1 /* Modified for SILC. -Pekka */
3 /* This is an independent implementation of the encryption algorithm: */
5 /* Twofish by Bruce Schneier and colleagues */
7 /* which is a candidate algorithm in the Advanced Encryption Standard */
8 /* programme of the US National Institute of Standards and Technology. */
10 /* Copyright in this implementation is held by Dr B R Gladman but I */
11 /* hereby give permission for its free direct or derivative use subject */
12 /* to acknowledgment of its origin and compliance with any conditions */
13 /* that the originators of t he algorithm place on its exploitation. */
15 /* My thanks to Doug Whiting and Niels Ferguson for comments that led */
16 /* to improvements in this implementation. */
18 /* Dr Brian Gladman (gladman@seven77.demon.co.uk) 14th January 1999 */
20 /* Timing data for Twofish (twofish.c)
23 Key Setup: 8414 cycles
24 Encrypt: 376 cycles = 68.1 mbits/sec
25 Decrypt: 374 cycles = 68.4 mbits/sec
26 Mean: 375 cycles = 68.3 mbits/sec
29 Key Setup: 11628 cycles
30 Encrypt: 376 cycles = 68.1 mbits/sec
31 Decrypt: 374 cycles = 68.4 mbits/sec
32 Mean: 375 cycles = 68.3 mbits/sec
35 Key Setup: 15457 cycles
36 Encrypt: 381 cycles = 67.2 mbits/sec
37 Decrypt: 374 cycles = 68.4 mbits/sec
38 Mean: 378 cycles = 67.8 mbits/sec
43 #include "twofish_internal.h"
47 * SILC Crypto API for Twofish
50 /* Sets the key for the cipher. */
52 SILC_CIPHER_API_SET_KEY(twofish_cbc)
56 SILC_GET_WORD_KEY(key, k, keylen);
57 twofish_set_key((TwofishContext *)context, k, keylen);
62 /* Returns the size of the cipher context. */
64 SILC_CIPHER_API_CONTEXT_LEN(twofish_cbc)
66 return sizeof(TwofishContext);
69 /* Encrypts with the cipher in CBC mode. Source and destination buffers
70 maybe one and same. */
72 SILC_CIPHER_API_ENCRYPT(twofish_cbc)
77 SILC_ASSERT((len & (16 - 1)) == 0);
80 SILC_CBC_GET_IV(tiv, iv);
82 SILC_CBC_ENC_PRE(tiv, src);
83 twofish_encrypt((TwofishContext *)context, tiv, tiv);
84 SILC_CBC_ENC_POST(tiv, dst, src);
86 for (i = 16; i < len; i += 16) {
87 SILC_CBC_ENC_PRE(tiv, src);
88 twofish_encrypt((TwofishContext *)context, tiv, tiv);
89 SILC_CBC_ENC_POST(tiv, dst, src);
92 SILC_CBC_PUT_IV(tiv, iv);
97 /* Decrypts with the cipher in CBC mode. Source and destination buffers
98 maybe one and same. */
100 SILC_CIPHER_API_DECRYPT(twofish_cbc)
102 SilcUInt32 tmp[4], tmp2[4], tiv[4];
108 SILC_CBC_GET_IV(tiv, iv);
110 SILC_CBC_DEC_PRE(tmp, src);
111 twofish_decrypt((TwofishContext *)context, tmp, tmp2);
112 SILC_CBC_DEC_POST(tmp2, dst, src, tmp, tiv);
114 for (i = 16; i < len; i += 16) {
115 SILC_CBC_DEC_PRE(tmp, src);
116 twofish_decrypt((TwofishContext *)context, tmp, tmp2);
117 SILC_CBC_DEC_POST(tmp2, dst, src, tmp, tiv);
120 SILC_CBC_PUT_IV(tiv, iv);
132 /* finite field arithmetic for GF(2**8) with the modular */
133 /* polynomial x^8 + x^6 + x^5 + x^3 + 1 (0x169) */
137 u1byte tab_5b[4] = { 0, G_M >> 2, G_M >> 1, (G_M >> 1) ^ (G_M >> 2) };
138 u1byte tab_ef[4] = { 0, (G_M >> 1) ^ (G_M >> 2), G_M >> 1, G_M >> 2 };
140 #define ffm_01(x) (x)
141 #define ffm_5b(x) ((x) ^ ((x) >> 2) ^ tab_5b[(x) & 3])
142 #define ffm_ef(x) ((x) ^ ((x) >> 1) ^ ((x) >> 2) ^ tab_ef[(x) & 3])
144 u1byte ror4[16] = { 0, 8, 1, 9, 2, 10, 3, 11, 4, 12, 5, 13, 6, 14, 7, 15 };
145 u1byte ashx[16] = { 0, 9, 2, 11, 4, 13, 6, 15, 8, 1, 10, 3, 12, 5, 14, 7 };
148 { { 8, 1, 7, 13, 6, 15, 3, 2, 0, 11, 5, 9, 14, 12, 10, 4 },
149 { 2, 8, 11, 13, 15, 7, 6, 14, 3, 1, 9, 4, 0, 10, 12, 5 }
153 { { 14, 12, 11, 8, 1, 2, 3, 5, 15, 4, 10, 6, 7, 0, 9, 13 },
154 { 1, 14, 2, 11, 4, 12, 3, 7, 6, 13, 10, 5, 15, 9, 0, 8 }
158 { { 11, 10, 5, 14, 6, 13, 9, 0, 12, 8, 15, 3, 2, 4, 7, 1 },
159 { 4, 12, 7, 5, 1, 6, 9, 10, 0, 14, 13, 8, 2, 11, 3, 15 }
163 { { 13, 7, 15, 4, 1, 2, 6, 14, 9, 11, 3, 0, 8, 5, 12, 10 },
164 { 11, 9, 5, 1, 12, 3, 13, 14, 6, 4, 7, 15, 2, 0, 8, 10 }
167 u1byte qp(const u4byte n, const u1byte x)
168 { u1byte a0, a1, a2, a3, a4, b0, b1, b2, b3, b4;
170 a0 = x >> 4; b0 = x & 15;
171 a1 = a0 ^ b0; b1 = ror4[b0] ^ ashx[a0];
172 a2 = qt0[n][a1]; b2 = qt1[n][b1];
173 a3 = a2 ^ b2; b3 = ror4[b2] ^ ashx[a2];
174 a4 = qt2[n][a3]; b4 = qt3[n][b3];
175 return (b4 << 4) | a4;
181 u1byte q_tab[2][256];
183 #define q(n,x) q_tab[n][x]
188 for(i = 0; i < 256; ++i)
190 q(0,i) = qp(0, (u1byte)i);
191 q(1,i) = qp(1, (u1byte)i);
197 #define q(n,x) qp(n, x)
204 u4byte m_tab[4][256];
207 { u4byte i, f01, f5b, fef;
209 for(i = 0; i < 256; ++i)
211 f01 = q(1,i); f5b = ffm_5b(f01); fef = ffm_ef(f01);
212 m_tab[0][i] = f01 + (f5b << 8) + (fef << 16) + (fef << 24);
213 m_tab[2][i] = f5b + (fef << 8) + (f01 << 16) + (fef << 24);
215 f01 = q(0,i); f5b = ffm_5b(f01); fef = ffm_ef(f01);
216 m_tab[1][i] = fef + (fef << 8) + (f5b << 16) + (f01 << 24);
217 m_tab[3][i] = f5b + (f01 << 8) + (fef << 16) + (f5b << 24);
221 #define mds(n,x) m_tab[n][x]
229 #define q_0(x) q(1,x)
235 #define q_1(x) q(0,x)
241 #define q_2(x) q(1,x)
247 #define q_3(x) q(0,x)
249 #define f_0(n,x) ((u4byte)fm_0##n(x))
250 #define f_1(n,x) ((u4byte)fm_1##n(x) << 8)
251 #define f_2(n,x) ((u4byte)fm_2##n(x) << 16)
252 #define f_3(n,x) ((u4byte)fm_3##n(x) << 24)
254 #define mds(n,x) f_0(n,q_##n(x)) ^ f_1(n,q_##n(x)) ^ f_2(n,q_##n(x)) ^ f_3(n,q_##n(x))
258 u4byte h_fun(TwofishContext *ctx, const u4byte x, const u4byte key[])
259 { u4byte b0, b1, b2, b3;
262 u4byte m5b_b0, m5b_b1, m5b_b2, m5b_b3;
263 u4byte mef_b0, mef_b1, mef_b2, mef_b3;
266 b0 = byte(x, 0); b1 = byte(x, 1); b2 = byte(x, 2); b3 = byte(x, 3);
270 case 4: b0 = q(1, b0) ^ byte(key[3],0);
271 b1 = q(0, b1) ^ byte(key[3],1);
272 b2 = q(0, b2) ^ byte(key[3],2);
273 b3 = q(1, b3) ^ byte(key[3],3);
274 case 3: b0 = q(1, b0) ^ byte(key[2],0);
275 b1 = q(1, b1) ^ byte(key[2],1);
276 b2 = q(0, b2) ^ byte(key[2],2);
277 b3 = q(0, b3) ^ byte(key[2],3);
278 case 2: b0 = q(0,q(0,b0) ^ byte(key[1],0)) ^ byte(key[0],0);
279 b1 = q(0,q(1,b1) ^ byte(key[1],1)) ^ byte(key[0],1);
280 b2 = q(1,q(0,b2) ^ byte(key[1],2)) ^ byte(key[0],2);
281 b3 = q(1,q(1,b3) ^ byte(key[1],3)) ^ byte(key[0],3);
285 return mds(0, b0) ^ mds(1, b1) ^ mds(2, b2) ^ mds(3, b3);
289 b0 = q(1, b0); b1 = q(0, b1); b2 = q(1, b2); b3 = q(0, b3);
290 m5b_b0 = ffm_5b(b0); m5b_b1 = ffm_5b(b1); m5b_b2 = ffm_5b(b2); m5b_b3 = ffm_5b(b3);
291 mef_b0 = ffm_ef(b0); mef_b1 = ffm_ef(b1); mef_b2 = ffm_ef(b2); mef_b3 = ffm_ef(b3);
292 b0 ^= mef_b1 ^ m5b_b2 ^ m5b_b3; b3 ^= m5b_b0 ^ mef_b1 ^ mef_b2;
293 b2 ^= mef_b0 ^ m5b_b1 ^ mef_b3; b1 ^= mef_b0 ^ mef_b2 ^ m5b_b3;
295 return b0 | (b3 << 8) | (b2 << 16) | (b1 << 24);
303 u4byte mk_tab[4][256];
308 #define q20(x) q(0,q(0,x) ^ byte(key[1],0)) ^ byte(key[0],0)
309 #define q21(x) q(0,q(1,x) ^ byte(key[1],1)) ^ byte(key[0],1)
310 #define q22(x) q(1,q(0,x) ^ byte(key[1],2)) ^ byte(key[0],2)
311 #define q23(x) q(1,q(1,x) ^ byte(key[1],3)) ^ byte(key[0],3)
313 #define q30(x) q(0,q(0,q(1, x) ^ byte(key[2],0)) ^ byte(key[1],0)) ^ byte(key[0],0)
314 #define q31(x) q(0,q(1,q(1, x) ^ byte(key[2],1)) ^ byte(key[1],1)) ^ byte(key[0],1)
315 #define q32(x) q(1,q(0,q(0, x) ^ byte(key[2],2)) ^ byte(key[1],2)) ^ byte(key[0],2)
316 #define q33(x) q(1,q(1,q(0, x) ^ byte(key[2],3)) ^ byte(key[1],3)) ^ byte(key[0],3)
318 #define q40(x) q(0,q(0,q(1, q(1, x) ^ byte(key[3],0)) ^ byte(key[2],0)) ^ byte(key[1],0)) ^ byte(key[0],0)
319 #define q41(x) q(0,q(1,q(1, q(0, x) ^ byte(key[3],1)) ^ byte(key[2],1)) ^ byte(key[1],1)) ^ byte(key[0],1)
320 #define q42(x) q(1,q(0,q(0, q(0, x) ^ byte(key[3],2)) ^ byte(key[2],2)) ^ byte(key[1],2)) ^ byte(key[0],2)
321 #define q43(x) q(1,q(1,q(0, q(1, x) ^ byte(key[3],3)) ^ byte(key[2],3)) ^ byte(key[1],3)) ^ byte(key[0],3)
323 void gen_mk_tab(TwofishContext *ctx, u4byte key[])
329 case 2: for(i = 0; i < 256; ++i)
333 mk_tab[0][i] = mds(0, q20(by)); mk_tab[1][i] = mds(1, q21(by));
334 mk_tab[2][i] = mds(2, q22(by)); mk_tab[3][i] = mds(3, q23(by));
336 sb[0][i] = q20(by); sb[1][i] = q21(by);
337 sb[2][i] = q22(by); sb[3][i] = q23(by);
342 case 3: for(i = 0; i < 256; ++i)
346 mk_tab[0][i] = mds(0, q30(by)); mk_tab[1][i] = mds(1, q31(by));
347 mk_tab[2][i] = mds(2, q32(by)); mk_tab[3][i] = mds(3, q33(by));
349 sb[0][i] = q30(by); sb[1][i] = q31(by);
350 sb[2][i] = q32(by); sb[3][i] = q33(by);
355 case 4: for(i = 0; i < 256; ++i)
359 mk_tab[0][i] = mds(0, q40(by)); mk_tab[1][i] = mds(1, q41(by));
360 mk_tab[2][i] = mds(2, q42(by)); mk_tab[3][i] = mds(3, q43(by));
362 sb[0][i] = q40(by); sb[1][i] = q41(by);
363 sb[2][i] = q42(by); sb[3][i] = q43(by);
370 # define g0_fun(x) ( mk_tab[0][byte(x,0)] ^ mk_tab[1][byte(x,1)] \
371 ^ mk_tab[2][byte(x,2)] ^ mk_tab[3][byte(x,3)] )
372 # define g1_fun(x) ( mk_tab[0][byte(x,3)] ^ mk_tab[1][byte(x,0)] \
373 ^ mk_tab[2][byte(x,1)] ^ mk_tab[3][byte(x,2)] )
375 # define g0_fun(x) ( mds(0, sb[0][byte(x,0)]) ^ mds(1, sb[1][byte(x,1)]) \
376 ^ mds(2, sb[2][byte(x,2)]) ^ mds(3, sb[3][byte(x,3)]) )
377 # define g1_fun(x) ( mds(0, sb[0][byte(x,3)]) ^ mds(1, sb[1][byte(x,0)]) \
378 ^ mds(2, sb[2][byte(x,1)]) ^ mds(3, sb[3][byte(x,2)]) )
383 #define g0_fun(x) h_fun(ctx,x,s_key)
384 #define g1_fun(x) h_fun(ctx,rotl(x,8),s_key)
388 /* The (12,8) Reed Soloman code has the generator polynomial
390 g(x) = x^4 + (a + 1/a) * x^3 + a * x^2 + (a + 1/a) * x + 1
392 where the coefficients are in the finite field GF(2^8) with a
393 modular polynomial a^8 + a^6 + a^3 + a^2 + 1. To generate the
394 remainder we have to start with a 12th order polynomial with our
395 eight input bytes as the coefficients of the 4th to 11th terms.
398 m[7] * x^11 + m[6] * x^10 ... + m[0] * x^4 + 0 * x^3 +... + 0
400 We then multiply the generator polynomial by m[7] * x^7 and subtract
401 it - xor in GF(2^8) - from the above to eliminate the x^7 term (the
402 artihmetic on the coefficients is done in GF(2^8). We then multiply
403 the generator polynomial by x^6 * coeff(x^10) and use this to remove
404 the x^10 term. We carry on in this way until the x^4 term is removed
405 so that we are left with:
407 r[3] * x^3 + r[2] * x^2 + r[1] 8 x^1 + r[0]
409 which give the resulting 4 bytes of the remainder. This is equivalent
410 to the matrix multiplication in the Twofish description but much faster
415 #define G_MOD 0x0000014d
417 u4byte mds_rem(u4byte p0, u4byte p1)
420 for(i = 0; i < 8; ++i)
422 t = p1 >> 24; /* get most significant coefficient */
424 p1 = (p1 << 8) | (p0 >> 24); p0 <<= 8; /* shift others up */
426 /* multiply t by a (the primitive element - i.e. left shift) */
430 if(t & 0x80) /* subtract modular polynomial on overflow */
434 p1 ^= t ^ (u << 16); /* remove t * (a * x^2 + 1) */
436 u ^= (t >> 1); /* form u = a * t + t / a = t * (a + 1 / a); */
438 if(t & 0x01) /* add the modular polynomial on underflow */
442 p1 ^= (u << 24) | (u << 8); /* remove t * (a + 1/a) * (x^3 + x) */
448 /* initialise the key schedule from the user supplied key */
450 u4byte *twofish_set_key(TwofishContext *ctx,
451 const u4byte in_key[], const u4byte key_len)
453 u4byte i, a, b, me_key[4], mo_key[4];
454 u4byte *l_key = ctx->l_key;
455 u4byte *s_key = ctx->s_key;
460 gen_qtab(); qt_gen = 1;
467 gen_mtab(); mt_gen = 1;
471 ctx->k_len = ctx->k_len = key_len / 64; /* 2, 3 or 4 */
473 for(i = 0; i < ctx->k_len; ++i)
475 a = in_key[i + i]; me_key[i] = a;
476 b = in_key[i + i + 1]; mo_key[i] = b;
477 s_key[ctx->k_len - i - 1] = mds_rem(a, b);
480 for(i = 0; i < 40; i += 2)
482 a = 0x01010101 * i; b = a + 0x01010101;
483 a = h_fun(ctx,a, me_key);
484 b = rotl(h_fun(ctx,b, mo_key), 8);
486 l_key[i + 1] = rotl(a + 2 * b, 9);
490 gen_mk_tab(ctx,s_key);
496 /* encrypt a block of text */
499 t1 = g1_fun(blk[1]); t0 = g0_fun(blk[0]); \
500 blk[2] = rotr(blk[2] ^ (t0 + t1 + l_key[4 * (i) + 8]), 1); \
501 blk[3] = rotl(blk[3], 1) ^ (t0 + 2 * t1 + l_key[4 * (i) + 9]); \
502 t1 = g1_fun(blk[3]); t0 = g0_fun(blk[2]); \
503 blk[0] = rotr(blk[0] ^ (t0 + t1 + l_key[4 * (i) + 10]), 1); \
504 blk[1] = rotl(blk[1], 1) ^ (t0 + 2 * t1 + l_key[4 * (i) + 11])
506 void twofish_encrypt(TwofishContext *ctx,
507 const u4byte in_blk[4], u4byte out_blk[])
509 u4byte t0, t1, blk[4];
510 u4byte *l_key = ctx->l_key;
511 u4byte *s_key = ctx->s_key;
513 blk[0] = in_blk[0] ^ l_key[0];
514 blk[1] = in_blk[1] ^ l_key[1];
515 blk[2] = in_blk[2] ^ l_key[2];
516 blk[3] = in_blk[3] ^ l_key[3];
518 f_rnd(0); f_rnd(1); f_rnd(2); f_rnd(3);
519 f_rnd(4); f_rnd(5); f_rnd(6); f_rnd(7);
521 out_blk[0] = blk[2] ^ l_key[4];
522 out_blk[1] = blk[3] ^ l_key[5];
523 out_blk[2] = blk[0] ^ l_key[6];
524 out_blk[3] = blk[1] ^ l_key[7];
527 /* decrypt a block of text */
530 t1 = g1_fun(blk[1]); t0 = g0_fun(blk[0]); \
531 blk[2] = rotl(blk[2], 1) ^ (t0 + t1 + l_key[4 * (i) + 10]); \
532 blk[3] = rotr(blk[3] ^ (t0 + 2 * t1 + l_key[4 * (i) + 11]), 1); \
533 t1 = g1_fun(blk[3]); t0 = g0_fun(blk[2]); \
534 blk[0] = rotl(blk[0], 1) ^ (t0 + t1 + l_key[4 * (i) + 8]); \
535 blk[1] = rotr(blk[1] ^ (t0 + 2 * t1 + l_key[4 * (i) + 9]), 1)
537 void twofish_decrypt(TwofishContext *ctx,
538 const u4byte in_blk[4], u4byte out_blk[4])
540 u4byte t0, t1, blk[4];
541 u4byte *l_key = ctx->l_key;
542 u4byte *s_key = ctx->s_key;
544 blk[0] = in_blk[0] ^ l_key[4];
545 blk[1] = in_blk[1] ^ l_key[5];
546 blk[2] = in_blk[2] ^ l_key[6];
547 blk[3] = in_blk[3] ^ l_key[7];
549 i_rnd(7); i_rnd(6); i_rnd(5); i_rnd(4);
550 i_rnd(3); i_rnd(2); i_rnd(1); i_rnd(0);
552 out_blk[0] = blk[2] ^ l_key[0];
553 out_blk[1] = blk[3] ^ l_key[1];
554 out_blk[2] = blk[0] ^ l_key[2];
555 out_blk[3] = blk[1] ^ l_key[3];