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
  65 #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))
  66
  67 #define opt_B(s) (((s->b >> 6) ^ (s->b >> 5) ^ (s->b >> 4) ^ (s->b)) & 0x1)
  68
  69 #define opt__select(x,y,r)  (4 & (((r & (r << 2)) >> 5) ^ ((r & ~(r << 2)) >> 4) ^ ( (r | r << 2) >> 3)))\
  70         |(2 & (((r | r << 2) >> 6) ^ ( (r | r << 2) >> 1) ^ (r >> 5) ^ r ^ ((x^y) << 1)))\
  71         |(1 & (((r & ~(r << 2)) >> 4) ^ ((r & (r << 2)) >> 3) ^ r ^ x))
  72
  73 /*
  74  * Some background on the expression above can be found here...
  75 uint8_t xopt__select(bool x, bool y, uint8_t r)
  76 {
  77         uint8_t r_ls2 = r << 2;
  78         uint8_t r_and_ls2 = r & r_ls2;
  79         uint8_t r_or_ls2  = r | r_ls2;
  80
  81         //r:      r0 r1 r2 r3 r4 r5 r6 r7
  82         //r_ls2:  r2 r3 r4 r5 r6 r7  0  0
  83         //                       z0
  84         //                          z1
  85
  86 //      uint8_t z0 = (r0 & r2) ^ (r1 & ~r3) ^ (r2 | r4); // <-- original
  87         uint8_t z0 = (r_and_ls2 >> 5) ^ ((r & ~r_ls2) >> 4) ^ ( r_or_ls2 >> 3);
  88
  89 //      uint8_t z1 = (r0 | r2) ^ ( r5 | r7) ^ r1 ^ r6 ^ x ^ y;  // <-- original
  90         uint8_t z1 = (r_or_ls2 >> 6) ^ ( r_or_ls2 >> 1) ^ (r >> 5) ^ r ^ ((x^y) << 1);
  91
  92 //      uint8_t z2 = (r3 & ~r5) ^ (r4 & r6 ) ^ r7 ^ x;  // <-- original
  93         uint8_t z2 = ((r & ~r_ls2) >> 4) ^ (r_and_ls2 >> 3) ^ r ^ x;
  94
  95         return (z0 & 4) | (z1 & 2) | (z2 & 1);
  96 }
  97 */
  98
  99 void opt_successor(const uint8_t* k, State *s, bool y, State* successor)
 100 {
 101
 102         uint8_t Tt = 1 & opt_T(s);
 103
 104         successor->t = (s->t >> 1);
 105         successor->t |= (Tt ^ (s->r >> 7 & 0x1) ^ (s->r >> 3 & 0x1)) << 15;
 106
 107         successor->b = s->b >> 1;
 108         successor->b |= (opt_B(s) ^ (s->r & 0x1)) << 7;
 109
 110         successor->r = (k[opt__select(Tt,y,s->r)] ^ successor->b) + s->l ;
 111         successor->l = successor->r+s->r;
 112
 113 }
 114
 115 void opt_suc(const uint8_t* k,State* s, uint8_t *in, uint8_t length, bool add32Zeroes)
 116 {
 117         State x2;
 118         int i;
 119         uint8_t head = 0;
 120         for(i =0 ; i < length  ; i++)
 121         {
 122                 head = 1 & (in[i] >> 7);
 123                 opt_successor(k,s,head,&x2);
 124
 125                 head = 1 & (in[i] >> 6);
 126                 opt_successor(k,&x2,head,s);
 127
 128                 head = 1 & (in[i] >> 5);
 129                 opt_successor(k,s,head,&x2);
 130
 131                 head = 1 & (in[i] >> 4);
 132                 opt_successor(k,&x2,head,s);
 133
 134                 head = 1 & (in[i] >> 3);
 135                 opt_successor(k,s,head,&x2);
 136
 137                 head = 1 & (in[i] >> 2);
 138                 opt_successor(k,&x2,head,s);
 139
 140                 head = 1 & (in[i] >> 1);
 141                 opt_successor(k,s,head,&x2);
 142
 143                 head = 1 & in[i];
 144                 opt_successor(k,&x2,head,s);
 145
 146         }
 147         //For tag MAC, an additional 32 zeroes
 148         if(add32Zeroes)
 149                 for(i =0 ; i < 16 ; i++)
 150                 {
 151                         opt_successor(k,s,0,&x2);
 152                         opt_successor(k,&x2,0,s);
 153                 }
 154 }
 155
 156 void opt_output(const uint8_t* k,State* s,  uint8_t *buffer)
 157 {
 158         uint8_t times = 0;
 159         uint8_t bout = 0;
 160         State temp = {0,0,0,0};
 161         for( ; times < 4 ; times++)
 162         {
 163                 bout =0;
 164                 bout |= (s->r & 0x4) << 5;
 165                 opt_successor(k,s,0,&temp);
 166                 bout |= (temp.r & 0x4) << 4;
 167                 opt_successor(k,&temp,0,s);
 168                 bout |= (s->r & 0x4) << 3;
 169                 opt_successor(k,s,0,&temp);
 170                 bout |= (temp.r & 0x4) << 2;
 171                 opt_successor(k,&temp,0,s);
 172                 bout |= (s->r & 0x4) << 1;
 173                 opt_successor(k,s,0,&temp);
 174                 bout |= (temp.r & 0x4) ;
 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) >> 2;
 179                 opt_successor(k,&temp,0,s);
 180                 buffer[times] = bout;
 181         }
 182
 183 }
 184
 185 void opt_MAC(uint8_t* k, uint8_t* input, uint8_t* out)
 186 {
 187         State _init  =  {
 188                         ((k[0] ^ 0x4c) + 0xEC) & 0xFF,// l
 189                         ((k[0] ^ 0x4c) + 0x21) & 0xFF,// r
 190                         0x4c, // b
 191                         0xE012 // t
 192                         };
 193
 194         opt_suc(k,&_init,input,12, false);
 195         //printf("\noutp ");
 196         opt_output(k,&_init, out);
 197 }
 198 uint8_t rev_byte(uint8_t b) {
 199         b = (b & 0xF0) >> 4 | (b & 0x0F) << 4;
 200         b = (b & 0xCC) >> 2 | (b & 0x33) << 2;
 201         b = (b & 0xAA) >> 1 | (b & 0x55) << 1;
 202    return b;
 203 }
 204 void opt_reverse_arraybytecpy(uint8_t* dest, uint8_t *src, size_t len)
 205 {
 206         uint8_t i;
 207         for( i =0; i< len ; i++)
 208                 dest[i] = rev_byte(src[i]);
 209 }
 210
 211 void opt_doReaderMAC(uint8_t *cc_nr_p, uint8_t *div_key_p, uint8_t mac[4])
 212 {
 213         static uint8_t cc_nr[12];
 214
 215         opt_reverse_arraybytecpy(cc_nr, cc_nr_p,12);
 216         uint8_t dest []= {0,0,0,0,0,0,0,0};
 217         opt_MAC(div_key_p,cc_nr, dest);
 218         //The output MAC must also be reversed
 219         opt_reverse_arraybytecpy(mac, dest,4);
 220         return;
 221 }
 222 void opt_doTagMAC(uint8_t *cc_p, const uint8_t *div_key_p, uint8_t mac[4])
 223 {
 224         static uint8_t cc_nr[8+4+4];
 225         opt_reverse_arraybytecpy(cc_nr, cc_p,12);
 226         State _init  =  {
 227                         ((div_key_p[0] ^ 0x4c) + 0xEC) & 0xFF,// l
 228                         ((div_key_p[0] ^ 0x4c) + 0x21) & 0xFF,// r
 229                         0x4c, // b
 230                         0xE012 // t
 231                         };
 232         opt_suc(div_key_p,&_init,cc_nr, 12,true);
 233         uint8_t dest []= {0,0,0,0};
 234         opt_output(div_key_p,&_init, dest);
 235         //The output MAC must also be reversed
 236         opt_reverse_arraybytecpy(mac, dest,4);
 237         return;
 238
 239 }
 240 /**
 241  * The tag MAC can be divided (both can, but no point in dividing the reader mac) into
 242  * two functions, since the first 8 bytes are known, we can pre-calculate the state
 243  * reached after feeding CC to the cipher.
 244  * @param cc_p
 245  * @param div_key_p
 246  * @return the cipher state
 247  */
 248 State opt_doTagMAC_1(uint8_t *cc_p, const uint8_t *div_key_p)
 249 {
 250         static uint8_t cc_nr[8];
 251         opt_reverse_arraybytecpy(cc_nr, cc_p,8);
 252         State _init  =  {
 253                         ((div_key_p[0] ^ 0x4c) + 0xEC) & 0xFF,// l
 254                         ((div_key_p[0] ^ 0x4c) + 0x21) & 0xFF,// r
 255                         0x4c, // b
 256                         0xE012 // t
 257                         };
 258         opt_suc(div_key_p,&_init,cc_nr, 8,false);
 259         return _init;
 260 }
 261 /**
 262  * The second part of the tag MAC calculation, since the CC is already calculated into the state,
 263  * this function is fed only the NR, and internally feeds the remaining 32 0-bits to generate the tag
 264  * MAC response.
 265  * @param _init - precalculated cipher state
 266  * @param nr - the reader challenge
 267  * @param mac - where to store the MAC
 268  * @param div_key_p - the key to use
 269  */
 270 void opt_doTagMAC_2(State _init,  uint8_t* nr, uint8_t mac[4], const uint8_t* div_key_p)
 271 {
 272         static uint8_t _nr [4];
 273         opt_reverse_arraybytecpy(_nr, nr, 4);
 274         opt_suc(div_key_p,&_init,_nr, 4, true);
 275         //opt_suc(div_key_p,&_init,nr, 4, false);
 276         uint8_t dest []= {0,0,0,0};
 277         opt_output(div_key_p,&_init, dest);
 278         //The output MAC must also be reversed
 279         opt_reverse_arraybytecpy(mac, dest,4);
 280         return;
 281 }