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 #define OPENSSL_FIPSAPI
  51
  52 #include <openssl/crypto.h>
  53 #include "modes_lcl.h"
  54 #include <string.h>
  55
  56 #ifndef MODES_DEBUG
  57 # ifndef NDEBUG
  58 #  define NDEBUG
  59 # endif
  60 #endif
  61 #include <assert.h>
  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 /*
  86  * Even though permitted values for TABLE_BITS are 8, 4 and 1, it should
  87  * never be set to 8. 8 is effectively reserved for testing purposes.
  88  * TABLE_BITS>1 are lookup-table-driven implementations referred to as
  89  * "Shoup's" in GCM specification. In other words OpenSSL does not cover
  90  * whole spectrum of possible table driven implementations. Why? In
  91  * non-"Shoup's" case memory access pattern is segmented in such manner,
  92  * that it's trivial to see that cache timing information can reveal
  93  * fair portion of intermediate hash value. Given that ciphertext is
  94  * always available to attacker, it's possible for him to attempt to
  95  * deduce secret parameter H and if successful, tamper with messages
  96  * [which is nothing but trivial in CTR mode]. In "Shoup's" case it's
  97  * not as trivial, but there is no reason to believe that it's resistant
  98  * to cache-timing attack. And the thing about "8-bit" implementation is
  99  * that it consumes 16 (sixteen) times more memory, 4KB per individual
 100  * key + 1KB shared. Well, on pros side it should be twice as fast as
 101  * "4-bit" version. And for gcc-generated x86[_64] code, "8-bit" version
 102  * was observed to run ~75% faster, closer to 100% for commercial
 103  * compilers... Yet "4-bit" procedure is preferred, because it's
 104  * believed to provide better security-performance balance and adequate
 105  * all-round performance. "All-round" refers to things like:
 106  *
 107  * - shorter setup time effectively improves overall timing for
 108  *   handling short messages;
 109  * - larger table allocation can become unbearable because of VM
 110  *   subsystem penalties (for example on Windows large enough free
 111  *   results in VM working set trimming, meaning that consequent
 112  *   malloc would immediately incur working set expansion);
 113  * - larger table has larger cache footprint, which can affect
 114  *   performance of other code paths (not necessarily even from same
 115  *   thread in Hyper-Threading world);
 116  *
 117  * Value of 1 is not appropriate for performance reasons.
 118  */
 119 #if     TABLE_BITS==8
 120
 121 static void gcm_init_8bit(u128 Htable[256], u64 H[2])
 122 {
 123         int  i, j;
 124         u128 V;
 125
 126         Htable[0].hi = 0;
 127         Htable[0].lo = 0;
 128         V.hi = H[0];
 129         V.lo = H[1];
 130
 131         for (Htable[128]=V, i=64; i>0; i>>=1) {
 132                 REDUCE1BIT(V);
 133                 Htable[i] = V;
 134         }
 135
 136         for (i=2; i<256; i<<=1) {
 137                 u128 *Hi = Htable+i, H0 = *Hi;
 138                 for (j=1; j<i; ++j) {
 139                         Hi[j].hi = H0.hi^Htable[j].hi;
 140                         Hi[j].lo = H0.lo^Htable[j].lo;
 141                 }
 142         }
 143 }
 144
 145 static void gcm_gmult_8bit(u64 Xi[2], const u128 Htable[256])
 146 {
 147         u128 Z = { 0, 0};
 148         const u8 *xi = (const u8 *)Xi+15;
 149         size_t rem, n = *xi;
 150         const union { long one; char little; } is_endian = {1};
 151         static const size_t rem_8bit[256] = {
 152                 PACK(0x0000), PACK(0x01C2), PACK(0x0384), PACK(0x0246),
 153                 PACK(0x0708), PACK(0x06CA), PACK(0x048C), PACK(0x054E),
 154                 PACK(0x0E10), PACK(0x0FD2), PACK(0x0D94), PACK(0x0C56),
 155                 PACK(0x0918), PACK(0x08DA), PACK(0x0A9C), PACK(0x0B5E),
 156                 PACK(0x1C20), PACK(0x1DE2), PACK(0x1FA4), PACK(0x1E66),
 157                 PACK(0x1B28), PACK(0x1AEA), PACK(0x18AC), PACK(0x196E),
 158                 PACK(0x1230), PACK(0x13F2), PACK(0x11B4), PACK(0x1076),
 159                 PACK(0x1538), PACK(0x14FA), PACK(0x16BC), PACK(0x177E),
 160                 PACK(0x3840), PACK(0x3982), PACK(0x3BC4), PACK(0x3A06),
 161                 PACK(0x3F48), PACK(0x3E8A), PACK(0x3CCC), PACK(0x3D0E),
 162                 PACK(0x3650), PACK(0x3792), PACK(0x35D4), PACK(0x3416),
 163                 PACK(0x3158), PACK(0x309A), PACK(0x32DC), PACK(0x331E),
 164                 PACK(0x2460), PACK(0x25A2), PACK(0x27E4), PACK(0x2626),
 165                 PACK(0x2368), PACK(0x22AA), PACK(0x20EC), PACK(0x212E),
 166                 PACK(0x2A70), PACK(0x2BB2), PACK(0x29F4), PACK(0x2836),
 167                 PACK(0x2D78), PACK(0x2CBA), PACK(0x2EFC), PACK(0x2F3E),
 168                 PACK(0x7080), PACK(0x7142), PACK(0x7304), PACK(0x72C6),
 169                 PACK(0x7788), PACK(0x764A), PACK(0x740C), PACK(0x75CE),
 170                 PACK(0x7E90), PACK(0x7F52), PACK(0x7D14), PACK(0x7CD6),
 171                 PACK(0x7998), PACK(0x785A), PACK(0x7A1C), PACK(0x7BDE),
 172                 PACK(0x6CA0), PACK(0x6D62), PACK(0x6F24), PACK(0x6EE6),
 173                 PACK(0x6BA8), PACK(0x6A6A), PACK(0x682C), PACK(0x69EE),
 174                 PACK(0x62B0), PACK(0x6372), PACK(0x6134), PACK(0x60F6),
 175                 PACK(0x65B8), PACK(0x647A), PACK(0x663C), PACK(0x67FE),
 176                 PACK(0x48C0), PACK(0x4902), PACK(0x4B44), PACK(0x4A86),
 177                 PACK(0x4FC8), PACK(0x4E0A), PACK(0x4C4C), PACK(0x4D8E),
 178                 PACK(0x46D0), PACK(0x4712), PACK(0x4554), PACK(0x4496),
 179                 PACK(0x41D8), PACK(0x401A), PACK(0x425C), PACK(0x439E),
 180                 PACK(0x54E0), PACK(0x5522), PACK(0x5764), PACK(0x56A6),
 181                 PACK(0x53E8), PACK(0x522A), PACK(0x506C), PACK(0x51AE),
 182                 PACK(0x5AF0), PACK(0x5B32), PACK(0x5974), PACK(0x58B6),
 183                 PACK(0x5DF8), PACK(0x5C3A), PACK(0x5E7C), PACK(0x5FBE),
 184                 PACK(0xE100), PACK(0xE0C2), PACK(0xE284), PACK(0xE346),
 185                 PACK(0xE608), PACK(0xE7CA), PACK(0xE58C), PACK(0xE44E),
 186                 PACK(0xEF10), PACK(0xEED2), PACK(0xEC94), PACK(0xED56),
 187                 PACK(0xE818), PACK(0xE9DA), PACK(0xEB9C), PACK(0xEA5E),
 188                 PACK(0xFD20), PACK(0xFCE2), PACK(0xFEA4), PACK(0xFF66),
 189                 PACK(0xFA28), PACK(0xFBEA), PACK(0xF9AC), PACK(0xF86E),
 190                 PACK(0xF330), PACK(0xF2F2), PACK(0xF0B4), PACK(0xF176),
 191                 PACK(0xF438), PACK(0xF5FA), PACK(0xF7BC), PACK(0xF67E),
 192                 PACK(0xD940), PACK(0xD882), PACK(0xDAC4), PACK(0xDB06),
 193                 PACK(0xDE48), PACK(0xDF8A), PACK(0xDDCC), PACK(0xDC0E),
 194                 PACK(0xD750), PACK(0xD692), PACK(0xD4D4), PACK(0xD516),
 195                 PACK(0xD058), PACK(0xD19A), PACK(0xD3DC), PACK(0xD21E),
 196                 PACK(0xC560), PACK(0xC4A2), PACK(0xC6E4), PACK(0xC726),
 197                 PACK(0xC268), PACK(0xC3AA), PACK(0xC1EC), PACK(0xC02E),
 198                 PACK(0xCB70), PACK(0xCAB2), PACK(0xC8F4), PACK(0xC936),
 199                 PACK(0xCC78), PACK(0xCDBA), PACK(0xCFFC), PACK(0xCE3E),
 200                 PACK(0x9180), PACK(0x9042), PACK(0x9204), PACK(0x93C6),
 201                 PACK(0x9688), PACK(0x974A), PACK(0x950C), PACK(0x94CE),
 202                 PACK(0x9F90), PACK(0x9E52), PACK(0x9C14), PACK(0x9DD6),
 203                 PACK(0x9898), PACK(0x995A), PACK(0x9B1C), PACK(0x9ADE),
 204                 PACK(0x8DA0), PACK(0x8C62), PACK(0x8E24), PACK(0x8FE6),
 205                 PACK(0x8AA8), PACK(0x8B6A), PACK(0x892C), PACK(0x88EE),
 206                 PACK(0x83B0), PACK(0x8272), PACK(0x8034), PACK(0x81F6),
 207                 PACK(0x84B8), PACK(0x857A), PACK(0x873C), PACK(0x86FE),
 208                 PACK(0xA9C0), PACK(0xA802), PACK(0xAA44), PACK(0xAB86),
 209                 PACK(0xAEC8), PACK(0xAF0A), PACK(0xAD4C), PACK(0xAC8E),
 210                 PACK(0xA7D0), PACK(0xA612), PACK(0xA454), PACK(0xA596),
 211                 PACK(0xA0D8), PACK(0xA11A), PACK(0xA35C), PACK(0xA29E),
 212                 PACK(0xB5E0), PACK(0xB422), PACK(0xB664), PACK(0xB7A6),
 213                 PACK(0xB2E8), PACK(0xB32A), PACK(0xB16C), PACK(0xB0AE),
 214                 PACK(0xBBF0), PACK(0xBA32), PACK(0xB874), PACK(0xB9B6),
 215                 PACK(0xBCF8), PACK(0xBD3A), PACK(0xBF7C), PACK(0xBEBE) };
 216
 217         while (1) {
 218                 Z.hi ^= Htable[n].hi;
 219                 Z.lo ^= Htable[n].lo;
 220
 221                 if ((u8 *)Xi==xi)       break;
 222
 223                 n = *(--xi);
 224
 225                 rem  = (size_t)Z.lo&0xff;
 226                 Z.lo = (Z.hi<<56)|(Z.lo>>8);
 227                 Z.hi = (Z.hi>>8);
 228                 if (sizeof(size_t)==8)
 229                         Z.hi ^= rem_8bit[rem];
 230                 else
 231                         Z.hi ^= (u64)rem_8bit[rem]<<32;
 232         }
 233
 234         if (is_endian.little) {
 235 #ifdef BSWAP8
 236                 Xi[0] = BSWAP8(Z.hi);
 237                 Xi[1] = BSWAP8(Z.lo);
 238 #else
 239                 u8 *p = (u8 *)Xi;
 240                 u32 v;
 241                 v = (u32)(Z.hi>>32);    PUTU32(p,v);
 242                 v = (u32)(Z.hi);        PUTU32(p+4,v);
 243                 v = (u32)(Z.lo>>32);    PUTU32(p+8,v);
 244                 v = (u32)(Z.lo);        PUTU32(p+12,v);
 245 #endif
 246         }
 247         else {
 248                 Xi[0] = Z.hi;
 249                 Xi[1] = Z.lo;
 250         }
 251 }
 252 #define GCM_MUL(ctx,Xi)   gcm_gmult_8bit(ctx->Xi.u,ctx->Htable)
 253
 254 #elif   TABLE_BITS==4
 255
 256 static void gcm_init_4bit(u128 Htable[16], u64 H[2])
 257 {
 258         u128 V;
 259 #if defined(OPENSSL_SMALL_FOOTPRINT)
 260         int  i;
 261 #endif
 262
 263         Htable[0].hi = 0;
 264         Htable[0].lo = 0;
 265         V.hi = H[0];
 266         V.lo = H[1];
 267
 268 #if defined(OPENSSL_SMALL_FOOTPRINT)
 269         for (Htable[8]=V, i=4; i>0; i>>=1) {
 270                 REDUCE1BIT(V);
 271                 Htable[i] = V;
 272         }
 273
 274         for (i=2; i<16; i<<=1) {
 275                 u128 *Hi = Htable+i;
 276                 int   j;
 277                 for (V=*Hi, j=1; j<i; ++j) {
 278                         Hi[j].hi = V.hi^Htable[j].hi;
 279                         Hi[j].lo = V.lo^Htable[j].lo;
 280                 }
 281         }
 282 #else
 283         Htable[8] = V;
 284         REDUCE1BIT(V);
 285         Htable[4] = V;
 286         REDUCE1BIT(V);
 287         Htable[2] = V;
 288         REDUCE1BIT(V);
 289         Htable[1] = V;
 290         Htable[3].hi  = V.hi^Htable[2].hi, Htable[3].lo  = V.lo^Htable[2].lo;
 291         V=Htable[4];
 292         Htable[5].hi  = V.hi^Htable[1].hi, Htable[5].lo  = V.lo^Htable[1].lo;
 293         Htable[6].hi  = V.hi^Htable[2].hi, Htable[6].lo  = V.lo^Htable[2].lo;
 294         Htable[7].hi  = V.hi^Htable[3].hi, Htable[7].lo  = V.lo^Htable[3].lo;
 295         V=Htable[8];
 296         Htable[9].hi  = V.hi^Htable[1].hi, Htable[9].lo  = V.lo^Htable[1].lo;
 297         Htable[10].hi = V.hi^Htable[2].hi, Htable[10].lo = V.lo^Htable[2].lo;
 298         Htable[11].hi = V.hi^Htable[3].hi, Htable[11].lo = V.lo^Htable[3].lo;
 299         Htable[12].hi = V.hi^Htable[4].hi, Htable[12].lo = V.lo^Htable[4].lo;
 300         Htable[13].hi = V.hi^Htable[5].hi, Htable[13].lo = V.lo^Htable[5].lo;
 301         Htable[14].hi = V.hi^Htable[6].hi, Htable[14].lo = V.lo^Htable[6].lo;
 302         Htable[15].hi = V.hi^Htable[7].hi, Htable[15].lo = V.lo^Htable[7].lo;
 303 #endif
 304 #if defined(GHASH_ASM) && (defined(__arm__) || defined(__arm))
 305         /*
 306          * ARM assembler expects specific dword order in Htable.
 307          */
 308         {
 309         int j;
 310         const union { long one; char little; } is_endian = {1};
 311
 312         if (is_endian.little)
 313                 for (j=0;j<16;++j) {
 314                         V = Htable[j];
 315                         Htable[j].hi = V.lo;
 316                         Htable[j].lo = V.hi;
 317                 }
 318         else
 319                 for (j=0;j<16;++j) {
 320                         V = Htable[j];
 321                         Htable[j].hi = V.lo<<32|V.lo>>32;
 322                         Htable[j].lo = V.hi<<32|V.hi>>32;
 323                 }
 324         }
 325 #endif
 326 }
 327
 328 #ifndef GHASH_ASM
 329 static const size_t rem_4bit[16] = {
 330         PACK(0x0000), PACK(0x1C20), PACK(0x3840), PACK(0x2460),
 331         PACK(0x7080), PACK(0x6CA0), PACK(0x48C0), PACK(0x54E0),
 332         PACK(0xE100), PACK(0xFD20), PACK(0xD940), PACK(0xC560),
 333         PACK(0x9180), PACK(0x8DA0), PACK(0xA9C0), PACK(0xB5E0) };
 334
 335 static void gcm_gmult_4bit(u64 Xi[2], const u128 Htable[16])
 336 {
 337         u128 Z;
 338         int cnt = 15;
 339         size_t rem, nlo, nhi;
 340         const union { long one; char little; } is_endian = {1};
 341
 342         nlo  = ((const u8 *)Xi)[15];
 343         nhi  = nlo>>4;
 344         nlo &= 0xf;
 345
 346         Z.hi = Htable[nlo].hi;
 347         Z.lo = Htable[nlo].lo;
 348
 349         while (1) {
 350                 rem  = (size_t)Z.lo&0xf;
 351                 Z.lo = (Z.hi<<60)|(Z.lo>>4);
 352                 Z.hi = (Z.hi>>4);
 353                 if (sizeof(size_t)==8)
 354                         Z.hi ^= rem_4bit[rem];
 355                 else
 356                         Z.hi ^= (u64)rem_4bit[rem]<<32;
 357
 358                 Z.hi ^= Htable[nhi].hi;
 359                 Z.lo ^= Htable[nhi].lo;
 360
 361                 if (--cnt<0)            break;
 362
 363                 nlo  = ((const u8 *)Xi)[cnt];
 364                 nhi  = nlo>>4;
 365                 nlo &= 0xf;
 366
 367                 rem  = (size_t)Z.lo&0xf;
 368                 Z.lo = (Z.hi<<60)|(Z.lo>>4);
 369                 Z.hi = (Z.hi>>4);
 370                 if (sizeof(size_t)==8)
 371                         Z.hi ^= rem_4bit[rem];
 372                 else
 373                         Z.hi ^= (u64)rem_4bit[rem]<<32;
 374
 375                 Z.hi ^= Htable[nlo].hi;
 376                 Z.lo ^= Htable[nlo].lo;
 377         }
 378
 379         if (is_endian.little) {
 380 #ifdef BSWAP8
 381                 Xi[0] = BSWAP8(Z.hi);
 382                 Xi[1] = BSWAP8(Z.lo);
 383 #else
 384                 u8 *p = (u8 *)Xi;
 385                 u32 v;
 386                 v = (u32)(Z.hi>>32);    PUTU32(p,v);
 387                 v = (u32)(Z.hi);        PUTU32(p+4,v);
 388                 v = (u32)(Z.lo>>32);    PUTU32(p+8,v);
 389                 v = (u32)(Z.lo);        PUTU32(p+12,v);
 390 #endif
 391         }
 392         else {
 393                 Xi[0] = Z.hi;
 394                 Xi[1] = Z.lo;
 395         }
 396 }
 397
 398 #if !defined(OPENSSL_SMALL_FOOTPRINT)
 399 /*
 400  * Streamed gcm_mult_4bit, see CRYPTO_gcm128_[en|de]crypt for
 401  * details... Compiler-generated code doesn't seem to give any
 402  * performance improvement, at least not on x86[_64]. It's here
 403  * mostly as reference and a placeholder for possible future
 404  * non-trivial optimization[s]...
 405  */
 406 static void gcm_ghash_4bit(u64 Xi[2],const u128 Htable[16],
 407                                 const u8 *inp,size_t len)
 408 {
 409     u128 Z;
 410     int cnt;
 411     size_t rem, nlo, nhi;
 412     const union { long one; char little; } is_endian = {1};
 413
 414 #if 1
 415     do {
 416         cnt  = 15;
 417         nlo  = ((const u8 *)Xi)[15];
 418         nlo ^= inp[15];
 419         nhi  = nlo>>4;
 420         nlo &= 0xf;
 421
 422         Z.hi = Htable[nlo].hi;
 423         Z.lo = Htable[nlo].lo;
 424
 425         while (1) {
 426                 rem  = (size_t)Z.lo&0xf;
 427                 Z.lo = (Z.hi<<60)|(Z.lo>>4);
 428                 Z.hi = (Z.hi>>4);
 429                 if (sizeof(size_t)==8)
 430                         Z.hi ^= rem_4bit[rem];
 431                 else
 432                         Z.hi ^= (u64)rem_4bit[rem]<<32;
 433
 434                 Z.hi ^= Htable[nhi].hi;
 435                 Z.lo ^= Htable[nhi].lo;
 436
 437                 if (--cnt<0)            break;
 438
 439                 nlo  = ((const u8 *)Xi)[cnt];
 440                 nlo ^= inp[cnt];
 441                 nhi  = nlo>>4;
 442                 nlo &= 0xf;
 443
 444                 rem  = (size_t)Z.lo&0xf;
 445                 Z.lo = (Z.hi<<60)|(Z.lo>>4);
 446                 Z.hi = (Z.hi>>4);
 447                 if (sizeof(size_t)==8)
 448                         Z.hi ^= rem_4bit[rem];
 449                 else
 450                         Z.hi ^= (u64)rem_4bit[rem]<<32;
 451
 452                 Z.hi ^= Htable[nlo].hi;
 453                 Z.lo ^= Htable[nlo].lo;
 454         }
 455 #else
 456     /*
 457      * Extra 256+16 bytes per-key plus 512 bytes shared tables
 458      * [should] give ~50% improvement... One could have PACK()-ed
 459      * the rem_8bit even here, but the priority is to minimize
 460      * cache footprint...
 461      */
 462     u128 Hshr4[16];     /* Htable shifted right by 4 bits */
 463     u8   Hshl4[16];     /* Htable shifted left  by 4 bits */
 464     static const unsigned short rem_8bit[256] = {
 465         0x0000, 0x01C2, 0x0384, 0x0246, 0x0708, 0x06CA, 0x048C, 0x054E,
 466         0x0E10, 0x0FD2, 0x0D94, 0x0C56, 0x0918, 0x08DA, 0x0A9C, 0x0B5E,
 467         0x1C20, 0x1DE2, 0x1FA4, 0x1E66, 0x1B28, 0x1AEA, 0x18AC, 0x196E,
 468         0x1230, 0x13F2, 0x11B4, 0x1076, 0x1538, 0x14FA, 0x16BC, 0x177E,
 469         0x3840, 0x3982, 0x3BC4, 0x3A06, 0x3F48, 0x3E8A, 0x3CCC, 0x3D0E,
 470         0x3650, 0x3792, 0x35D4, 0x3416, 0x3158, 0x309A, 0x32DC, 0x331E,
 471         0x2460, 0x25A2, 0x27E4, 0x2626, 0x2368, 0x22AA, 0x20EC, 0x212E,
 472         0x2A70, 0x2BB2, 0x29F4, 0x2836, 0x2D78, 0x2CBA, 0x2EFC, 0x2F3E,
 473         0x7080, 0x7142, 0x7304, 0x72C6, 0x7788, 0x764A, 0x740C, 0x75CE,
 474         0x7E90, 0x7F52, 0x7D14, 0x7CD6, 0x7998, 0x785A, 0x7A1C, 0x7BDE,
 475         0x6CA0, 0x6D62, 0x6F24, 0x6EE6, 0x6BA8, 0x6A6A, 0x682C, 0x69EE,
 476         0x62B0, 0x6372, 0x6134, 0x60F6, 0x65B8, 0x647A, 0x663C, 0x67FE,
 477         0x48C0, 0x4902, 0x4B44, 0x4A86, 0x4FC8, 0x4E0A, 0x4C4C, 0x4D8E,
 478         0x46D0, 0x4712, 0x4554, 0x4496, 0x41D8, 0x401A, 0x425C, 0x439E,
 479         0x54E0, 0x5522, 0x5764, 0x56A6, 0x53E8, 0x522A, 0x506C, 0x51AE,
 480         0x5AF0, 0x5B32, 0x5974, 0x58B6, 0x5DF8, 0x5C3A, 0x5E7C, 0x5FBE,
 481         0xE100, 0xE0C2, 0xE284, 0xE346, 0xE608, 0xE7CA, 0xE58C, 0xE44E,
 482         0xEF10, 0xEED2, 0xEC94, 0xED56, 0xE818, 0xE9DA, 0xEB9C, 0xEA5E,
 483         0xFD20, 0xFCE2, 0xFEA4, 0xFF66, 0xFA28, 0xFBEA, 0xF9AC, 0xF86E,
 484         0xF330, 0xF2F2, 0xF0B4, 0xF176, 0xF438, 0xF5FA, 0xF7BC, 0xF67E,
 485         0xD940, 0xD882, 0xDAC4, 0xDB06, 0xDE48, 0xDF8A, 0xDDCC, 0xDC0E,
 486         0xD750, 0xD692, 0xD4D4, 0xD516, 0xD058, 0xD19A, 0xD3DC, 0xD21E,
 487         0xC560, 0xC4A2, 0xC6E4, 0xC726, 0xC268, 0xC3AA, 0xC1EC, 0xC02E,
 488         0xCB70, 0xCAB2, 0xC8F4, 0xC936, 0xCC78, 0xCDBA, 0xCFFC, 0xCE3E,
 489         0x9180, 0x9042, 0x9204, 0x93C6, 0x9688, 0x974A, 0x950C, 0x94CE,
 490         0x9F90, 0x9E52, 0x9C14, 0x9DD6, 0x9898, 0x995A, 0x9B1C, 0x9ADE,
 491         0x8DA0, 0x8C62, 0x8E24, 0x8FE6, 0x8AA8, 0x8B6A, 0x892C, 0x88EE,
 492         0x83B0, 0x8272, 0x8034, 0x81F6, 0x84B8, 0x857A, 0x873C, 0x86FE,
 493         0xA9C0, 0xA802, 0xAA44, 0xAB86, 0xAEC8, 0xAF0A, 0xAD4C, 0xAC8E,
 494         0xA7D0, 0xA612, 0xA454, 0xA596, 0xA0D8, 0xA11A, 0xA35C, 0xA29E,
 495         0xB5E0, 0xB422, 0xB664, 0xB7A6, 0xB2E8, 0xB32A, 0xB16C, 0xB0AE,
 496         0xBBF0, 0xBA32, 0xB874, 0xB9B6, 0xBCF8, 0xBD3A, 0xBF7C, 0xBEBE };
 497     /*
 498      * This pre-processing phase slows down procedure by approximately
 499      * same time as it makes each loop spin faster. In other words
 500      * single block performance is approximately same as straightforward
 501      * "4-bit" implementation, and then it goes only faster...
 502      */
 503     for (cnt=0; cnt<16; ++cnt) {
 504         Z.hi = Htable[cnt].hi;
 505         Z.lo = Htable[cnt].lo;
 506         Hshr4[cnt].lo = (Z.hi<<60)|(Z.lo>>4);
 507         Hshr4[cnt].hi = (Z.hi>>4);
 508         Hshl4[cnt]    = (u8)(Z.lo<<4);
 509     }
 510
 511     do {
 512         for (Z.lo=0, Z.hi=0, cnt=15; cnt; --cnt) {
 513                 nlo  = ((const u8 *)Xi)[cnt];
 514                 nlo ^= inp[cnt];
 515                 nhi  = nlo>>4;
 516                 nlo &= 0xf;
 517
 518                 Z.hi ^= Htable[nlo].hi;
 519                 Z.lo ^= Htable[nlo].lo;
 520
 521                 rem = (size_t)Z.lo&0xff;
 522
 523                 Z.lo = (Z.hi<<56)|(Z.lo>>8);
 524                 Z.hi = (Z.hi>>8);
 525
 526                 Z.hi ^= Hshr4[nhi].hi;
 527                 Z.lo ^= Hshr4[nhi].lo;
 528                 Z.hi ^= (u64)rem_8bit[rem^Hshl4[nhi]]<<48;
 529         }
 530
 531         nlo  = ((const u8 *)Xi)[0];
 532         nlo ^= inp[0];
 533         nhi  = nlo>>4;
 534         nlo &= 0xf;
 535
 536         Z.hi ^= Htable[nlo].hi;
 537         Z.lo ^= Htable[nlo].lo;
 538
 539         rem = (size_t)Z.lo&0xf;
 540
 541         Z.lo = (Z.hi<<60)|(Z.lo>>4);
 542         Z.hi = (Z.hi>>4);
 543
 544         Z.hi ^= Htable[nhi].hi;
 545         Z.lo ^= Htable[nhi].lo;
 546         Z.hi ^= ((u64)rem_8bit[rem<<4])<<48;
 547 #endif
 548
 549         if (is_endian.little) {
 550 #ifdef BSWAP8
 551                 Xi[0] = BSWAP8(Z.hi);
 552                 Xi[1] = BSWAP8(Z.lo);
 553 #else
 554                 u8 *p = (u8 *)Xi;
 555                 u32 v;
 556                 v = (u32)(Z.hi>>32);    PUTU32(p,v);
 557                 v = (u32)(Z.hi);        PUTU32(p+4,v);
 558                 v = (u32)(Z.lo>>32);    PUTU32(p+8,v);
 559                 v = (u32)(Z.lo);        PUTU32(p+12,v);
 560 #endif
 561         }
 562         else {
 563                 Xi[0] = Z.hi;
 564                 Xi[1] = Z.lo;
 565         }
 566     } while (inp+=16, len-=16);
 567 }
 568 #endif
 569 #else
 570 void gcm_gmult_4bit(u64 Xi[2],const u128 Htable[16]);
 571 void gcm_ghash_4bit(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
 572 #endif
 573
 574 #define GCM_MUL(ctx,Xi)   gcm_gmult_4bit(ctx->Xi.u,ctx->Htable)
 575 #if defined(GHASH_ASM) || !defined(OPENSSL_SMALL_FOOTPRINT)
 576 #define GHASH(ctx,in,len) gcm_ghash_4bit((ctx)->Xi.u,(ctx)->Htable,in,len)
 577 /* GHASH_CHUNK is "stride parameter" missioned to mitigate cache
 578  * trashing effect. In other words idea is to hash data while it's
 579  * still in L1 cache after encryption pass... */
 580 #define GHASH_CHUNK       (3*1024)
 581 #endif
 582
 583 #else   /* TABLE_BITS */
 584
 585 static void gcm_gmult_1bit(u64 Xi[2],const u64 H[2])
 586 {
 587         u128 V,Z = { 0,0 };
 588         long X;
 589         int  i,j;
 590         const long *xi = (const long *)Xi;
 591         const union { long one; char little; } is_endian = {1};
 592
 593         V.hi = H[0];    /* H is in host byte order, no byte swapping */
 594         V.lo = H[1];
 595
 596         for (j=0; j<16/sizeof(long); ++j) {
 597                 if (is_endian.little) {
 598                         if (sizeof(long)==8) {
 599 #ifdef BSWAP8
 600                                 X = (long)(BSWAP8(xi[j]));
 601 #else
 602                                 const u8 *p = (const u8 *)(xi+j);
 603                                 X = (long)((u64)GETU32(p)<<32|GETU32(p+4));
 604 #endif
 605                         }
 606                         else {
 607                                 const u8 *p = (const u8 *)(xi+j);
 608                                 X = (long)GETU32(p);
 609                         }
 610                 }
 611                 else
 612                         X = xi[j];
 613
 614                 for (i=0; i<8*sizeof(long); ++i, X<<=1) {
 615                         u64 M = (u64)(X>>(8*sizeof(long)-1));
 616                         Z.hi ^= V.hi&M;
 617                         Z.lo ^= V.lo&M;
 618
 619                         REDUCE1BIT(V);
 620                 }
 621         }
 622
 623         if (is_endian.little) {
 624 #ifdef BSWAP8
 625                 Xi[0] = BSWAP8(Z.hi);
 626                 Xi[1] = BSWAP8(Z.lo);
 627 #else
 628                 u8 *p = (u8 *)Xi;
 629                 u32 v;
 630                 v = (u32)(Z.hi>>32);    PUTU32(p,v);
 631                 v = (u32)(Z.hi);        PUTU32(p+4,v);
 632                 v = (u32)(Z.lo>>32);    PUTU32(p+8,v);
 633                 v = (u32)(Z.lo);        PUTU32(p+12,v);
 634 #endif
 635         }
 636         else {
 637                 Xi[0] = Z.hi;
 638                 Xi[1] = Z.lo;
 639         }
 640 }
 641 #define GCM_MUL(ctx,Xi)   gcm_gmult_1bit(ctx->Xi.u,ctx->H.u)
 642
 643 #endif
 644
 645 #if     TABLE_BITS==4 && defined(GHASH_ASM)
 646 # if    !defined(I386_ONLY) && \
 647         (defined(__i386)        || defined(__i386__)    || \
 648          defined(__x86_64)      || defined(__x86_64__)  || \
 649          defined(_M_IX86)       || defined(_M_AMD64)    || defined(_M_X64))
 650 #  define GHASH_ASM_X86_OR_64
 651 #  define GCM_FUNCREF_4BIT
 652 extern unsigned int OPENSSL_ia32cap_P[2];
 653
 654 void gcm_init_clmul(u128 Htable[16],const u64 Xi[2]);
 655 void gcm_gmult_clmul(u64 Xi[2],const u128 Htable[16]);
 656 void gcm_ghash_clmul(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
 657
 658 #  if   defined(__i386) || defined(__i386__) || defined(_M_IX86)
 659 #   define GHASH_ASM_X86
 660 void gcm_gmult_4bit_mmx(u64 Xi[2],const u128 Htable[16]);
 661 void gcm_ghash_4bit_mmx(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
 662
 663 void gcm_gmult_4bit_x86(u64 Xi[2],const u128 Htable[16]);
 664 void gcm_ghash_4bit_x86(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
 665 #  endif
 666 # elif defined(__arm__) || defined(__arm)
 667 #  include "arm_arch.h"
 668 #  if __ARM_ARCH__>=7
 669 #   define GHASH_ASM_ARM
 670 #   define GCM_FUNCREF_4BIT
 671 void gcm_gmult_neon(u64 Xi[2],const u128 Htable[16]);
 672 void gcm_ghash_neon(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len);
 673 #  endif
 674 # endif
 675 #endif
 676
 677 #ifdef GCM_FUNCREF_4BIT
 678 # undef  GCM_MUL
 679 # define GCM_MUL(ctx,Xi)        (*gcm_gmult_p)(ctx->Xi.u,ctx->Htable)
 680 # ifdef GHASH
 681 #  undef  GHASH
 682 #  define GHASH(ctx,in,len)     (*gcm_ghash_p)(ctx->Xi.u,ctx->Htable,in,len)
 683 # endif
 684 #endif
 685
 686 void CRYPTO_gcm128_init(GCM128_CONTEXT *ctx,void *key,block128_f block)
 687 {
 688         const union { long one; char little; } is_endian = {1};
 689
 690         memset(ctx,0,sizeof(*ctx));
 691         ctx->block = block;
 692         ctx->key   = key;
 693
 694         (*block)(ctx->H.c,ctx->H.c,key);
 695
 696         if (is_endian.little) {
 697                 /* H is stored in host byte order */
 698 #ifdef BSWAP8
 699                 ctx->H.u[0] = BSWAP8(ctx->H.u[0]);
 700                 ctx->H.u[1] = BSWAP8(ctx->H.u[1]);
 701 #else
 702                 u8 *p = ctx->H.c;
 703                 u64 hi,lo;
 704                 hi = (u64)GETU32(p)  <<32|GETU32(p+4);
 705                 lo = (u64)GETU32(p+8)<<32|GETU32(p+12);
 706                 ctx->H.u[0] = hi;
 707                 ctx->H.u[1] = lo;
 708 #endif
 709         }
 710
 711 #if     TABLE_BITS==8
 712         gcm_init_8bit(ctx->Htable,ctx->H.u);
 713 #elif   TABLE_BITS==4
 714 # if    defined(GHASH_ASM_X86_OR_64)
 715 #  if   !defined(GHASH_ASM_X86) || defined(OPENSSL_IA32_SSE2)
 716         if (OPENSSL_ia32cap_P[0]&(1<<24) &&     /* check FXSR bit */
 717             OPENSSL_ia32cap_P[1]&(1<<1) ) {     /* check PCLMULQDQ bit */
 718                 gcm_init_clmul(ctx->Htable,ctx->H.u);
 719                 ctx->gmult = gcm_gmult_clmul;
 720                 ctx->ghash = gcm_ghash_clmul;
 721                 return;
 722         }
 723 #  endif
 724         gcm_init_4bit(ctx->Htable,ctx->H.u);
 725 #  if   defined(GHASH_ASM_X86)                  /* x86 only */
 726 #   if  defined(OPENSSL_IA32_SSE2)
 727         if (OPENSSL_ia32cap_P[0]&(1<<25)) {     /* check SSE bit */
 728 #   else
 729         if (OPENSSL_ia32cap_P[0]&(1<<23)) {     /* check MMX bit */
 730 #   endif
 731                 ctx->gmult = gcm_gmult_4bit_mmx;
 732                 ctx->ghash = gcm_ghash_4bit_mmx;
 733         } else {
 734                 ctx->gmult = gcm_gmult_4bit_x86;
 735                 ctx->ghash = gcm_ghash_4bit_x86;
 736         }
 737 #  else
 738         ctx->gmult = gcm_gmult_4bit;
 739         ctx->ghash = gcm_ghash_4bit;
 740 #  endif
 741 # elif  defined(GHASH_ASM_ARM)
 742         if (OPENSSL_armcap_P & ARMV7_NEON) {
 743                 ctx->gmult = gcm_gmult_neon;
 744                 ctx->ghash = gcm_ghash_neon;
 745         } else {
 746                 gcm_init_4bit(ctx->Htable,ctx->H.u);
 747                 ctx->gmult = gcm_gmult_4bit;
 748                 ctx->ghash = gcm_ghash_4bit;
 749         }
 750 # else
 751         gcm_init_4bit(ctx->Htable,ctx->H.u);
 752 # endif
 753 #endif
 754 }
 755
 756 void CRYPTO_gcm128_setiv(GCM128_CONTEXT *ctx,const unsigned char *iv,size_t len)
 757 {
 758         const union { long one; char little; } is_endian = {1};
 759         unsigned int ctr;
 760 #ifdef GCM_FUNCREF_4BIT
 761         void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16])    = ctx->gmult;
 762 #endif
 763
 764         ctx->Yi.u[0]  = 0;
 765         ctx->Yi.u[1]  = 0;
 766         ctx->Xi.u[0]  = 0;
 767         ctx->Xi.u[1]  = 0;
 768         ctx->len.u[0] = 0;      /* AAD length */
 769         ctx->len.u[1] = 0;      /* message length */
 770         ctx->ares = 0;
 771         ctx->mres = 0;
 772
 773         if (len==12) {
 774                 memcpy(ctx->Yi.c,iv,12);
 775                 ctx->Yi.c[15]=1;
 776                 ctr=1;
 777         }
 778         else {
 779                 size_t i;
 780                 u64 len0 = len;
 781
 782                 while (len>=16) {
 783                         for (i=0; i<16; ++i) ctx->Yi.c[i] ^= iv[i];
 784                         GCM_MUL(ctx,Yi);
 785                         iv += 16;
 786                         len -= 16;
 787                 }
 788                 if (len) {
 789                         for (i=0; i<len; ++i) ctx->Yi.c[i] ^= iv[i];
 790                         GCM_MUL(ctx,Yi);
 791                 }
 792                 len0 <<= 3;
 793                 if (is_endian.little) {
 794 #ifdef BSWAP8
 795                         ctx->Yi.u[1]  ^= BSWAP8(len0);
 796 #else
 797                         ctx->Yi.c[8]  ^= (u8)(len0>>56);
 798                         ctx->Yi.c[9]  ^= (u8)(len0>>48);
 799                         ctx->Yi.c[10] ^= (u8)(len0>>40);
 800                         ctx->Yi.c[11] ^= (u8)(len0>>32);
 801                         ctx->Yi.c[12] ^= (u8)(len0>>24);
 802                         ctx->Yi.c[13] ^= (u8)(len0>>16);
 803                         ctx->Yi.c[14] ^= (u8)(len0>>8);
 804                         ctx->Yi.c[15] ^= (u8)(len0);
 805 #endif
 806                 }
 807                 else
 808                         ctx->Yi.u[1]  ^= len0;
 809
 810                 GCM_MUL(ctx,Yi);
 811
 812                 if (is_endian.little)
 813                         ctr = GETU32(ctx->Yi.c+12);
 814                 else
 815                         ctr = ctx->Yi.d[3];
 816         }
 817
 818         (*ctx->block)(ctx->Yi.c,ctx->EK0.c,ctx->key);
 819         ++ctr;
 820         if (is_endian.little)
 821                 PUTU32(ctx->Yi.c+12,ctr);
 822         else
 823                 ctx->Yi.d[3] = ctr;
 824 }
 825
 826 int CRYPTO_gcm128_aad(GCM128_CONTEXT *ctx,const unsigned char *aad,size_t len)
 827 {
 828         size_t i;
 829         unsigned int n;
 830         u64 alen = ctx->len.u[0];
 831 #ifdef GCM_FUNCREF_4BIT
 832         void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16])    = ctx->gmult;
 833 # ifdef GHASH
 834         void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16],
 835                                 const u8 *inp,size_t len)       = ctx->ghash;
 836 # endif
 837 #endif
 838
 839         if (ctx->len.u[1]) return -2;
 840
 841         alen += len;
 842         if (alen>(U64(1)<<61) || (sizeof(len)==8 && alen<len))
 843                 return -1;
 844         ctx->len.u[0] = alen;
 845
 846         n = ctx->ares;
 847         if (n) {
 848                 while (n && len) {
 849                         ctx->Xi.c[n] ^= *(aad++);
 850                         --len;
 851                         n = (n+1)%16;
 852                 }
 853                 if (n==0) GCM_MUL(ctx,Xi);
 854                 else {
 855                         ctx->ares = n;
 856                         return 0;
 857                 }
 858         }
 859
 860 #ifdef GHASH
 861         if ((i = (len&(size_t)-16))) {
 862                 GHASH(ctx,aad,i);
 863                 aad += i;
 864                 len -= i;
 865         }
 866 #else
 867         while (len>=16) {
 868                 for (i=0; i<16; ++i) ctx->Xi.c[i] ^= aad[i];
 869                 GCM_MUL(ctx,Xi);
 870                 aad += 16;
 871                 len -= 16;
 872         }
 873 #endif
 874         if (len) {
 875                 n = (unsigned int)len;
 876                 for (i=0; i<len; ++i) ctx->Xi.c[i] ^= aad[i];
 877         }
 878
 879         ctx->ares = n;
 880         return 0;
 881 }
 882
 883 int CRYPTO_gcm128_encrypt(GCM128_CONTEXT *ctx,
 884                 const unsigned char *in, unsigned char *out,
 885                 size_t len)
 886 {
 887         const union { long one; char little; } is_endian = {1};
 888         unsigned int n, ctr;
 889         size_t i;
 890         u64        mlen  = ctx->len.u[1];
 891         block128_f block = ctx->block;
 892         void      *key   = ctx->key;
 893 #ifdef GCM_FUNCREF_4BIT
 894         void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16])    = ctx->gmult;
 895 # ifdef GHASH
 896         void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16],
 897                                 const u8 *inp,size_t len)       = ctx->ghash;
 898 # endif
 899 #endif
 900
 901 #if 0
 902         n = (unsigned int)mlen%16; /* alternative to ctx->mres */
 903 #endif
 904         mlen += len;
 905         if (mlen>((U64(1)<<36)-32) || (sizeof(len)==8 && mlen<len))
 906                 return -1;
 907         ctx->len.u[1] = mlen;
 908
 909         if (ctx->ares) {
 910                 /* First call to encrypt finalizes GHASH(AAD) */
 911                 GCM_MUL(ctx,Xi);
 912                 ctx->ares = 0;
 913         }
 914
 915         if (is_endian.little)
 916                 ctr = GETU32(ctx->Yi.c+12);
 917         else
 918                 ctr = ctx->Yi.d[3];
 919
 920         n = ctx->mres;
 921 #if !defined(OPENSSL_SMALL_FOOTPRINT)
 922         if (16%sizeof(size_t) == 0) do {        /* always true actually */
 923                 if (n) {
 924                         while (n && len) {
 925                                 ctx->Xi.c[n] ^= *(out++) = *(in++)^ctx->EKi.c[n];
 926                                 --len;
 927                                 n = (n+1)%16;
 928                         }
 929                         if (n==0) GCM_MUL(ctx,Xi);
 930                         else {
 931                                 ctx->mres = n;
 932                                 return 0;
 933                         }
 934                 }
 935 #if defined(STRICT_ALIGNMENT)
 936                 if (((size_t)in|(size_t)out)%sizeof(size_t) != 0)
 937                         break;
 938 #endif
 939 #if defined(GHASH) && defined(GHASH_CHUNK)
 940                 while (len>=GHASH_CHUNK) {
 941                     size_t j=GHASH_CHUNK;
 942
 943                     while (j) {
 944                         (*block)(ctx->Yi.c,ctx->EKi.c,key);
 945                         ++ctr;
 946                         if (is_endian.little)
 947                                 PUTU32(ctx->Yi.c+12,ctr);
 948                         else
 949                                 ctx->Yi.d[3] = ctr;
 950                         for (i=0; i<16; i+=sizeof(size_t))
 951                                 *(size_t *)(out+i) =
 952                                 *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
 953                         out += 16;
 954                         in  += 16;
 955                         j   -= 16;
 956                     }
 957                     GHASH(ctx,out-GHASH_CHUNK,GHASH_CHUNK);
 958                     len -= GHASH_CHUNK;
 959                 }
 960                 if ((i = (len&(size_t)-16))) {
 961                     size_t j=i;
 962
 963                     while (len>=16) {
 964                         (*block)(ctx->Yi.c,ctx->EKi.c,key);
 965                         ++ctr;
 966                         if (is_endian.little)
 967                                 PUTU32(ctx->Yi.c+12,ctr);
 968                         else
 969                                 ctx->Yi.d[3] = ctr;
 970                         for (i=0; i<16; i+=sizeof(size_t))
 971                                 *(size_t *)(out+i) =
 972                                 *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
 973                         out += 16;
 974                         in  += 16;
 975                         len -= 16;
 976                     }
 977                     GHASH(ctx,out-j,j);
 978                 }
 979 #else
 980                 while (len>=16) {
 981                         (*block)(ctx->Yi.c,ctx->EKi.c,key);
 982                         ++ctr;
 983                         if (is_endian.little)
 984                                 PUTU32(ctx->Yi.c+12,ctr);
 985                         else
 986                                 ctx->Yi.d[3] = ctr;
 987                         for (i=0; i<16; i+=sizeof(size_t))
 988                                 *(size_t *)(ctx->Xi.c+i) ^=
 989                                 *(size_t *)(out+i) =
 990                                 *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
 991                         GCM_MUL(ctx,Xi);
 992                         out += 16;
 993                         in  += 16;
 994                         len -= 16;
 995                 }
 996 #endif
 997                 if (len) {
 998                         (*block)(ctx->Yi.c,ctx->EKi.c,key);
 999                         ++ctr;
1000                         if (is_endian.little)
1001                                 PUTU32(ctx->Yi.c+12,ctr);
1002                         else
1003                                 ctx->Yi.d[3] = ctr;
1004                         while (len--) {
1005                                 ctx->Xi.c[n] ^= out[n] = in[n]^ctx->EKi.c[n];
1006                                 ++n;
1007                         }
1008                 }
1009
1010                 ctx->mres = n;
1011                 return 0;
1012         } while(0);
1013 #endif
1014         for (i=0;i<len;++i) {
1015                 if (n==0) {
1016                         (*block)(ctx->Yi.c,ctx->EKi.c,key);
1017                         ++ctr;
1018                         if (is_endian.little)
1019                                 PUTU32(ctx->Yi.c+12,ctr);
1020                         else
1021                                 ctx->Yi.d[3] = ctr;
1022                 }
1023                 ctx->Xi.c[n] ^= out[i] = in[i]^ctx->EKi.c[n];
1024                 n = (n+1)%16;
1025                 if (n==0)
1026                         GCM_MUL(ctx,Xi);
1027         }
1028
1029         ctx->mres = n;
1030         return 0;
1031 }
1032
1033 int CRYPTO_gcm128_decrypt(GCM128_CONTEXT *ctx,
1034                 const unsigned char *in, unsigned char *out,
1035                 size_t len)
1036 {
1037         const union { long one; char little; } is_endian = {1};
1038         unsigned int n, ctr;
1039         size_t i;
1040         u64        mlen  = ctx->len.u[1];
1041         block128_f block = ctx->block;
1042         void      *key   = ctx->key;
1043 #ifdef GCM_FUNCREF_4BIT
1044         void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16])    = ctx->gmult;
1045 # ifdef GHASH
1046         void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16],
1047                                 const u8 *inp,size_t len)       = ctx->ghash;
1048 # endif
1049 #endif
1050
1051         mlen += len;
1052         if (mlen>((U64(1)<<36)-32) || (sizeof(len)==8 && mlen<len))
1053                 return -1;
1054         ctx->len.u[1] = mlen;
1055
1056         if (ctx->ares) {
1057                 /* First call to decrypt finalizes GHASH(AAD) */
1058                 GCM_MUL(ctx,Xi);
1059                 ctx->ares = 0;
1060         }
1061
1062         if (is_endian.little)
1063                 ctr = GETU32(ctx->Yi.c+12);
1064         else
1065                 ctr = ctx->Yi.d[3];
1066
1067         n = ctx->mres;
1068 #if !defined(OPENSSL_SMALL_FOOTPRINT)
1069         if (16%sizeof(size_t) == 0) do {        /* always true actually */
1070                 if (n) {
1071                         while (n && len) {
1072                                 u8 c = *(in++);
1073                                 *(out++) = c^ctx->EKi.c[n];
1074                                 ctx->Xi.c[n] ^= c;
1075                                 --len;
1076                                 n = (n+1)%16;
1077                         }
1078                         if (n==0) GCM_MUL (ctx,Xi);
1079                         else {
1080                                 ctx->mres = n;
1081                                 return 0;
1082                         }
1083                 }
1084 #if defined(STRICT_ALIGNMENT)
1085                 if (((size_t)in|(size_t)out)%sizeof(size_t) != 0)
1086                         break;
1087 #endif
1088 #if defined(GHASH) && defined(GHASH_CHUNK)
1089                 while (len>=GHASH_CHUNK) {
1090                     size_t j=GHASH_CHUNK;
1091
1092                     GHASH(ctx,in,GHASH_CHUNK);
1093                     while (j) {
1094                         (*block)(ctx->Yi.c,ctx->EKi.c,key);
1095                         ++ctr;
1096                         if (is_endian.little)
1097                                 PUTU32(ctx->Yi.c+12,ctr);
1098                         else
1099                                 ctx->Yi.d[3] = ctr;
1100                         for (i=0; i<16; i+=sizeof(size_t))
1101                                 *(size_t *)(out+i) =
1102                                 *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
1103                         out += 16;
1104                         in  += 16;
1105                         j   -= 16;
1106                     }
1107                     len -= GHASH_CHUNK;
1108                 }
1109                 if ((i = (len&(size_t)-16))) {
1110                     GHASH(ctx,in,i);
1111                     while (len>=16) {
1112                         (*block)(ctx->Yi.c,ctx->EKi.c,key);
1113                         ++ctr;
1114                         if (is_endian.little)
1115                                 PUTU32(ctx->Yi.c+12,ctr);
1116                         else
1117                                 ctx->Yi.d[3] = ctr;
1118                         for (i=0; i<16; i+=sizeof(size_t))
1119                                 *(size_t *)(out+i) =
1120                                 *(size_t *)(in+i)^*(size_t *)(ctx->EKi.c+i);
1121                         out += 16;
1122                         in  += 16;
1123                         len -= 16;
1124                     }
1125                 }
1126 #else
1127                 while (len>=16) {
1128                         (*block)(ctx->Yi.c,ctx->EKi.c,key);
1129                         ++ctr;
1130                         if (is_endian.little)
1131                                 PUTU32(ctx->Yi.c+12,ctr);
1132                         else
1133                                 ctx->Yi.d[3] = ctr;
1134                         for (i=0; i<16; i+=sizeof(size_t)) {
1135                                 size_t c = *(size_t *)(in+i);
1136                                 *(size_t *)(out+i) = c^*(size_t *)(ctx->EKi.c+i);
1137                                 *(size_t *)(ctx->Xi.c+i) ^= c;
1138                         }
1139                         GCM_MUL(ctx,Xi);
1140                         out += 16;
1141                         in  += 16;
1142                         len -= 16;
1143                 }
1144 #endif
1145                 if (len) {
1146                         (*block)(ctx->Yi.c,ctx->EKi.c,key);
1147                         ++ctr;
1148                         if (is_endian.little)
1149                                 PUTU32(ctx->Yi.c+12,ctr);
1150                         else
1151                                 ctx->Yi.d[3] = ctr;
1152                         while (len--) {
1153                                 u8 c = in[n];
1154                                 ctx->Xi.c[n] ^= c;
1155                                 out[n] = c^ctx->EKi.c[n];
1156                                 ++n;
1157                         }
1158                 }
1159
1160                 ctx->mres = n;
1161                 return 0;
1162         } while(0);
1163 #endif
1164         for (i=0;i<len;++i) {
1165                 u8 c;
1166                 if (n==0) {
1167                         (*block)(ctx->Yi.c,ctx->EKi.c,key);
1168                         ++ctr;
1169                         if (is_endian.little)
1170                                 PUTU32(ctx->Yi.c+12,ctr);
1171                         else
1172                                 ctx->Yi.d[3] = ctr;
1173                 }
1174                 c = in[i];
1175                 out[i] = c^ctx->EKi.c[n];
1176                 ctx->Xi.c[n] ^= c;
1177                 n = (n+1)%16;
1178                 if (n==0)
1179                         GCM_MUL(ctx,Xi);
1180         }
1181
1182         ctx->mres = n;
1183         return 0;
1184 }
1185
1186 int CRYPTO_gcm128_encrypt_ctr32(GCM128_CONTEXT *ctx,
1187                 const unsigned char *in, unsigned char *out,
1188                 size_t len, ctr128_f stream)
1189 {
1190         const union { long one; char little; } is_endian = {1};
1191         unsigned int n, ctr;
1192         size_t i;
1193         u64   mlen = ctx->len.u[1];
1194         void *key  = ctx->key;
1195 #ifdef GCM_FUNCREF_4BIT
1196         void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16])    = ctx->gmult;
1197 # ifdef GHASH
1198         void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16],
1199                                 const u8 *inp,size_t len)       = ctx->ghash;
1200 # endif
1201 #endif
1202
1203         mlen += len;
1204         if (mlen>((U64(1)<<36)-32) || (sizeof(len)==8 && mlen<len))
1205                 return -1;
1206         ctx->len.u[1] = mlen;
1207
1208         if (ctx->ares) {
1209                 /* First call to encrypt finalizes GHASH(AAD) */
1210                 GCM_MUL(ctx,Xi);
1211                 ctx->ares = 0;
1212         }
1213
1214         if (is_endian.little)
1215                 ctr = GETU32(ctx->Yi.c+12);
1216         else
1217                 ctr = ctx->Yi.d[3];
1218
1219         n = ctx->mres;
1220         if (n) {
1221                 while (n && len) {
1222                         ctx->Xi.c[n] ^= *(out++) = *(in++)^ctx->EKi.c[n];
1223                         --len;
1224                         n = (n+1)%16;
1225                 }
1226                 if (n==0) GCM_MUL(ctx,Xi);
1227                 else {
1228                         ctx->mres = n;
1229                         return 0;
1230                 }
1231         }
1232 #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
1233         while (len>=GHASH_CHUNK) {
1234                 (*stream)(in,out,GHASH_CHUNK/16,key,ctx->Yi.c);
1235                 ctr += GHASH_CHUNK/16;
1236                 if (is_endian.little)
1237                         PUTU32(ctx->Yi.c+12,ctr);
1238                 else
1239                         ctx->Yi.d[3] = ctr;
1240                 GHASH(ctx,out,GHASH_CHUNK);
1241                 out += GHASH_CHUNK;
1242                 in  += GHASH_CHUNK;
1243                 len -= GHASH_CHUNK;
1244         }
1245 #endif
1246         if ((i = (len&(size_t)-16))) {
1247                 size_t j=i/16;
1248
1249                 (*stream)(in,out,j,key,ctx->Yi.c);
1250                 ctr += (unsigned int)j;
1251                 if (is_endian.little)
1252                         PUTU32(ctx->Yi.c+12,ctr);
1253                 else
1254                         ctx->Yi.d[3] = ctr;
1255                 in  += i;
1256                 len -= i;
1257 #if defined(GHASH)
1258                 GHASH(ctx,out,i);
1259                 out += i;
1260 #else
1261                 while (j--) {
1262                         for (i=0;i<16;++i) ctx->Xi.c[i] ^= out[i];
1263                         GCM_MUL(ctx,Xi);
1264                         out += 16;
1265                 }
1266 #endif
1267         }
1268         if (len) {
1269                 (*ctx->block)(ctx->Yi.c,ctx->EKi.c,key);
1270                 ++ctr;
1271                 if (is_endian.little)
1272                         PUTU32(ctx->Yi.c+12,ctr);
1273                 else
1274                         ctx->Yi.d[3] = ctr;
1275                 while (len--) {
1276                         ctx->Xi.c[n] ^= out[n] = in[n]^ctx->EKi.c[n];
1277                         ++n;
1278                 }
1279         }
1280
1281         ctx->mres = n;
1282         return 0;
1283 }
1284
1285 int CRYPTO_gcm128_decrypt_ctr32(GCM128_CONTEXT *ctx,
1286                 const unsigned char *in, unsigned char *out,
1287                 size_t len,ctr128_f stream)
1288 {
1289         const union { long one; char little; } is_endian = {1};
1290         unsigned int n, ctr;
1291         size_t i;
1292         u64   mlen = ctx->len.u[1];
1293         void *key  = ctx->key;
1294 #ifdef GCM_FUNCREF_4BIT
1295         void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16])    = ctx->gmult;
1296 # ifdef GHASH
1297         void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16],
1298                                 const u8 *inp,size_t len)       = ctx->ghash;
1299 # endif
1300 #endif
1301
1302         mlen += len;
1303         if (mlen>((U64(1)<<36)-32) || (sizeof(len)==8 && mlen<len))
1304                 return -1;
1305         ctx->len.u[1] = mlen;
1306
1307         if (ctx->ares) {
1308                 /* First call to decrypt finalizes GHASH(AAD) */
1309                 GCM_MUL(ctx,Xi);
1310                 ctx->ares = 0;
1311         }
1312
1313         if (is_endian.little)
1314                 ctr = GETU32(ctx->Yi.c+12);
1315         else
1316                 ctr = ctx->Yi.d[3];
1317
1318         n = ctx->mres;
1319         if (n) {
1320                 while (n && len) {
1321                         u8 c = *(in++);
1322                         *(out++) = c^ctx->EKi.c[n];
1323                         ctx->Xi.c[n] ^= c;
1324                         --len;
1325                         n = (n+1)%16;
1326                 }
1327                 if (n==0) GCM_MUL (ctx,Xi);
1328                 else {
1329                         ctx->mres = n;
1330                         return 0;
1331                 }
1332         }
1333 #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
1334         while (len>=GHASH_CHUNK) {
1335                 GHASH(ctx,in,GHASH_CHUNK);
1336                 (*stream)(in,out,GHASH_CHUNK/16,key,ctx->Yi.c);
1337                 ctr += GHASH_CHUNK/16;
1338                 if (is_endian.little)
1339                         PUTU32(ctx->Yi.c+12,ctr);
1340                 else
1341                         ctx->Yi.d[3] = ctr;
1342                 out += GHASH_CHUNK;
1343                 in  += GHASH_CHUNK;
1344                 len -= GHASH_CHUNK;
1345         }
1346 #endif
1347         if ((i = (len&(size_t)-16))) {
1348                 size_t j=i/16;
1349
1350 #if defined(GHASH)
1351                 GHASH(ctx,in,i);
1352 #else
1353                 while (j--) {
1354                         size_t k;
1355                         for (k=0;k<16;++k) ctx->Xi.c[k] ^= in[k];
1356                         GCM_MUL(ctx,Xi);
1357                         in += 16;
1358                 }
1359                 j   = i/16;
1360                 in -= i;
1361 #endif
1362                 (*stream)(in,out,j,key,ctx->Yi.c);
1363                 ctr += (unsigned int)j;
1364                 if (is_endian.little)
1365                         PUTU32(ctx->Yi.c+12,ctr);
1366                 else
1367                         ctx->Yi.d[3] = ctr;
1368                 out += i;
1369                 in  += i;
1370                 len -= i;
1371         }
1372         if (len) {
1373                 (*ctx->block)(ctx->Yi.c,ctx->EKi.c,key);
1374                 ++ctr;
1375                 if (is_endian.little)
1376                         PUTU32(ctx->Yi.c+12,ctr);
1377                 else
1378                         ctx->Yi.d[3] = ctr;
1379                 while (len--) {
1380                         u8 c = in[n];
1381                         ctx->Xi.c[n] ^= c;
1382                         out[n] = c^ctx->EKi.c[n];
1383                         ++n;
1384                 }
1385         }
1386
1387         ctx->mres = n;
1388         return 0;
1389 }
1390
1391 int CRYPTO_gcm128_finish(GCM128_CONTEXT *ctx,const unsigned char *tag,
1392                         size_t len)
1393 {
1394         const union { long one; char little; } is_endian = {1};
1395         u64 alen = ctx->len.u[0]<<3;
1396         u64 clen = ctx->len.u[1]<<3;
1397 #ifdef GCM_FUNCREF_4BIT
1398         void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16])    = ctx->gmult;
1399 #endif
1400
1401         if (ctx->mres)
1402                 GCM_MUL(ctx,Xi);
1403
1404         if (is_endian.little) {
1405 #ifdef BSWAP8
1406                 alen = BSWAP8(alen);
1407                 clen = BSWAP8(clen);
1408 #else
1409                 u8 *p = ctx->len.c;
1410
1411                 ctx->len.u[0] = alen;
1412                 ctx->len.u[1] = clen;
1413
1414                 alen = (u64)GETU32(p)  <<32|GETU32(p+4);
1415                 clen = (u64)GETU32(p+8)<<32|GETU32(p+12);
1416 #endif
1417         }
1418
1419         ctx->Xi.u[0] ^= alen;
1420         ctx->Xi.u[1] ^= clen;
1421         GCM_MUL(ctx,Xi);
1422
1423         ctx->Xi.u[0] ^= ctx->EK0.u[0];
1424         ctx->Xi.u[1] ^= ctx->EK0.u[1];
1425
1426         if (tag && len<=sizeof(ctx->Xi))
1427                 return memcmp(ctx->Xi.c,tag,len);
1428         else
1429                 return -1;
1430 }
1431
1432 void CRYPTO_gcm128_tag(GCM128_CONTEXT *ctx, unsigned char *tag, size_t len)
1433 {
1434         CRYPTO_gcm128_finish(ctx, NULL, 0);
1435         memcpy(tag, ctx->Xi.c, len<=sizeof(ctx->Xi.c)?len:sizeof(ctx->Xi.c));
1436 }
1437
1438 GCM128_CONTEXT *CRYPTO_gcm128_new(void *key, block128_f block)
1439 {
1440         GCM128_CONTEXT *ret;
1441
1442         if ((ret = (GCM128_CONTEXT *)OPENSSL_malloc(sizeof(GCM128_CONTEXT))))
1443                 CRYPTO_gcm128_init(ret,key,block);
1444
1445         return ret;
1446 }
1447
1448 void CRYPTO_gcm128_release(GCM128_CONTEXT *ctx)
1449 {
1450         if (ctx) {
1451                 OPENSSL_cleanse(ctx,sizeof(*ctx));
1452                 OPENSSL_free(ctx);
1453         }
1454 }
1455
1456 #if defined(SELFTEST)
1457 #include <stdio.h>
1458 #include <openssl/aes.h>
1459
1460 /* Test Case 1 */
1461 static const u8 K1[16],
1462                 *P1=NULL,
1463                 *A1=NULL,
1464                 IV1[12],
1465                 *C1=NULL,
1466                 T1[]=  {0x58,0xe2,0xfc,0xce,0xfa,0x7e,0x30,0x61,0x36,0x7f,0x1d,0x57,0xa4,0xe7,0x45,0x5a};
1467
1468 /* Test Case 2 */
1469 #define K2 K1
1470 #define A2 A1
1471 #define IV2 IV1
1472 static const u8 P2[16],
1473                 C2[]=  {0x03,0x88,0xda,0xce,0x60,0xb6,0xa3,0x92,0xf3,0x28,0xc2,0xb9,0x71,0xb2,0xfe,0x78},
1474                 T2[]=  {0xab,0x6e,0x47,0xd4,0x2c,0xec,0x13,0xbd,0xf5,0x3a,0x67,0xb2,0x12,0x57,0xbd,0xdf};
1475
1476 /* Test Case 3 */
1477 #define A3 A2
1478 static const u8 K3[]=  {0xfe,0xff,0xe9,0x92,0x86,0x65,0x73,0x1c,0x6d,0x6a,0x8f,0x94,0x67,0x30,0x83,0x08},
1479                 P3[]=  {0xd9,0x31,0x32,0x25,0xf8,0x84,0x06,0xe5,0xa5,0x59,0x09,0xc5,0xaf,0xf5,0x26,0x9a,
1480                         0x86,0xa7,0xa9,0x53,0x15,0x34,0xf7,0xda,0x2e,0x4c,0x30,0x3d,0x8a,0x31,0x8a,0x72,
1481                         0x1c,0x3c,0x0c,0x95,0x95,0x68,0x09,0x53,0x2f,0xcf,0x0e,0x24,0x49,0xa6,0xb5,0x25,
1482                         0xb1,0x6a,0xed,0xf5,0xaa,0x0d,0xe6,0x57,0xba,0x63,0x7b,0x39,0x1a,0xaf,0xd2,0x55},
1483                 IV3[]= {0xca,0xfe,0xba,0xbe,0xfa,0xce,0xdb,0xad,0xde,0xca,0xf8,0x88},
1484                 C3[]=  {0x42,0x83,0x1e,0xc2,0x21,0x77,0x74,0x24,0x4b,0x72,0x21,0xb7,0x84,0xd0,0xd4,0x9c,
1485                         0xe3,0xaa,0x21,0x2f,0x2c,0x02,0xa4,0xe0,0x35,0xc1,0x7e,0x23,0x29,0xac,0xa1,0x2e,
1486                         0x21,0xd5,0x14,0xb2,0x54,0x66,0x93,0x1c,0x7d,0x8f,0x6a,0x5a,0xac,0x84,0xaa,0x05,
1487                         0x1b,0xa3,0x0b,0x39,0x6a,0x0a,0xac,0x97,0x3d,0x58,0xe0,0x91,0x47,0x3f,0x59,0x85},
1488                 T3[]=  {0x4d,0x5c,0x2a,0xf3,0x27,0xcd,0x64,0xa6,0x2c,0xf3,0x5a,0xbd,0x2b,0xa6,0xfa,0xb4};
1489
1490 /* Test Case 4 */
1491 #define K4 K3
1492 #define IV4 IV3
1493 static const u8 P4[]=  {0xd9,0x31,0x32,0x25,0xf8,0x84,0x06,0xe5,0xa5,0x59,0x09,0xc5,0xaf,0xf5,0x26,0x9a,
1494                         0x86,0xa7,0xa9,0x53,0x15,0x34,0xf7,0xda,0x2e,0x4c,0x30,0x3d,0x8a,0x31,0x8a,0x72,
1495                         0x1c,0x3c,0x0c,0x95,0x95,0x68,0x09,0x53,0x2f,0xcf,0x0e,0x24,0x49,0xa6,0xb5,0x25,
1496                         0xb1,0x6a,0xed,0xf5,0xaa,0x0d,0xe6,0x57,0xba,0x63,0x7b,0x39},
1497                 A4[]=  {0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,
1498                         0xab,0xad,0xda,0xd2},
1499                 C4[]=  {0x42,0x83,0x1e,0xc2,0x21,0x77,0x74,0x24,0x4b,0x72,0x21,0xb7,0x84,0xd0,0xd4,0x9c,
1500                         0xe3,0xaa,0x21,0x2f,0x2c,0x02,0xa4,0xe0,0x35,0xc1,0x7e,0x23,0x29,0xac,0xa1,0x2e,
1501                         0x21,0xd5,0x14,0xb2,0x54,0x66,0x93,0x1c,0x7d,0x8f,0x6a,0x5a,0xac,0x84,0xaa,0x05,
1502                         0x1b,0xa3,0x0b,0x39,0x6a,0x0a,0xac,0x97,0x3d,0x58,0xe0,0x91},
1503                 T4[]=  {0x5b,0xc9,0x4f,0xbc,0x32,0x21,0xa5,0xdb,0x94,0xfa,0xe9,0x5a,0xe7,0x12,0x1a,0x47};
1504
1505 /* Test Case 5 */
1506 #define K5 K4
1507 #define P5 P4
1508 #define A5 A4
1509 static const u8 IV5[]= {0xca,0xfe,0xba,0xbe,0xfa,0xce,0xdb,0xad},
1510                 C5[]=  {0x61,0x35,0x3b,0x4c,0x28,0x06,0x93,0x4a,0x77,0x7f,0xf5,0x1f,0xa2,0x2a,0x47,0x55,
1511                         0x69,0x9b,0x2a,0x71,0x4f,0xcd,0xc6,0xf8,0x37,0x66,0xe5,0xf9,0x7b,0x6c,0x74,0x23,
1512                         0x73,0x80,0x69,0x00,0xe4,0x9f,0x24,0xb2,0x2b,0x09,0x75,0x44,0xd4,0x89,0x6b,0x42,
1513                         0x49,0x89,0xb5,0xe1,0xeb,0xac,0x0f,0x07,0xc2,0x3f,0x45,0x98},
1514                 T5[]=  {0x36,0x12,0xd2,0xe7,0x9e,0x3b,0x07,0x85,0x56,0x1b,0xe1,0x4a,0xac,0xa2,0xfc,0xcb};
1515
1516 /* Test Case 6 */
1517 #define K6 K5
1518 #define P6 P5
1519 #define A6 A5
1520 static const u8 IV6[]= {0x93,0x13,0x22,0x5d,0xf8,0x84,0x06,0xe5,0x55,0x90,0x9c,0x5a,0xff,0x52,0x69,0xaa,
1521                         0x6a,0x7a,0x95,0x38,0x53,0x4f,0x7d,0xa1,0xe4,0xc3,0x03,0xd2,0xa3,0x18,0xa7,0x28,
1522                         0xc3,0xc0,0xc9,0x51,0x56,0x80,0x95,0x39,0xfc,0xf0,0xe2,0x42,0x9a,0x6b,0x52,0x54,
1523                         0x16,0xae,0xdb,0xf5,0xa0,0xde,0x6a,0x57,0xa6,0x37,0xb3,0x9b},
1524                 C6[]=  {0x8c,0xe2,0x49,0x98,0x62,0x56,0x15,0xb6,0x03,0xa0,0x33,0xac,0xa1,0x3f,0xb8,0x94,
1525                         0xbe,0x91,0x12,0xa5,0xc3,0xa2,0x11,0xa8,0xba,0x26,0x2a,0x3c,0xca,0x7e,0x2c,0xa7,
1526                         0x01,0xe4,0xa9,0xa4,0xfb,0xa4,0x3c,0x90,0xcc,0xdc,0xb2,0x81,0xd4,0x8c,0x7c,0x6f,
1527                         0xd6,0x28,0x75,0xd2,0xac,0xa4,0x17,0x03,0x4c,0x34,0xae,0xe5},
1528                 T6[]=  {0x61,0x9c,0xc5,0xae,0xff,0xfe,0x0b,0xfa,0x46,0x2a,0xf4,0x3c,0x16,0x99,0xd0,0x50};
1529
1530 /* Test Case 7 */
1531 static const u8 K7[24],
1532                 *P7=NULL,
1533                 *A7=NULL,
1534                 IV7[12],
1535                 *C7=NULL,
1536                 T7[]=  {0xcd,0x33,0xb2,0x8a,0xc7,0x73,0xf7,0x4b,0xa0,0x0e,0xd1,0xf3,0x12,0x57,0x24,0x35};
1537
1538 /* Test Case 8 */
1539 #define K8 K7
1540 #define IV8 IV7
1541 #define A8 A7
1542 static const u8 P8[16],
1543                 C8[]=  {0x98,0xe7,0x24,0x7c,0x07,0xf0,0xfe,0x41,0x1c,0x26,0x7e,0x43,0x84,0xb0,0xf6,0x00},
1544                 T8[]=  {0x2f,0xf5,0x8d,0x80,0x03,0x39,0x27,0xab,0x8e,0xf4,0xd4,0x58,0x75,0x14,0xf0,0xfb};
1545
1546 /* Test Case 9 */
1547 #define A9 A8
1548 static const u8 K9[]=  {0xfe,0xff,0xe9,0x92,0x86,0x65,0x73,0x1c,0x6d,0x6a,0x8f,0x94,0x67,0x30,0x83,0x08,
1549                         0xfe,0xff,0xe9,0x92,0x86,0x65,0x73,0x1c},
1550                 P9[]=  {0xd9,0x31,0x32,0x25,0xf8,0x84,0x06,0xe5,0xa5,0x59,0x09,0xc5,0xaf,0xf5,0x26,0x9a,
1551                         0x86,0xa7,0xa9,0x53,0x15,0x34,0xf7,0xda,0x2e,0x4c,0x30,0x3d,0x8a,0x31,0x8a,0x72,
1552                         0x1c,0x3c,0x0c,0x95,0x95,0x68,0x09,0x53,0x2f,0xcf,0x0e,0x24,0x49,0xa6,0xb5,0x25,
1553                         0xb1,0x6a,0xed,0xf5,0xaa,0x0d,0xe6,0x57,0xba,0x63,0x7b,0x39,0x1a,0xaf,0xd2,0x55},
1554                 IV9[]= {0xca,0xfe,0xba,0xbe,0xfa,0xce,0xdb,0xad,0xde,0xca,0xf8,0x88},
1555                 C9[]=  {0x39,0x80,0xca,0x0b,0x3c,0x00,0xe8,0x41,0xeb,0x06,0xfa,0xc4,0x87,0x2a,0x27,0x57,
1556                         0x85,0x9e,0x1c,0xea,0xa6,0xef,0xd9,0x84,0x62,0x85,0x93,0xb4,0x0c,0xa1,0xe1,0x9c,
1557                         0x7d,0x77,0x3d,0x00,0xc1,0x44,0xc5,0x25,0xac,0x61,0x9d,0x18,0xc8,0x4a,0x3f,0x47,
1558                         0x18,0xe2,0x44,0x8b,0x2f,0xe3,0x24,0xd9,0xcc,0xda,0x27,0x10,0xac,0xad,0xe2,0x56},
1559                 T9[]=  {0x99,0x24,0xa7,0xc8,0x58,0x73,0x36,0xbf,0xb1,0x18,0x02,0x4d,0xb8,0x67,0x4a,0x14};
1560
1561 /* Test Case 10 */
1562 #define K10 K9
1563 #define IV10 IV9
1564 static const u8 P10[]= {0xd9,0x31,0x32,0x25,0xf8,0x84,0x06,0xe5,0xa5,0x59,0x09,0xc5,0xaf,0xf5,0x26,0x9a,
1565                         0x86,0xa7,0xa9,0x53,0x15,0x34,0xf7,0xda,0x2e,0x4c,0x30,0x3d,0x8a,0x31,0x8a,0x72,
1566                         0x1c,0x3c,0x0c,0x95,0x95,0x68,0x09,0x53,0x2f,0xcf,0x0e,0x24,0x49,0xa6,0xb5,0x25,
1567                         0xb1,0x6a,0xed,0xf5,0xaa,0x0d,0xe6,0x57,0xba,0x63,0x7b,0x39},
1568                 A10[]= {0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,
1569                         0xab,0xad,0xda,0xd2},
1570                 C10[]= {0x39,0x80,0xca,0x0b,0x3c,0x00,0xe8,0x41,0xeb,0x06,0xfa,0xc4,0x87,0x2a,0x27,0x57,
1571                         0x85,0x9e,0x1c,0xea,0xa6,0xef,0xd9,0x84,0x62,0x85,0x93,0xb4,0x0c,0xa1,0xe1,0x9c,
1572                         0x7d,0x77,0x3d,0x00,0xc1,0x44,0xc5,0x25,0xac,0x61,0x9d,0x18,0xc8,0x4a,0x3f,0x47,
1573                         0x18,0xe2,0x44,0x8b,0x2f,0xe3,0x24,0xd9,0xcc,0xda,0x27,0x10},
1574                 T10[]= {0x25,0x19,0x49,0x8e,0x80,0xf1,0x47,0x8f,0x37,0xba,0x55,0xbd,0x6d,0x27,0x61,0x8c};
1575
1576 /* Test Case 11 */
1577 #define K11 K10
1578 #define P11 P10
1579 #define A11 A10
1580 static const u8 IV11[]={0xca,0xfe,0xba,0xbe,0xfa,0xce,0xdb,0xad},
1581                 C11[]= {0x0f,0x10,0xf5,0x99,0xae,0x14,0xa1,0x54,0xed,0x24,0xb3,0x6e,0x25,0x32,0x4d,0xb8,
1582                         0xc5,0x66,0x63,0x2e,0xf2,0xbb,0xb3,0x4f,0x83,0x47,0x28,0x0f,0xc4,0x50,0x70,0x57,
1583                         0xfd,0xdc,0x29,0xdf,0x9a,0x47,0x1f,0x75,0xc6,0x65,0x41,0xd4,0xd4,0xda,0xd1,0xc9,
1584                         0xe9,0x3a,0x19,0xa5,0x8e,0x8b,0x47,0x3f,0xa0,0xf0,0x62,0xf7},
1585                 T11[]= {0x65,0xdc,0xc5,0x7f,0xcf,0x62,0x3a,0x24,0x09,0x4f,0xcc,0xa4,0x0d,0x35,0x33,0xf8};
1586
1587 /* Test Case 12 */
1588 #define K12 K11
1589 #define P12 P11
1590 #define A12 A11
1591 static const u8 IV12[]={0x93,0x13,0x22,0x5d,0xf8,0x84,0x06,0xe5,0x55,0x90,0x9c,0x5a,0xff,0x52,0x69,0xaa,
1592                         0x6a,0x7a,0x95,0x38,0x53,0x4f,0x7d,0xa1,0xe4,0xc3,0x03,0xd2,0xa3,0x18,0xa7,0x28,
1593                         0xc3,0xc0,0xc9,0x51,0x56,0x80,0x95,0x39,0xfc,0xf0,0xe2,0x42,0x9a,0x6b,0x52,0x54,
1594                         0x16,0xae,0xdb,0xf5,0xa0,0xde,0x6a,0x57,0xa6,0x37,0xb3,0x9b},
1595                 C12[]= {0xd2,0x7e,0x88,0x68,0x1c,0xe3,0x24,0x3c,0x48,0x30,0x16,0x5a,0x8f,0xdc,0xf9,0xff,
1596                         0x1d,0xe9,0xa1,0xd8,0xe6,0xb4,0x47,0xef,0x6e,0xf7,0xb7,0x98,0x28,0x66,0x6e,0x45,
1597                         0x81,0xe7,0x90,0x12,0xaf,0x34,0xdd,0xd9,0xe2,0xf0,0x37,0x58,0x9b,0x29,0x2d,0xb3,
1598                         0xe6,0x7c,0x03,0x67,0x45,0xfa,0x22,0xe7,0xe9,0xb7,0x37,0x3b},
1599                 T12[]= {0xdc,0xf5,0x66,0xff,0x29,0x1c,0x25,0xbb,0xb8,0x56,0x8f,0xc3,0xd3,0x76,0xa6,0xd9};
1600
1601 /* Test Case 13 */
1602 static const u8 K13[32],
1603                 *P13=NULL,
1604                 *A13=NULL,
1605                 IV13[12],
1606                 *C13=NULL,
1607                 T13[]={0x53,0x0f,0x8a,0xfb,0xc7,0x45,0x36,0xb9,0xa9,0x63,0xb4,0xf1,0xc4,0xcb,0x73,0x8b};
1608
1609 /* Test Case 14 */
1610 #define K14 K13
1611 #define A14 A13
1612 static const u8 P14[16],
1613                 IV14[12],
1614                 C14[]= {0xce,0xa7,0x40,0x3d,0x4d,0x60,0x6b,0x6e,0x07,0x4e,0xc5,0xd3,0xba,0xf3,0x9d,0x18},
1615                 T14[]= {0xd0,0xd1,0xc8,0xa7,0x99,0x99,0x6b,0xf0,0x26,0x5b,0x98,0xb5,0xd4,0x8a,0xb9,0x19};
1616
1617 /* Test Case 15 */
1618 #define A15 A14
1619 static const u8 K15[]= {0xfe,0xff,0xe9,0x92,0x86,0x65,0x73,0x1c,0x6d,0x6a,0x8f,0x94,0x67,0x30,0x83,0x08,
1620                         0xfe,0xff,0xe9,0x92,0x86,0x65,0x73,0x1c,0x6d,0x6a,0x8f,0x94,0x67,0x30,0x83,0x08},
1621                 P15[]= {0xd9,0x31,0x32,0x25,0xf8,0x84,0x06,0xe5,0xa5,0x59,0x09,0xc5,0xaf,0xf5,0x26,0x9a,
1622                         0x86,0xa7,0xa9,0x53,0x15,0x34,0xf7,0xda,0x2e,0x4c,0x30,0x3d,0x8a,0x31,0x8a,0x72,
1623                         0x1c,0x3c,0x0c,0x95,0x95,0x68,0x09,0x53,0x2f,0xcf,0x0e,0x24,0x49,0xa6,0xb5,0x25,
1624                         0xb1,0x6a,0xed,0xf5,0xaa,0x0d,0xe6,0x57,0xba,0x63,0x7b,0x39,0x1a,0xaf,0xd2,0x55},
1625                 IV15[]={0xca,0xfe,0xba,0xbe,0xfa,0xce,0xdb,0xad,0xde,0xca,0xf8,0x88},
1626                 C15[]= {0x52,0x2d,0xc1,0xf0,0x99,0x56,0x7d,0x07,0xf4,0x7f,0x37,0xa3,0x2a,0x84,0x42,0x7d,
1627                         0x64,0x3a,0x8c,0xdc,0xbf,0xe5,0xc0,0xc9,0x75,0x98,0xa2,0xbd,0x25,0x55,0xd1,0xaa,
1628                         0x8c,0xb0,0x8e,0x48,0x59,0x0d,0xbb,0x3d,0xa7,0xb0,0x8b,0x10,0x56,0x82,0x88,0x38,
1629                         0xc5,0xf6,0x1e,0x63,0x93,0xba,0x7a,0x0a,0xbc,0xc9,0xf6,0x62,0x89,0x80,0x15,0xad},
1630                 T15[]= {0xb0,0x94,0xda,0xc5,0xd9,0x34,0x71,0xbd,0xec,0x1a,0x50,0x22,0x70,0xe3,0xcc,0x6c};
1631
1632 /* Test Case 16 */
1633 #define K16 K15
1634 #define IV16 IV15
1635 static const u8 P16[]= {0xd9,0x31,0x32,0x25,0xf8,0x84,0x06,0xe5,0xa5,0x59,0x09,0xc5,0xaf,0xf5,0x26,0x9a,
1636                         0x86,0xa7,0xa9,0x53,0x15,0x34,0xf7,0xda,0x2e,0x4c,0x30,0x3d,0x8a,0x31,0x8a,0x72,
1637                         0x1c,0x3c,0x0c,0x95,0x95,0x68,0x09,0x53,0x2f,0xcf,0x0e,0x24,0x49,0xa6,0xb5,0x25,
1638                         0xb1,0x6a,0xed,0xf5,0xaa,0x0d,0xe6,0x57,0xba,0x63,0x7b,0x39},
1639                 A16[]= {0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,0xfe,0xed,0xfa,0xce,0xde,0xad,0xbe,0xef,
1640                         0xab,0xad,0xda,0xd2},
1641                 C16[]= {0x52,0x2d,0xc1,0xf0,0x99,0x56,0x7d,0x07,0xf4,0x7f,0x37,0xa3,0x2a,0x84,0x42,0x7d,
1642                         0x64,0x3a,0x8c,0xdc,0xbf,0xe5,0xc0,0xc9,0x75,0x98,0xa2,0xbd,0x25,0x55,0xd1,0xaa,
1643                         0x8c,0xb0,0x8e,0x48,0x59,0x0d,0xbb,0x3d,0xa7,0xb0,0x8b,0x10,0x56,0x82,0x88,0x38,
1644                         0xc5,0xf6,0x1e,0x63,0x93,0xba,0x7a,0x0a,0xbc,0xc9,0xf6,0x62},
1645                 T16[]= {0x76,0xfc,0x6e,0xce,0x0f,0x4e,0x17,0x68,0xcd,0xdf,0x88,0x53,0xbb,0x2d,0x55,0x1b};
1646
1647 /* Test Case 17 */
1648 #define K17 K16
1649 #define P17 P16
1650 #define A17 A16
1651 static const u8 IV17[]={0xca,0xfe,0xba,0xbe,0xfa,0xce,0xdb,0xad},
1652                 C17[]= {0xc3,0x76,0x2d,0xf1,0xca,0x78,0x7d,0x32,0xae,0x47,0xc1,0x3b,0xf1,0x98,0x44,0xcb,
1653                         0xaf,0x1a,0xe1,0x4d,0x0b,0x97,0x6a,0xfa,0xc5,0x2f,0xf7,0xd7,0x9b,0xba,0x9d,0xe0,
1654                         0xfe,0xb5,0x82,0xd3,0x39,0x34,0xa4,0xf0,0x95,0x4c,0xc2,0x36,0x3b,0xc7,0x3f,0x78,
1655                         0x62,0xac,0x43,0x0e,0x64,0xab,0xe4,0x99,0xf4,0x7c,0x9b,0x1f},
1656                 T17[]= {0x3a,0x33,0x7d,0xbf,0x46,0xa7,0x92,0xc4,0x5e,0x45,0x49,0x13,0xfe,0x2e,0xa8,0xf2};
1657
1658 /* Test Case 18 */
1659 #define K18 K17
1660 #define P18 P17
1661 #define A18 A17
1662 static const u8 IV18[]={0x93,0x13,0x22,0x5d,0xf8,0x84,0x06,0xe5,0x55,0x90,0x9c,0x5a,0xff,0x52,0x69,0xaa,
1663                         0x6a,0x7a,0x95,0x38,0x53,0x4f,0x7d,0xa1,0xe4,0xc3,0x03,0xd2,0xa3,0x18,0xa7,0x28,
1664                         0xc3,0xc0,0xc9,0x51,0x56,0x80,0x95,0x39,0xfc,0xf0,0xe2,0x42,0x9a,0x6b,0x52,0x54,
1665                         0x16,0xae,0xdb,0xf5,0xa0,0xde,0x6a,0x57,0xa6,0x37,0xb3,0x9b},
1666                 C18[]= {0x5a,0x8d,0xef,0x2f,0x0c,0x9e,0x53,0xf1,0xf7,0x5d,0x78,0x53,0x65,0x9e,0x2a,0x20,
1667                         0xee,0xb2,0xb2,0x2a,0xaf,0xde,0x64,0x19,0xa0,0x58,0xab,0x4f,0x6f,0x74,0x6b,0xf4,
1668                         0x0f,0xc0,0xc3,0xb7,0x80,0xf2,0x44,0x45,0x2d,0xa3,0xeb,0xf1,0xc5,0xd8,0x2c,0xde,
1669                         0xa2,0x41,0x89,0x97,0x20,0x0e,0xf8,0x2e,0x44,0xae,0x7e,0x3f},
1670                 T18[]= {0xa4,0x4a,0x82,0x66,0xee,0x1c,0x8e,0xb0,0xc8,0xb5,0xd4,0xcf,0x5a,0xe9,0xf1,0x9a};
1671
1672 #define TEST_CASE(n)    do {                                    \
1673         u8 out[sizeof(P##n)];                                   \
1674         AES_set_encrypt_key(K##n,sizeof(K##n)*8,&key);          \
1675         CRYPTO_gcm128_init(&ctx,&key,(block128_f)AES_encrypt);  \
1676         CRYPTO_gcm128_setiv(&ctx,IV##n,sizeof(IV##n));          \
1677         memset(out,0,sizeof(out));                              \
1678         if (A##n) CRYPTO_gcm128_aad(&ctx,A##n,sizeof(A##n));    \
1679         if (P##n) CRYPTO_gcm128_encrypt(&ctx,P##n,out,sizeof(out));     \
1680         if (CRYPTO_gcm128_finish(&ctx,T##n,16) ||               \
1681             (C##n && memcmp(out,C##n,sizeof(out))))             \
1682                 ret++, printf ("encrypt test#%d failed.\n",n);  \
1683         CRYPTO_gcm128_setiv(&ctx,IV##n,sizeof(IV##n));          \
1684         memset(out,0,sizeof(out));                              \
1685         if (A##n) CRYPTO_gcm128_aad(&ctx,A##n,sizeof(A##n));    \
1686         if (C##n) CRYPTO_gcm128_decrypt(&ctx,C##n,out,sizeof(out));     \
1687         if (CRYPTO_gcm128_finish(&ctx,T##n,16) ||               \
1688             (P##n && memcmp(out,P##n,sizeof(out))))             \
1689                 ret++, printf ("decrypt test#%d failed.\n",n);  \
1690         } while(0)
1691
1692 int main()
1693 {
1694         GCM128_CONTEXT ctx;
1695         AES_KEY key;
1696         int ret=0;
1697
1698         TEST_CASE(1);
1699         TEST_CASE(2);
1700         TEST_CASE(3);
1701         TEST_CASE(4);
1702         TEST_CASE(5);
1703         TEST_CASE(6);
1704         TEST_CASE(7);
1705         TEST_CASE(8);
1706         TEST_CASE(9);
1707         TEST_CASE(10);
1708         TEST_CASE(11);
1709         TEST_CASE(12);
1710         TEST_CASE(13);
1711         TEST_CASE(14);
1712         TEST_CASE(15);
1713         TEST_CASE(16);
1714         TEST_CASE(17);
1715         TEST_CASE(18);
1716
1717 #ifdef OPENSSL_CPUID_OBJ
1718         {
1719         size_t start,stop,gcm_t,ctr_t,OPENSSL_rdtsc();
1720         union { u64 u; u8 c[1024]; } buf;
1721         int i;
1722
1723         AES_set_encrypt_key(K1,sizeof(K1)*8,&key);
1724         CRYPTO_gcm128_init(&ctx,&key,(block128_f)AES_encrypt);
1725         CRYPTO_gcm128_setiv(&ctx,IV1,sizeof(IV1));
1726
1727         CRYPTO_gcm128_encrypt(&ctx,buf.c,buf.c,sizeof(buf));
1728         start = OPENSSL_rdtsc();
1729         CRYPTO_gcm128_encrypt(&ctx,buf.c,buf.c,sizeof(buf));
1730         gcm_t = OPENSSL_rdtsc() - start;
1731
1732         CRYPTO_ctr128_encrypt(buf.c,buf.c,sizeof(buf),
1733                         &key,ctx.Yi.c,ctx.EKi.c,&ctx.mres,
1734                         (block128_f)AES_encrypt);
1735         start = OPENSSL_rdtsc();
1736         CRYPTO_ctr128_encrypt(buf.c,buf.c,sizeof(buf),
1737                         &key,ctx.Yi.c,ctx.EKi.c,&ctx.mres,
1738                         (block128_f)AES_encrypt);
1739         ctr_t = OPENSSL_rdtsc() - start;
1740
1741         printf("%.2f-%.2f=%.2f\n",
1742                         gcm_t/(double)sizeof(buf),
1743                         ctr_t/(double)sizeof(buf),
1744                         (gcm_t-ctr_t)/(double)sizeof(buf));
1745 #ifdef GHASH
1746         GHASH(&ctx,buf.c,sizeof(buf));
1747         start = OPENSSL_rdtsc();
1748         for (i=0;i<100;++i) GHASH(&ctx,buf.c,sizeof(buf));
1749         gcm_t = OPENSSL_rdtsc() - start;
1750         printf("%.2f\n",gcm_t/(double)sizeof(buf)/(double)i);
1751 #endif
1752         }
1753 #endif
1754
1755         return ret;
1756 }
1757 #endif