1 /* $OpenBSD: rc4.c,v 1.13 2025/01/27 14:02:32 jsing Exp $ */ 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 <endian.h> 60 61 #include <openssl/rc4.h> 62 63 #include "crypto_arch.h" 64 65 /* RC4 as implemented from a posting from 66 * Newsgroups: sci.crypt 67 * From: sterndark@netcom.com (David Sterndark) 68 * Subject: RC4 Algorithm revealed. 69 * Message-ID: <sternCvKL4B.Hyy@netcom.com> 70 * Date: Wed, 14 Sep 1994 06:35:31 GMT 71 */ 72 73 #ifdef HAVE_RC4_INTERNAL 74 void rc4_internal(RC4_KEY *key, size_t len, const unsigned char *indata, 75 unsigned char *outdata); 76 77 #else 78 static void 79 rc4_internal(RC4_KEY *key, size_t len, const unsigned char *indata, 80 unsigned char *outdata) 81 { 82 RC4_INT *d; 83 RC4_INT x, y,tx, ty; 84 size_t i; 85 86 x = key->x; 87 y = key->y; 88 d = key->data; 89 90 #if defined(RC4_CHUNK) 91 /* 92 * The original reason for implementing this(*) was the fact that 93 * pre-21164a Alpha CPUs don't have byte load/store instructions 94 * and e.g. a byte store has to be done with 64-bit load, shift, 95 * and, or and finally 64-bit store. Peaking data and operating 96 * at natural word size made it possible to reduce amount of 97 * instructions as well as to perform early read-ahead without 98 * suffering from RAW (read-after-write) hazard. This resulted 99 * in ~40%(**) performance improvement on 21064 box with gcc. 100 * But it's not only Alpha users who win here:-) Thanks to the 101 * early-n-wide read-ahead this implementation also exhibits 102 * >40% speed-up on SPARC and 20-30% on 64-bit MIPS (depending 103 * on sizeof(RC4_INT)). 104 * 105 * (*) "this" means code which recognizes the case when input 106 * and output pointers appear to be aligned at natural CPU 107 * word boundary 108 * (**) i.e. according to 'apps/openssl speed rc4' benchmark, 109 * crypto/rc4/rc4speed.c exhibits almost 70% speed-up... 110 * 111 * Caveats. 112 * 113 * - RC4_CHUNK="unsigned long long" should be a #1 choice for 114 * UltraSPARC. Unfortunately gcc generates very slow code 115 * (2.5-3 times slower than one generated by Sun's WorkShop 116 * C) and therefore gcc (at least 2.95 and earlier) should 117 * always be told that RC4_CHUNK="unsigned long". 118 * 119 * <appro@fy.chalmers.se> 120 */ 121 122 # define RC4_STEP ( \ 123 x=(x+1) &0xff, \ 124 tx=d[x], \ 125 y=(tx+y)&0xff, \ 126 ty=d[y], \ 127 d[y]=tx, \ 128 d[x]=ty, \ 129 (RC4_CHUNK)d[(tx+ty)&0xff]\ 130 ) 131 132 if ((((size_t)indata & (sizeof(RC4_CHUNK) - 1)) | 133 ((size_t)outdata & (sizeof(RC4_CHUNK) - 1))) == 0 ) { 134 RC4_CHUNK ichunk, otp; 135 136 /* 137 * I reckon we can afford to implement both endian 138 * cases and to decide which way to take at run-time 139 * because the machine code appears to be very compact 140 * and redundant 1-2KB is perfectly tolerable (i.e. 141 * in case the compiler fails to eliminate it:-). By 142 * suggestion from Terrel Larson <terr@terralogic.net>. 143 * 144 * Special notes. 145 * 146 * - compilers (those I've tried) don't seem to have 147 * problems eliminating either the operators guarded 148 * by "if (sizeof(RC4_CHUNK)==8)" or the condition 149 * expressions themselves so I've got 'em to replace 150 * corresponding #ifdefs from the previous version; 151 * - I chose to let the redundant switch cases when 152 * sizeof(RC4_CHUNK)!=8 be (were also #ifdefed 153 * before); 154 * - in case you wonder "&(sizeof(RC4_CHUNK)*8-1)" in 155 * [LB]ESHFT guards against "shift is out of range" 156 * warnings when sizeof(RC4_CHUNK)!=8 157 * 158 * <appro@fy.chalmers.se> 159 */ 160 #if BYTE_ORDER == BIG_ENDIAN 161 # define BESHFT(c) (((sizeof(RC4_CHUNK)-(c)-1)*8)&(sizeof(RC4_CHUNK)*8-1)) 162 for (; len & (0 - sizeof(RC4_CHUNK)); len -= sizeof(RC4_CHUNK)) { 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 otp |= RC4_STEP << BESHFT(4); 170 otp |= RC4_STEP << BESHFT(5); 171 otp |= RC4_STEP << BESHFT(6); 172 otp |= RC4_STEP << BESHFT(7); 173 } 174 *(RC4_CHUNK *)outdata = otp^ichunk; 175 indata += sizeof(RC4_CHUNK); 176 outdata += sizeof(RC4_CHUNK); 177 } 178 #else 179 # define LESHFT(c) (((c)*8)&(sizeof(RC4_CHUNK)*8-1)) 180 for (; len & (0 - sizeof(RC4_CHUNK)); len -= sizeof(RC4_CHUNK)) { 181 ichunk = *(RC4_CHUNK *)indata; 182 otp = RC4_STEP; 183 otp |= RC4_STEP << 8; 184 otp |= RC4_STEP << 16; 185 otp |= RC4_STEP << 24; 186 if (sizeof(RC4_CHUNK) == 8) { 187 otp |= RC4_STEP << LESHFT(4); 188 otp |= RC4_STEP << LESHFT(5); 189 otp |= RC4_STEP << LESHFT(6); 190 otp |= RC4_STEP << LESHFT(7); 191 } 192 *(RC4_CHUNK *)outdata = otp ^ ichunk; 193 indata += sizeof(RC4_CHUNK); 194 outdata += sizeof(RC4_CHUNK); 195 } 196 #endif 197 } 198 #endif 199 #define RC4_LOOP(in,out) \ 200 x=((x+1)&0xff); \ 201 tx=d[x]; \ 202 y=(tx+y)&0xff; \ 203 d[x]=ty=d[y]; \ 204 d[y]=tx; \ 205 (out) = d[(tx+ty)&0xff]^ (in); 206 207 i = len >> 3; 208 if (i) { 209 for (;;) { 210 RC4_LOOP(indata[0], outdata[0]); 211 RC4_LOOP(indata[1], outdata[1]); 212 RC4_LOOP(indata[2], outdata[2]); 213 RC4_LOOP(indata[3], outdata[3]); 214 RC4_LOOP(indata[4], outdata[4]); 215 RC4_LOOP(indata[5], outdata[5]); 216 RC4_LOOP(indata[6], outdata[6]); 217 RC4_LOOP(indata[7], outdata[7]); 218 219 indata += 8; 220 outdata += 8; 221 222 if (--i == 0) 223 break; 224 } 225 } 226 i = len&0x07; 227 if (i) { 228 for (;;) { 229 RC4_LOOP(indata[0], outdata[0]); 230 if (--i == 0) 231 break; 232 RC4_LOOP(indata[1], outdata[1]); 233 if (--i == 0) 234 break; 235 RC4_LOOP(indata[2], outdata[2]); 236 if (--i == 0) 237 break; 238 RC4_LOOP(indata[3], outdata[3]); 239 if (--i == 0) 240 break; 241 RC4_LOOP(indata[4], outdata[4]); 242 if (--i == 0) 243 break; 244 RC4_LOOP(indata[5], outdata[5]); 245 if (--i == 0) 246 break; 247 RC4_LOOP(indata[6], outdata[6]); 248 if (--i == 0) 249 break; 250 } 251 } 252 key->x = x; 253 key->y = y; 254 } 255 #endif 256 257 #ifdef HAVE_RC4_SET_KEY_INTERNAL 258 void rc4_set_key_internal(RC4_KEY *key, int len, const unsigned char *data); 259 260 #else 261 static inline void 262 rc4_set_key_internal(RC4_KEY *key, int len, const unsigned char *data) 263 { 264 RC4_INT tmp; 265 int id1, id2; 266 RC4_INT *d; 267 unsigned int i; 268 269 d = &(key->data[0]); 270 key->x = 0; 271 key->y = 0; 272 id1 = id2 = 0; 273 274 #define SK_LOOP(d,n) { \ 275 tmp=d[(n)]; \ 276 id2 = (data[id1] + tmp + id2) & 0xff; \ 277 if (++id1 == len) id1=0; \ 278 d[(n)]=d[id2]; \ 279 d[id2]=tmp; } 280 281 for (i = 0; i < 256; i++) 282 d[i] = i; 283 for (i = 0; i < 256; i += 4) { 284 SK_LOOP(d, i + 0); 285 SK_LOOP(d, i + 1); 286 SK_LOOP(d, i + 2); 287 SK_LOOP(d, i + 3); 288 } 289 } 290 #endif 291 292 void 293 RC4(RC4_KEY *key, size_t len, const unsigned char *indata, 294 unsigned char *outdata) 295 { 296 rc4_internal(key, len, indata, outdata); 297 } 298 LCRYPTO_ALIAS(RC4); 299 300 void 301 RC4_set_key(RC4_KEY *key, int len, const unsigned char *data) 302 { 303 rc4_set_key_internal(key, len, data); 304 } 305 LCRYPTO_ALIAS(RC4_set_key); 306