git.zerfleddert.de Git - proxmark3-svn/blob - armsrc/optimized_cipher.c

   1 /*****************************************************************************
   2  * WARNING
   3  *
   4  * THIS CODE IS CREATED FOR EXPERIMENTATION AND EDUCATIONAL USE ONLY.
   5  *
   6  * USAGE OF THIS CODE IN OTHER WAYS MAY INFRINGE UPON THE INTELLECTUAL
   7  * PROPERTY OF OTHER PARTIES, SUCH AS INSIDE SECURE AND HID GLOBAL,
   8  * AND MAY EXPOSE YOU TO AN INFRINGEMENT ACTION FROM THOSE PARTIES.
   9  *
  10  * THIS CODE SHOULD NEVER BE USED TO INFRINGE PATENTS OR INTELLECTUAL PROPERTY RIGHTS.
  11  *
  12  *****************************************************************************
  13  *
  14  * This file is part of loclass. It is a reconstructon of the cipher engine
  15  * used in iClass, and RFID techology.
  16  *
  17  * The implementation is based on the work performed by
  18  * Flavio D. Garcia, Gerhard de Koning Gans, Roel Verdult and
  19  * Milosch Meriac in the paper "Dismantling IClass".
  20  *
  21  * Copyright (C) 2014 Martin Holst Swende
  22  *
  23  * This is free software: you can redistribute it and/or modify
  24  * it under the terms of the GNU General Public License version 2 as published
  25  * by the Free Software Foundation.
  26  *
  27  * This file is distributed in the hope that it will be useful,
  28  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  29  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  30  * GNU General Public License for more details.
  31  *
  32  * You should have received a copy of the GNU General Public License
  33  * along with loclass.  If not, see <http://www.gnu.org/licenses/>.
  34  *
  35  *
  36  *
  37  ****************************************************************************/
  38
  39 /**
  40
  41   This file contains an optimized version of the MAC-calculation algorithm. Some measurements on
  42   a std laptop showed it runs in about 1/3 of the time:
  43
  44         Std: 0.428962
  45         Opt: 0.151609
  46
  47   Additionally, it is self-reliant, not requiring e.g. bitstreams from the cipherutils, thus can
  48   be easily dropped into a code base.
  49
  50   The optimizations have been performed in the following steps:
  51   * Parameters passed by reference instead of by value.
  52   * Iteration instead of recursion, un-nesting recursive loops into for-loops.
  53   * Handling of bytes instead of individual bits, for less shuffling and masking
  54   * Less creation of "objects", structs, and instead reuse of alloc:ed memory
  55   * Inlining some functions via #define:s
  56
  57   As a consequence, this implementation is less generic. Also, I haven't bothered documenting this.
  58   For a thorough documentation, check out the MAC-calculation within cipher.c instead.
  59
  60   -- MHS 2015
  61 **/
  62
  63 #include "optimized_cipher.h"
  64 #include <stddef.h>
  65 #include <stdbool.h>
  66 #include <stdint.h>
  67
  68
  69 #define opt_T(s) (0x1 & ((s->t >> 15) ^ (s->t >> 14)^ (s->t >> 10)^ (s->t >> 8)^ (s->t >> 5)^ (s->t >> 4)^ (s->t >> 1)^ s->t))
  70
  71 #define opt_B(s) (((s->b >> 6) ^ (s->b >> 5) ^ (s->b >> 4) ^ (s->b)) & 0x1)
  72
  73 #define opt__select(x,y,r)  (4 & (((r & (r << 2)) >> 5) ^ ((r & ~(r << 2)) >> 4) ^ ( (r | r << 2) >> 3)))\
  74         |(2 & (((r | r << 2) >> 6) ^ ( (r | r << 2) >> 1) ^ (r >> 5) ^ r ^ ((x^y) << 1)))\
  75         |(1 & (((r & ~(r << 2)) >> 4) ^ ((r & (r << 2)) >> 3) ^ r ^ x))
  76
  77 /*
  78  * Some background on the expression above can be found here...
  79 uint8_t xopt__select(bool x, bool y, uint8_t r)
  80 {
  81         uint8_t r_ls2 = r << 2;
  82         uint8_t r_and_ls2 = r & r_ls2;
  83         uint8_t r_or_ls2  = r | r_ls2;
  84
  85         //r:      r0 r1 r2 r3 r4 r5 r6 r7
  86         //r_ls2:  r2 r3 r4 r5 r6 r7  0  0
  87         //                       z0
  88         //                          z1
  89
  90 //      uint8_t z0 = (r0 & r2) ^ (r1 & ~r3) ^ (r2 | r4); // <-- original
  91         uint8_t z0 = (r_and_ls2 >> 5) ^ ((r & ~r_ls2) >> 4) ^ ( r_or_ls2 >> 3);
  92
  93 //      uint8_t z1 = (r0 | r2) ^ ( r5 | r7) ^ r1 ^ r6 ^ x ^ y;  // <-- original
  94         uint8_t z1 = (r_or_ls2 >> 6) ^ ( r_or_ls2 >> 1) ^ (r >> 5) ^ r ^ ((x^y) << 1);
  95
  96 //      uint8_t z2 = (r3 & ~r5) ^ (r4 & r6 ) ^ r7 ^ x;  // <-- original
  97         uint8_t z2 = ((r & ~r_ls2) >> 4) ^ (r_and_ls2 >> 3) ^ r ^ x;
  98
  99         return (z0 & 4) | (z1 & 2) | (z2 & 1);
 100 }
 101 */
 102
 103 void opt_successor(const uint8_t* k, State *s, bool y, State* successor)
 104 {
 105
 106         uint8_t Tt = 1 & opt_T(s);
 107
 108         successor->t = (s->t >> 1);
 109         successor->t |= (Tt ^ (s->r >> 7 & 0x1) ^ (s->r >> 3 & 0x1)) << 15;
 110
 111         successor->b = s->b >> 1;
 112         successor->b |= (opt_B(s) ^ (s->r & 0x1)) << 7;
 113
 114         successor->r = (k[opt__select(Tt,y,s->r)] ^ successor->b) + s->l ;
 115         successor->l = successor->r+s->r;
 116
 117 }
 118
 119 void opt_suc(const uint8_t* k,State* s, uint8_t *in, uint8_t length, bool add32Zeroes)
 120 {
 121         State x2;
 122         int i;
 123         uint8_t head = 0;
 124         for(i =0 ; i < length  ; i++)
 125         {
 126                 head = 1 & (in[i] >> 7);
 127                 opt_successor(k,s,head,&x2);
 128
 129                 head = 1 & (in[i] >> 6);
 130                 opt_successor(k,&x2,head,s);
 131
 132                 head = 1 & (in[i] >> 5);
 133                 opt_successor(k,s,head,&x2);
 134
 135                 head = 1 & (in[i] >> 4);
 136                 opt_successor(k,&x2,head,s);
 137
 138                 head = 1 & (in[i] >> 3);
 139                 opt_successor(k,s,head,&x2);
 140
 141                 head = 1 & (in[i] >> 2);
 142                 opt_successor(k,&x2,head,s);
 143
 144                 head = 1 & (in[i] >> 1);
 145                 opt_successor(k,s,head,&x2);
 146
 147                 head = 1 & in[i];
 148                 opt_successor(k,&x2,head,s);
 149
 150         }
 151         //For tag MAC, an additional 32 zeroes
 152         if(add32Zeroes)
 153                 for(i =0 ; i < 16 ; i++)
 154                 {
 155                         opt_successor(k,s,0,&x2);
 156                         opt_successor(k,&x2,0,s);
 157                 }
 158 }
 159
 160 void opt_output(const uint8_t* k,State* s,  uint8_t *buffer)
 161 {
 162         uint8_t times = 0;
 163         uint8_t bout = 0;
 164         State temp = {0,0,0,0};
 165         for( ; times < 4 ; times++)
 166         {
 167                 bout =0;
 168                 bout |= (s->r & 0x4) << 5;
 169                 opt_successor(k,s,0,&temp);
 170                 bout |= (temp.r & 0x4) << 4;
 171                 opt_successor(k,&temp,0,s);
 172                 bout |= (s->r & 0x4) << 3;
 173                 opt_successor(k,s,0,&temp);
 174                 bout |= (temp.r & 0x4) << 2;
 175                 opt_successor(k,&temp,0,s);
 176                 bout |= (s->r & 0x4) << 1;
 177                 opt_successor(k,s,0,&temp);
 178                 bout |= (temp.r & 0x4) ;
 179                 opt_successor(k,&temp,0,s);
 180                 bout |= (s->r & 0x4) >> 1;
 181                 opt_successor(k,s,0,&temp);
 182                 bout |= (temp.r & 0x4) >> 2;
 183                 opt_successor(k,&temp,0,s);
 184                 buffer[times] = bout;
 185         }
 186
 187 }
 188
 189 void opt_MAC(uint8_t* k, uint8_t* input, uint8_t* out)
 190 {
 191         State _init  =  {
 192                         ((k[0] ^ 0x4c) + 0xEC) & 0xFF,// l
 193                         ((k[0] ^ 0x4c) + 0x21) & 0xFF,// r
 194                         0x4c, // b
 195                         0xE012 // t
 196                         };
 197
 198         opt_suc(k,&_init,input,12, false);
 199         //printf("\noutp ");
 200         opt_output(k,&_init, out);
 201 }
 202 uint8_t rev_byte(uint8_t b) {
 203         b = (b & 0xF0) >> 4 | (b & 0x0F) << 4;
 204         b = (b & 0xCC) >> 2 | (b & 0x33) << 2;
 205         b = (b & 0xAA) >> 1 | (b & 0x55) << 1;
 206    return b;
 207 }
 208 void opt_reverse_arraybytecpy(uint8_t* dest, uint8_t *src, size_t len)
 209 {
 210         uint8_t i;
 211         for( i =0; i< len ; i++)
 212                 dest[i] = rev_byte(src[i]);
 213 }
 214
 215 void opt_doReaderMAC(uint8_t *cc_nr_p, uint8_t *div_key_p, uint8_t mac[4])
 216 {
 217         static uint8_t cc_nr[12];
 218
 219         opt_reverse_arraybytecpy(cc_nr, cc_nr_p,12);
 220         uint8_t dest []= {0,0,0,0,0,0,0,0};
 221         opt_MAC(div_key_p,cc_nr, dest);
 222         //The output MAC must also be reversed
 223         opt_reverse_arraybytecpy(mac, dest,4);
 224         return;
 225 }
 226 void opt_doTagMAC(uint8_t *cc_p, const uint8_t *div_key_p, uint8_t mac[4])
 227 {
 228         static uint8_t cc_nr[8+4+4];
 229         opt_reverse_arraybytecpy(cc_nr, cc_p,12);
 230         State _init  =  {
 231                         ((div_key_p[0] ^ 0x4c) + 0xEC) & 0xFF,// l
 232                         ((div_key_p[0] ^ 0x4c) + 0x21) & 0xFF,// r
 233                         0x4c, // b
 234                         0xE012 // t
 235                         };
 236         opt_suc(div_key_p,&_init,cc_nr, 12,true);
 237         uint8_t dest []= {0,0,0,0};
 238         opt_output(div_key_p,&_init, dest);
 239         //The output MAC must also be reversed
 240         opt_reverse_arraybytecpy(mac, dest,4);
 241         return;
 242
 243 }
 244 /**
 245  * The tag MAC can be divided (both can, but no point in dividing the reader mac) into
 246  * two functions, since the first 8 bytes are known, we can pre-calculate the state
 247  * reached after feeding CC to the cipher.
 248  * @param cc_p
 249  * @param div_key_p
 250  * @return the cipher state
 251  */
 252 State opt_doTagMAC_1(uint8_t *cc_p, const uint8_t *div_key_p)
 253 {
 254         static uint8_t cc_nr[8];
 255         opt_reverse_arraybytecpy(cc_nr, cc_p,8);
 256         State _init  =  {
 257                         ((div_key_p[0] ^ 0x4c) + 0xEC) & 0xFF,// l
 258                         ((div_key_p[0] ^ 0x4c) + 0x21) & 0xFF,// r
 259                         0x4c, // b
 260                         0xE012 // t
 261                         };
 262         opt_suc(div_key_p,&_init,cc_nr, 8,false);
 263         return _init;
 264 }
 265 /**
 266  * The second part of the tag MAC calculation, since the CC is already calculated into the state,
 267  * this function is fed only the NR, and internally feeds the remaining 32 0-bits to generate the tag
 268  * MAC response.
 269  * @param _init - precalculated cipher state
 270  * @param nr - the reader challenge
 271  * @param mac - where to store the MAC
 272  * @param div_key_p - the key to use
 273  */
 274 void opt_doTagMAC_2(State _init,  uint8_t* nr, uint8_t mac[4], const uint8_t* div_key_p)
 275 {
 276         static uint8_t _nr [4];
 277         opt_reverse_arraybytecpy(_nr, nr, 4);
 278         opt_suc(div_key_p,&_init,_nr, 4, true);
 279         //opt_suc(div_key_p,&_init,nr, 4, false);
 280         uint8_t dest []= {0,0,0,0};
 281         opt_output(div_key_p,&_init, dest);
 282         //The output MAC must also be reversed
 283         opt_reverse_arraybytecpy(mac, dest,4);
 284         return;
 285 }