,0x16,0x99,0xd0,0x50};
1432
1433 /* Test Case 7 */
1434 static const u8 K7[24],
1435                 *P7=NULL,
1436                 *A7=NULL,
1437                 IV7[12],
1438                 *C7=NULL,
1439                 T7[]=  {0xcd,0x33,0xb2,0x8a,0xc7,0x73,0xf7,0x4b,0xa0,0x0e,0xd1,0xf3,0x12,0x57,0x24,0x35};
1440
1441 /* Test Case 8 */
1442 #define K8 K7
1443 #define IV8 IV7
1444 #define A8 A7
1445 static const u8 P8[16],
1446                 C8[]=  {0x98,0xe7,0x24,0x7c,0x07,0xf0,0xfe,0x41,0x1c,0x26,0x7e,0x43,0x84,0xb0,0xf6,0x00},
1447                 T8[]=  {0x2f,0xf5,0x8d,0x80,0x03,0x39,0x27,0xab,0x8e,0xf4,0xd4,0x58,0x75,0x14,0xf0,0xfb};
1448
1449 /* Test Case 9 */
1450 #define A9 A8
1451 static const u8 K9[]=  {0xfe,0xff,0xe9,0x92,0x86,0x65,0x73,0x1c,0x6d,0x6a,0x8f,0x94,0x67,0x30,0x83,0x08,
1452                         0xfe,0xff,0xe9,0x92,0x86,0x65,0x73,0x1c},
1453                 P9[]=  {0xd9,0x31,0x32,0x25,0xf8,0x84,0x06,0xe5,0xa5,0x59,0x09,0xc5,0xaf,0xf5,0x26,0x9a,
1454                         0x86,0xa7,0xa9,0x53,0x15,0x34,0xf7,0xda,0x2e,0x4c,0x30,0x3d,0x8a,0x31,0x8a,0x72,
1455                         0x1c,0x3c,0x0c,0x95,0x95,0x68,0x09,0x53,0x2f,0xcf,0x0e,0x24,0x49,0xa6,0xb5,0x25,
1456                         0xb1,0x6a,0xed,0xf5,0xaa,0x0d,0xe6,0x57,0xba,0x63,0x7b,0x39,0x1a,0xaf,0xd2,0x55},
1457                 IV9[]= {0xca,0xfe,0xba,0xbe,0xfa,0xce,0xdb,0xad,0xde,0xca,0xf8,0x88},
1458                 C9[]=  {0x39,0x80,0xca,0x0b,0x3c,0x00,0xe8,0x41,0xeb,0x06,0xfa,0xc4,0x87,0x2a,0x27,0x57,
1459                         0x85,0x9e,0x1c,0xea,0xa6,0xef,0xd9,0x84,0x62,0x85,0x93,0xb4,0x0c,0xa1,0xe1,0x9c,
1460                         0x7d,0x77,0x3d,0x00,0xc1,0x44,0xc5,0x25,0xac,0x61,0x9d,0x18,0xc8,0x4a,0x3f,0x47,
1461                         0x18,0xe2,0x44,0x8b,0x2f,0xe3,0x24,0xd9,0xcc,0xda,0x27,0x10,0xac,0xad,0xe2,0x56},
1462                 T9[]=  {0x99,0x24,0xa7,0xc8,0x58,0x73,0x36,0xbf,0xb1,0x18,0x02,0x4d,0xb8,0x67,0x4a,0x14};
1463
1464 /* Test Case 10 */
1465 #define K10 K9
1466 #define IV10 IV9
1467 static const u8 P10[]= {0xd9,0x31,0x32,0x25,0xf8,0x84,0x06,0xe5,0xa5,0x59,0x09,0xc5,0xaf,0xf5,0x26,0x9a,
1468                         0x86,0xa7,0xa9,0x53,0x15,0x34,0xf7,0xda,0x2e,0x4c,0x30,0x3d,0x8a,0x31,0x8a,0x72,
1469                         0x1c,0x3c,0x0c,0x95,0x95,0x68,0x09,0x53,0x2f,0xcf,0x0e,0x24,0x49,0xa6,0xb5,0x25,
1470                         0xb1,0x6a,0xed,0xf5,0xaa,0x0d,0xe6,0x57,0xba,0x63,0x7b,0x39},
1471                 A10[]= {0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,
1472                         0xab,0xad,0xda,0xd2},
1473                 C10[]= {0x39,0x80,0xca,0x0b,0x3c,0x00,0xe8,0x41,0xeb,0x06,0xfa,0xc4,0x87,0x2a,0x27,0x57,
1474                         0x85,0x9e,0x1c,0xea,0xa6,0xef,0xd9,0x84,0x62,0x85,0x93,0xb4,0x0c,0xa1,0xe1,0x9c,
1475                         0x7d,0x77,0x3d,0x00,0xc1,0x44,0xc5,0x25,0xac,0x61,0x9d,0x18,0xc8,0x4a,0x3f,0x47,
1476                         0x18,0xe2,0x44,0x8b,0x2f,0xe3,0x24,0xd9,0xcc,0xda,0x27,0x10},
1477                 T10[]= {0x25,0x19,0x49,0x8e,0x80,0xf1,0x47,0x8f,0x37,0xba,0x55,0xbd,0x6d,0x27,0x61,0x8c};
1478
1479 /* Test Case 11 */
1480 #define K11 K10
1481 #define P11 P10
1482 #define A11 A10
1483 static const u8 IV11[]={0xca,0xfe,0xba,0xbe,0xfa,0xce,0xdb,0xad},
1484                 C11[]= {0x0f,0x10,0xf5,0x99,0xae,0x14,0xa1,0x54,0xed,0x24,0xb3,0x6e,0x25,0x32,0x4d,0xb8,
1485                         0xc5,0x66,0x63,0x2e,0xf2,0xbb,0xb3,0x4f,0x83,0x47,0x28,0x0f,0xc4,0x50,0x70,0x57,
1486                         0xfd,0xdc,0x29,0xdf,0x9a,0x47,0x1f,0x75,0xc6,0x65,0x41,0xd4,0xd4,0xda,0xd1,0xc9,
1487                         0xe9,0x3a,0x19,0xa5,0x8e,0x8b,0x47,0x3f,0xa0,0xf0,0x62,0xf7},
1488                 T11[]= {0x65,0xdc,0xc5,0x7f,0xcf,0x62,0x3a,0x24,0x09,0x4f,0xcc,0xa4,0x0d,0x35,0x33,0xf8};
1489
1490 /* Test Case 12 */
1491 #define K12 K11
1492 #define P12 P11
1493 #define A12 A11
1494 static const u8 IV12[]={0x93,0x13,0x22,0x5d,0xf8,0x84,0x06,0xe5,0x55,0x90,0x9c,0x5a,0xff,0x52,0x69,0xaa,
1495                         0x6a,0x7a,0x95,0x38,0x53,0x4f,0x7d,0xa1,0xe4,0xc3,0x03,0xd2,0xa3,0x18,0xa7,0x28,
1496                         0xc3,0xc0,0xc9,0x51,0x56,0x80,0x95,0x39,0xfc,0xf0,0xe2,0x42,0x9a,0x6b,0x52,0x54,
1497                         0x16,0xae,0xdb,0xf5,0xa0,0xde,0x6a,0x57,0xa6,0x37,0xb3,0x9b},
1498                 C12[]= {0xd2,0x7e,0x88,0x68,0x1c,0xe3,0x24,0x3c,0x48,0x30,0x16,0x5a,0x8f,0xdc,0xf9,0xff,
1499                         0x1d,0xe9,0xa1,0xd8,0xe6,0xb4,0x47,0xef,0x6e,0xf7,0xb7,0x98,0x28,0x66,0x6e,0x45,
1500                         0x81,0xe7,0x90,0x12,0xaf,0x34,0xdd,0xd9,0xe2,0xf0,0x37,0x58,0x9b,0x29,0x2d,0xb3,
1501                         0xe6,0x7c,0x03,0x67,0x45,0xfa,0x22,0xe7,0xe9,0xb7,0x37,0x3b},
1502                 T12[]= {0xdc,0xf5,0x66,0xff,0x29,0x1c,0x25,0xbb,0xb8,0x56,0x8f,0xc3,0xd3,0x76,0xa6,0xd9};
1503
1504 /* Test Case 13 */
1505 static const u8 K13[32],
1506                 *P13=NULL,
1507                 *A13=NULL,
1508                 IV13[12],
1509                 *C13=NULL,
1510                 T13[]={0x53,0x0f,0x8a,0xfb,0xc7,0x45,0x36,0xb9,0xa9,0x63,0xb4,0xf1,0xc4,0xcb,0x73,0x8b};
1511
1512 /* Test Case 14 */
1513 #define K14 K13
1514 #define A14 A13
1515 static const u8 P14[16],
1516                 IV14[12],
1517                 C14[]= {0xce,0xa7,0x40,0x3d,0x4d,0x60,0x6b,0x6e,0x07,0x4e,0xc5,0xd3,0xba,0xf3,0x9d,0x18},
1518                 T14[]= {0xd0,0xd1,0xc8,0xa7,0x99,0x99,0x6b,0xf0,0x26,0x5b,0x98,0xb5,0xd4,0x8a,0xb9,0x19};
1519
1520 /* Test Case 15 */
1521 #define A15 A14
1522 static const u8 K15[]= {0xfe,0xff,0xe9,0x92,0x86,0x65,0x73,0x1c,0x6d,0x6a,0x8f,0x94,0x67,0x30,0x83,0x08,
1523                         0xfe,0xff,0xe9,0x92,0x86,0x65,0x73,0x1c,0x6d,0x6a,0x8f,0x94,0x67,0x30,0x83,0x08},
1524                 P15[]= {0xd9,0x31,0x32,0x25,0xf8,0x84,0x06,0xe5,0xa5,0x59,0x09,0xc5,0xaf,0xf5,0x26,0x9a,
1525                         0x86,0xa7,0xa9,0x53,0x15,0x34,0xf7,0xda,0x2e,0x4c,0x30,0x3d,0x8a,0x31,0x8a,0x72,
1526                         0x1c,0x3c,0x0c,0x95,0x95,0x68,0x09,0x53,0x2f,0xcf,0x0e,0x24,0x49,0xa6,0xb5,0x25,
1527                         0xb1,0x6a,0xed,0xf5,0xaa,0x0d,0xe6,0x57,0xba,0x63,0x7b,0x39,0x1a,0xaf,0xd2,0x55},
1528                 IV15[]={0xca,0xfe,0xba,0xbe,0xfa,0xce,0xdb,0xad,0xde,0xca,0xf8,0x88},
1529                 C15[]= {0x52,0x2d,0xc1,0xf0,0x99,0x56,0x7d,0x07,0xf4,0x7f,0x37,0xa3,0x2a,0x84,0x42,0x7d,
1530                         0x64,0x3a,0x8c,0xdc,0xbf,0xe5,0xc0,0xc9,0x75,0x98,0xa2,0xbd,0x25,0x55,0xd1,0xaa,
1531                         0x8c,0xb0,0x8e,0x48,0x59,0x0d,0xbb,0x3d,0xa7,0xb0,0x8b,0x10,0x56,0x82,0x88,0x38,
1532                         0xc5,0xf6,0x1e,0x63,0x93,0xba,0x7a,0x0a,0xbc,0xc9,0xf6,0x62,0x89,0x80,0x15,0xad},
1533                 T15[]= {0xb0,0x94,0xda,0xc5,0xd9,0x34,0x71,0xbd,0xec,0x1a,0x50,0x22,0x70,0xe3,0xcc,0x6c};
1534
1535 /* Test Case 16 */
1536 #define K16 K15
1537 #define IV16 IV15
1538 static const u8 P16[]= {0xd9,0x31,0x32,0x25,0xf8,0x84,0x06,0xe5,0xa5,0x59,0x09,0xc5,0xaf,0xf5,0x26,0x9a,
1539                         0x86,0xa7,0xa9,0x53,0x15,0x34,0xf7,0xda,0x2e,0x4c,0x30,0x3d,0x8a,0x31,0x8a,0x72,
1540                         0x1c,0x3c,0x0c,0x95,0x95,0x68,0x09,0x53,0x2f,0xcf,0x0e,0x24,0x49,0xa6,0xb5,0x25,
1541                         0xb1,0x6a,0xed,0xf5,0xaa,0x0d,0xe6,0x57,0xba,0x63,0x7b,0x39},
1542                 A16[]= {0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,
1543                         0xab,0xad,0xda,0xd2},
1544                 C16[]= {0x52,0x2d,0xc1,0xf0,0x99,0x56,0x7d,0x07,0xf4,0x7f,0x37,0xa3,0x2a,0x84,0x42,0x7d,
1545                         0x64,0x3a,0x8c,0xdc,0xbf,0xe5,0xc0,0xc9,0x75,0x98,0xa2,0xbd,0x25,0x55,0xd1,0xaa,
1546                         0x8c,0xb0,0x8e,0x48,0x59,0x0d,0xbb,0x3d,0xa7,0xb0,0x8b,0x10,0x56,0x82,0x88,0x38,
1547                         0xc5,0xf6,0x1e,0x63,0x93,0xba,0x7a,0x0a,0xbc,0xc9,0xf6,0x62},
1548                 T16[]= {0x76,0xfc,0x6e,0xce,0x0f,0x4e,0x17,0x68,0xcd,0xdf,0x88,0x53,0xbb,0x2d,0x55,0x1b};
1549
1550 /* Test Case 17 */
1551 #define K17 K16
1552 #define P17 P16
1553 #define A17 A16
1554 static const u8 IV17[]={0xca,0xfe,0xba,0xbe,0xfa,0xce,0xdb,0xad},
1555                 C17[]= {0xc3,0x76,0x2d,0xf1,0xca,0x78,0x7d,0x32,0xae,0x47,0xc1,0x3b,0xf1,0x98,0x44,0xcb,
1556                         0xaf,0x1a,0xe1,0x4d,0x0b,0x97,0x6a,0xfa,0xc5,0x2f,0xf7,0xd7,0x9b,0xba,0x9d,0xe0,
1557                         0xfe,0xb5,0x82,0xd3,0x39,0x34,0xa4,0xf0,0x95,0x4c,0xc2,0x36,0x3b,0xc7,0x3f,0x78,
1558                         0x62,0xac,0x43,0x0e,0x64,0xab,0xe4,0x99,0xf4,0x7c,0x9b,0x1f},
1559                 T17[]= {0x3a,0x33,0x7d,0xbf,0x46,0xa7,0x92,0xc4,0x5e,0x45,0x49,0x13,0xfe,0x2e,0xa8,0xf2};
1560
1561 /* Test Case 18 */
1562 #define K18 K17
1563 #define P18 P17
1564 #define A18 A17
1565 static const u8 IV18[]={0x93,0x13,0x22,0x5d,0xf8,0x84,0x06,0xe5,0x55,0x90,0x9c,0x5a,0xff,0x52,0x69,0xaa,
1566                         0x6a,0x7a,0x95,0x38,0x53,0x4f,0x7d,0xa1,0xe4,0xc3,0x03,0xd2,0xa3,0x18,0xa7,0x28,
1567                         0xc3,0xc0,0xc9,0x51,0x56,0x80,0x95,0x39,0xfc,0xf0,0xe2,0x42,0x9a,0x6b,0x52,0x54,
1568                         0x16,0xae,0xdb,0xf5,0xa0,0xde,0x6a,0x57,0xa6,0x37,0xb3,0x9b},
1569                 C18[]= {0x5a,0x8d,0xef,0x2f,0x0c,0x9e,0x53,0xf1,0xf7,0x5d,0x78,0x53,0x65,0x9e,0x2a,0x20,
1570                         0xee,0xb2,0xb2,0x2a,0xaf,0xde,0x64,0x19,0xa0,0x58,0xab,0x4f,0x6f,0x74,0x6b,0xf4,
1571                         0x0f,0xc0,0xc3,0xb7,0x80,0xf2,0x44,0x45,0x2d,0xa3,0xeb,0xf1,0xc5,0xd8,0x2c,0xde,
1572                         0xa2,0x41,0x89,0x97,0x20,0x0e,0xf8,0x2e,0x44,0xae,0x7e,0x3f},
1573                 T18[]= {0xa4,0x4a,0x82,0x66,0xee,0x1c,0x8e,0xb0,0xc8,0xb5,0xd4,0xcf,0x5a,0xe9,0xf1,0x9a};
1574
1575 #define TEST_CASE(n)    do {                                    \
1576         u8 out[sizeof(P##n)];                                   \
1577         AES_set_encrypt_key(K##n,sizeof(K##n)*8,&key);          \
1578         CRYPTO_gcm128_init(&ctx,&key,(block128_f)AES_encrypt);  \
1579         CRYPTO_gcm128_setiv(&ctx,IV##n,sizeof(IV##n));          \
1580         memset(out,0,sizeof(out));                              \
1581         if (A##n) CRYPTO_gcm128_aad(&ctx,A##n,sizeof(A##n));    \
1582         if (P##n) CRYPTO_gcm128_encrypt(&ctx,P##n,out,sizeof(out));     \
1583         if (CRYPTO_gcm128_finish(&ctx,T##n,16) ||               \
1584             (C##n && memcmp(out,C##n,sizeof(out))))             \
1585                 ret++, printf ("encrypt test#%d failed.\n",n);  \
1586         CRYPTO_gcm128_setiv(&ctx,IV##n,sizeof(IV##n));          \
1587         memset(out,0,sizeof(out));                              \
1588         if (A##n) CRYPTO_gcm128_aad(&ctx,A##n,sizeof(A##n));    \
1589         if (C##n) CRYPTO_gcm128_decrypt(&ctx,C##n,out,sizeof(out));     \
1590         if (CRYPTO_gcm128_finish(&ctx,T##n,16) ||               \
1591             (P##n && memcmp(out,P##n,sizeof(out))))             \
1592                 ret++, printf ("decrypt test#%d failed.\n",n);  \
1593         } while(0)
1594
1595 int main()
1596 {
1597         GCM128_CONTEXT ctx;
1598         AES_KEY key;
1599         int ret=0;
1600
1601         TEST_CASE(1);
1602         TEST_CASE(2);
1603         TEST_CASE(3);
1604         TEST_CASE(4);
1605         TEST_CASE(5);
1606         TEST_CASE(6);
1607         TEST_CASE(7);
1608         TEST_CASE(8);
1609         TEST_CASE(9);
1610         TEST_CASE(10);
1611         TEST_CASE(11);
1612         TEST_CASE(12);
1613         TEST_CASE(13);
1614         TEST_CASE(14);
1615         TEST_CASE(15);
1616         TEST_CASE(16);
1617         TEST_CASE(17);
1618         TEST_CASE(18);
1619
1620 #ifdef OPENSSL_CPUID_OBJ
1621         {
1622         size_t start,stop,gcm_t,ctr_t,OPENSSL_rdtsc();
1623         union { u64 u; u8 c[1024]; } buf;
1624         int i;
1625
1626         AES_set_encrypt_key(K1,sizeof(K1)*8,&key);
1627         CRYPTO_gcm128_init(&ctx,&key,(block128_f)AES_encrypt);
1628         CRYPTO_gcm128_setiv(&ctx,IV1,sizeof(IV1));
1629
1630         CRYPTO_gcm128_encrypt(&ctx,buf.c,buf.c,sizeof(buf));
1631         start = OPENSSL_rdtsc();
1632         CRYPTO_gcm128_encrypt(&ctx,buf.c,buf.c,sizeof(buf));
1633         gcm_t = OPENSSL_rdtsc() - start;
1634
1635         CRYPTO_ctr128_encrypt(buf.c,buf.c,sizeof(buf),
1636                         &key,ctx.Yi.c,ctx.EKi.c,&ctx.mres,
1637                         (block128_f)AES_encrypt);
1638         start = OPENSSL_rdtsc();
1639         CRYPTO_ctr128_encrypt(buf.c,buf.c,sizeof(buf),
1640                         &key,ctx.Yi.c,ctx.EKi.c,&ctx.mres,
1641                         (block128_f)AES_encrypt);
1642         ctr_t = OPENSSL_rdtsc() - start;
1643
1644         printf("%.2f-%.2f=%.2f\n",
1645                         gcm_t/(double)sizeof(buf),
1646                         ctr_t/(double)sizeof(buf),
1647                         (gcm_t-ctr_t)/(double)sizeof(buf));
1648 #ifdef GHASH
1649         GHASH(&ctx,buf.c,sizeof(buf));
1650         start = OPENSSL_rdtsc();
1651         for (i=0;i<100;++i) GHASH(&ctx,buf.c,sizeof(buf));
1652         gcm_t = OPENSSL_rdtsc() - start;
1653         printf("%.2f\n",gcm_t/(double)sizeof(buf)/(double)i);
1654 #endif
1655         }
1656 #endif
1657
1658         return ret;
1659 }
1660 #endif