crypto/rc4/rc4_enc.c

   1 /* crypto/rc4/rc4_enc.c */
   2 /* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com)
   3  * All rights reserved.
   4  *
   5  * This package is an SSL implementation written
   6  * by Eric Young (eay@cryptsoft.com).
   7  * The implementation was written so as to conform with Netscapes SSL.
   8  *
   9  * This library is free for commercial and non-commercial use as long as
  10  * the following conditions are aheared to.  The following conditions
  11  * apply to all code found in this distribution, be it the RC4, RSA,
  12  * lhash, DES, etc., code; not just the SSL code.  The SSL documentation
  13  * included with this distribution is covered by the same copyright terms
  14  * except that the holder is Tim Hudson (tjh@cryptsoft.com).
  15  *
  16  * Copyright remains Eric Young's, and as such any Copyright notices in
  17  * the code are not to be removed.
  18  * If this package is used in a product, Eric Young should be given attribution
  19  * as the author of the parts of the library used.
  20  * This can be in the form of a textual message at program startup or
  21  * in documentation (online or textual) provided with the package.
  22  *
  23  * Redistribution and use in source and binary forms, with or without
  24  * modification, are permitted provided that the following conditions
  25  * are met:
  26  * 1. Redistributions of source code must retain the copyright
  27  *    notice, this list of conditions and the following disclaimer.
  28  * 2. Redistributions in binary form must reproduce the above copyright
  29  *    notice, this list of conditions and the following disclaimer in the
  30  *    documentation and/or other materials provided with the distribution.
  31  * 3. All advertising materials mentioning features or use of this software
  32  *    must display the following acknowledgement:
  33  *    "This product includes cryptographic software written by
  34  *     Eric Young (eay@cryptsoft.com)"
  35  *    The word 'cryptographic' can be left out if the rouines from the library
  36  *    being used are not cryptographic related :-).
  37  * 4. If you include any Windows specific code (or a derivative thereof) from
  38  *    the apps directory (application code) you must include an acknowledgement:
  39  *    "This product includes software written by Tim Hudson (tjh@cryptsoft.com)"
  40  *
  41  * THIS SOFTWARE IS PROVIDED BY ERIC YOUNG ``AS IS'' AND
  42  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  43  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  44  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  45  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  46  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  47  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  48  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  49  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  50  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  51  * SUCH DAMAGE.
  52  *
  53  * The licence and distribution terms for any publically available version or
  54  * derivative of this code cannot be changed.  i.e. this code cannot simply be
  55  * copied and put under another distribution licence
  56  * [including the GNU Public Licence.]
  57  */
  58
  59 #include <openssl/rc4.h>
  60 #include "rc4_locl.h"
  61
  62 /*-
  63  * RC4 as implemented from a posting from
  64  * Newsgroups: sci.crypt
  65  * From: sterndark@netcom.com (David Sterndark)
  66  * Subject: RC4 Algorithm revealed.
  67  * Message-ID: <sternCvKL4B.Hyy@netcom.com>
  68  * Date: Wed, 14 Sep 1994 06:35:31 GMT
  69  */
  70
  71 void RC4(RC4_KEY *key, size_t len, const unsigned char *indata,
  72              unsigned char *outdata)
  73         {
  74         register RC4_INT *d;
  75         register RC4_INT x,y,tx,ty;
  76         size_t i;
  77
  78         x=key->x;
  79         y=key->y;
  80         d=key->data;
  81
  82 #if defined(RC4_CHUNK) && !defined(PEDANTIC)
  83         /*-
  84          * The original reason for implementing this(*) was the fact that
  85          * pre-21164a Alpha CPUs don't have byte load/store instructions
  86          * and e.g. a byte store has to be done with 64-bit load, shift,
  87          * and, or and finally 64-bit store. Peaking data and operating
  88          * at natural word size made it possible to reduce amount of
  89          * instructions as well as to perform early read-ahead without
  90          * suffering from RAW (read-after-write) hazard. This resulted
  91          * in ~40%(**) performance improvement on 21064 box with gcc.
  92          * But it's not only Alpha users who win here:-) Thanks to the
  93          * early-n-wide read-ahead this implementation also exhibits
  94          * >40% speed-up on SPARC and 20-30% on 64-bit MIPS (depending
  95          * on sizeof(RC4_INT)).
  96          *
  97          * (*)  "this" means code which recognizes the case when input
  98          *      and output pointers appear to be aligned at natural CPU
  99          *      word boundary
 100          * (**) i.e. according to 'apps/openssl speed rc4' benchmark,
 101          *      crypto/rc4/rc4speed.c exhibits almost 70% speed-up...
 102          *
 103          * Cavets.
 104          *
 105          * - RC4_CHUNK="unsigned long long" should be a #1 choice for
 106          *   UltraSPARC. Unfortunately gcc generates very slow code
 107          *   (2.5-3 times slower than one generated by Sun's WorkShop
 108          *   C) and therefore gcc (at least 2.95 and earlier) should
 109          *   always be told that RC4_CHUNK="unsigned long".
 110          *
 111          *                                      <appro@fy.chalmers.se>
 112          */
 113
 114 # define RC4_STEP       ( \
 115                         x=(x+1) &0xff,  \
 116                         tx=d[x],        \
 117                         y=(tx+y)&0xff,  \
 118                         ty=d[y],        \
 119                         d[y]=tx,        \
 120                         d[x]=ty,        \
 121                         (RC4_CHUNK)d[(tx+ty)&0xff]\
 122                         )
 123
 124         if ( ( ((size_t)indata  & (sizeof(RC4_CHUNK)-1)) |
 125                ((size_t)outdata & (sizeof(RC4_CHUNK)-1)) ) == 0 )
 126                 {
 127                 RC4_CHUNK ichunk,otp;
 128                 const union { long one; char little; } is_endian = {1};
 129
 130                 /*-
 131                  * I reckon we can afford to implement both endian
 132                  * cases and to decide which way to take at run-time
 133                  * because the machine code appears to be very compact
 134                  * and redundant 1-2KB is perfectly tolerable (i.e.
 135                  * in case the compiler fails to eliminate it:-). By
 136                  * suggestion from Terrel Larson <terr@terralogic.net>
 137                  * who also stands for the is_endian union:-)
 138                  *
 139                  * Special notes.
 140                  *
 141                  * - is_endian is declared automatic as doing otherwise
 142                  *   (declaring static) prevents gcc from eliminating
 143                  *   the redundant code;
 144                  * - compilers (those I've tried) don't seem to have
 145                  *   problems eliminating either the operators guarded
 146                  *   by "if (sizeof(RC4_CHUNK)==8)" or the condition
 147                  *   expressions themselves so I've got 'em to replace
 148                  *   corresponding #ifdefs from the previous version;
 149                  * - I chose to let the redundant switch cases when
 150                  *   sizeof(RC4_CHUNK)!=8 be (were also #ifdefed
 151                  *   before);
 152                  * - in case you wonder "&(sizeof(RC4_CHUNK)*8-1)" in
 153                  *   [LB]ESHFT guards against "shift is out of range"
 154                  *   warnings when sizeof(RC4_CHUNK)!=8
 155                  *
 156                  *                      <appro@fy.chalmers.se>
 157                  */
 158                 if (!is_endian.little)
 159                         {       /* BIG-ENDIAN CASE */
 160 # define BESHFT(c)      (((sizeof(RC4_CHUNK)-(c)-1)*8)&(sizeof(RC4_CHUNK)*8-1))
 161                         for (;len&(0-sizeof(RC4_CHUNK));len-=sizeof(RC4_CHUNK))
 162                                 {
 163                                 ichunk  = *(RC4_CHUNK *)indata;
 164                                 otp  = RC4_STEP<<BESHFT(0);
 165                                 otp |= RC4_STEP<<BESHFT(1);
 166                                 otp |= RC4_STEP<<BESHFT(2);
 167                                 otp |= RC4_STEP<<BESHFT(3);
 168                                 if (sizeof(RC4_CHUNK)==8)
 169                                         {
 170                                         otp |= RC4_STEP<<BESHFT(4);
 171                                         otp |= RC4_STEP<<BESHFT(5);
 172                                         otp |= RC4_STEP<<BESHFT(6);
 173                                         otp |= RC4_STEP<<BESHFT(7);
 174                                         }
 175                                 *(RC4_CHUNK *)outdata = otp^ichunk;
 176                                 indata  += sizeof(RC4_CHUNK);
 177                                 outdata += sizeof(RC4_CHUNK);
 178                                 }
 179                         if (len)
 180                                 {
 181                                 RC4_CHUNK mask=(RC4_CHUNK)-1, ochunk;
 182
 183                                 ichunk = *(RC4_CHUNK *)indata;
 184                                 ochunk = *(RC4_CHUNK *)outdata;
 185                                 otp = 0;
 186                                 i = BESHFT(0);
 187                                 mask <<= (sizeof(RC4_CHUNK)-len)<<3;
 188                                 switch (len&(sizeof(RC4_CHUNK)-1))
 189                                         {
 190                                         case 7: otp  = RC4_STEP<<i, i-=8;
 191                                         case 6: otp |= RC4_STEP<<i, i-=8;
 192                                         case 5: otp |= RC4_STEP<<i, i-=8;
 193                                         case 4: otp |= RC4_STEP<<i, i-=8;
 194                                         case 3: otp |= RC4_STEP<<i, i-=8;
 195                                         case 2: otp |= RC4_STEP<<i, i-=8;
 196                                         case 1: otp |= RC4_STEP<<i, i-=8;
 197                                         case 0: ; /*
 198                                                    * it's never the case,
 199                                                    * but it has to be here
 200                                                    * for ultrix?
 201                                                    */
 202                                         }
 203                                 ochunk &= ~mask;
 204                                 ochunk |= (otp^ichunk) & mask;
 205                                 *(RC4_CHUNK *)outdata = ochunk;
 206                                 }
 207                         key->x=x;
 208                         key->y=y;
 209                         return;
 210                         }
 211                 else
 212                         {       /* LITTLE-ENDIAN CASE */
 213 # define LESHFT(c)      (((c)*8)&(sizeof(RC4_CHUNK)*8-1))
 214                         for (;len&(0-sizeof(RC4_CHUNK));len-=sizeof(RC4_CHUNK))
 215                                 {
 216                                 ichunk  = *(RC4_CHUNK *)indata;
 217                                 otp  = RC4_STEP;
 218                                 otp |= RC4_STEP<<8;
 219                                 otp |= RC4_STEP<<16;
 220                                 otp |= RC4_STEP<<24;
 221                                 if (sizeof(RC4_CHUNK)==8)
 222                                         {
 223                                         otp |= RC4_STEP<<LESHFT(4);
 224                                         otp |= RC4_STEP<<LESHFT(5);
 225                                         otp |= RC4_STEP<<LESHFT(6);
 226                                         otp |= RC4_STEP<<LESHFT(7);
 227                                         }
 228                                 *(RC4_CHUNK *)outdata = otp^ichunk;
 229                                 indata  += sizeof(RC4_CHUNK);
 230                                 outdata += sizeof(RC4_CHUNK);
 231                                 }
 232                         if (len)
 233                                 {
 234                                 RC4_CHUNK mask=(RC4_CHUNK)-1, ochunk;
 235
 236                                 ichunk = *(RC4_CHUNK *)indata;
 237                                 ochunk = *(RC4_CHUNK *)outdata;
 238                                 otp = 0;
 239                                 i   = 0;
 240                                 mask >>= (sizeof(RC4_CHUNK)-len)<<3;
 241                                 switch (len&(sizeof(RC4_CHUNK)-1))
 242                                         {
 243                                         case 7: otp  = RC4_STEP,    i+=8;
 244                                         case 6: otp |= RC4_STEP<<i, i+=8;
 245                                         case 5: otp |= RC4_STEP<<i, i+=8;
 246                                         case 4: otp |= RC4_STEP<<i, i+=8;
 247                                         case 3: otp |= RC4_STEP<<i, i+=8;
 248                                         case 2: otp |= RC4_STEP<<i, i+=8;
 249                                         case 1: otp |= RC4_STEP<<i, i+=8;
 250                                         case 0: ; /*
 251                                                    * it's never the case,
 252                                                    * but it has to be here
 253                                                    * for ultrix?
 254                                                    */
 255                                         }
 256                                 ochunk &= ~mask;
 257                                 ochunk |= (otp^ichunk) & mask;
 258                                 *(RC4_CHUNK *)outdata = ochunk;
 259                                 }
 260                         key->x=x;
 261                         key->y=y;
 262                         return;
 263                         }
 264                 }
 265 #endif
 266 #define LOOP(in,out) \
 267                 x=((x+1)&0xff); \
 268                 tx=d[x]; \
 269                 y=(tx+y)&0xff; \
 270                 d[x]=ty=d[y]; \
 271                 d[y]=tx; \
 272                 (out) = d[(tx+ty)&0xff]^ (in);
 273
 274 #ifndef RC4_INDEX
 275 #define RC4_LOOP(a,b,i) LOOP(*((a)++),*((b)++))
 276 #else
 277 #define RC4_LOOP(a,b,i) LOOP(a[i],b[i])
 278 #endif
 279
 280         i=len>>3;
 281         if (i)
 282                 {
 283                 for (;;)
 284                         {
 285                         RC4_LOOP(indata,outdata,0);
 286                         RC4_LOOP(indata,outdata,1);
 287                         RC4_LOOP(indata,outdata,2);
 288                         RC4_LOOP(indata,outdata,3);
 289                         RC4_LOOP(indata,outdata,4);
 290                         RC4_LOOP(indata,outdata,5);
 291                         RC4_LOOP(indata,outdata,6);
 292                         RC4_LOOP(indata,outdata,7);
 293 #ifdef RC4_INDEX
 294                         indata+=8;
 295                         outdata+=8;
 296 #endif
 297                         if (--i == 0) break;
 298                         }
 299                 }
 300         i=len&0x07;
 301         if (i)
 302                 {
 303                 for (;;)
 304                         {
 305                         RC4_LOOP(indata,outdata,0); if (--i == 0) break;
 306                         RC4_LOOP(indata,outdata,1); if (--i == 0) break;
 307                         RC4_LOOP(indata,outdata,2); if (--i == 0) break;
 308                         RC4_LOOP(indata,outdata,3); if (--i == 0) break;
 309                         RC4_LOOP(indata,outdata,4); if (--i == 0) break;
 310                         RC4_LOOP(indata,outdata,5); if (--i == 0) break;
 311                         RC4_LOOP(indata,outdata,6); if (--i == 0) break;
 312                         }
 313                 }
 314         key->x=x;
 315         key->y=y;
 316         }