serpent_generic.c 21.2 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8
/*
 * Cryptographic API.
 *
 * Serpent Cipher Algorithm.
 *
 * Copyright (C) 2002 Dag Arne Osvik <osvik@ii.uib.no>
 *               2003 Herbert Valerio Riedel <hvr@gnu.org>
 *
9 10 11
 * Added tnepres support:
 *		Ruben Jesus Garcia Hernandez <ruben@ugr.es>, 18.10.2004
 *              Based on code by hvr
L
Linus Torvalds 已提交
12 13 14 15 16 17 18 19 20 21 22 23
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 */

#include <linux/init.h>
#include <linux/module.h>
#include <linux/errno.h>
#include <asm/byteorder.h>
#include <linux/crypto.h>
24
#include <linux/types.h>
25
#include <crypto/serpent.h>
L
Linus Torvalds 已提交
26 27 28 29 30 31 32

/* Key is padded to the maximum of 256 bits before round key generation.
 * Any key length <= 256 bits (32 bytes) is allowed by the algorithm.
 */

#define PHI 0x9e3779b9UL

33 34
#define keyiter(a, b, c, d, i, j) \
	({ b ^= d; b ^= c; b ^= a; b ^= PHI ^ i; b = rol32(b, 11); k[j] = b; })
L
Linus Torvalds 已提交
35

36 37
#define loadkeys(x0, x1, x2, x3, i) \
	({ x0 = k[i]; x1 = k[i+1]; x2 = k[i+2]; x3 = k[i+3]; })
L
Linus Torvalds 已提交
38

39 40
#define storekeys(x0, x1, x2, x3, i) \
	({ k[i] = x0; k[i+1] = x1; k[i+2] = x2; k[i+3] = x3; })
L
Linus Torvalds 已提交
41

42 43 44 45
#define store_and_load_keys(x0, x1, x2, x3, s, l) \
	({ storekeys(x0, x1, x2, x3, s); loadkeys(x0, x1, x2, x3, l); })

#define K(x0, x1, x2, x3, i) ({				\
L
Linus Torvalds 已提交
46
	x3 ^= k[4*(i)+3];        x2 ^= k[4*(i)+2];	\
47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72
	x1 ^= k[4*(i)+1];        x0 ^= k[4*(i)+0];	\
	})

#define LK(x0, x1, x2, x3, x4, i) ({					   \
							x0 = rol32(x0, 13);\
	x2 = rol32(x2, 3);	x1 ^= x0;		x4  = x0 << 3;	   \
	x3 ^= x2;		x1 ^= x2;				   \
	x1 = rol32(x1, 1);	x3 ^= x4;				   \
	x3 = rol32(x3, 7);	x4  = x1;				   \
	x0 ^= x1;		x4 <<= 7;		x2 ^= x3;	   \
	x0 ^= x3;		x2 ^= x4;		x3 ^= k[4*i+3];	   \
	x1 ^= k[4*i+1];		x0 = rol32(x0, 5);	x2 = rol32(x2, 22);\
	x0 ^= k[4*i+0];		x2 ^= k[4*i+2];				   \
	})

#define KL(x0, x1, x2, x3, x4, i) ({					   \
	x0 ^= k[4*i+0];		x1 ^= k[4*i+1];		x2 ^= k[4*i+2];	   \
	x3 ^= k[4*i+3];		x0 = ror32(x0, 5);	x2 = ror32(x2, 22);\
	x4 =  x1;		x2 ^= x3;		x0 ^= x3;	   \
	x4 <<= 7;		x0 ^= x1;		x1 = ror32(x1, 1); \
	x2 ^= x4;		x3 = ror32(x3, 7);	x4 = x0 << 3;	   \
	x1 ^= x0;		x3 ^= x4;		x0 = ror32(x0, 13);\
	x1 ^= x2;		x3 ^= x2;		x2 = ror32(x2, 3); \
	})

#define S0(x0, x1, x2, x3, x4) ({			\
L
Linus Torvalds 已提交
73 74
					x4  = x3;	\
	x3 |= x0;	x0 ^= x4;	x4 ^= x2;	\
75
	x4 = ~x4;	x3 ^= x1;	x1 &= x0;	\
L
Linus Torvalds 已提交
76 77
	x1 ^= x4;	x2 ^= x0;	x0 ^= x3;	\
	x4 |= x0;	x0 ^= x2;	x2 &= x1;	\
78 79 80
	x3 ^= x2;	x1 = ~x1;	x2 ^= x4;	\
	x1 ^= x2;					\
	})
L
Linus Torvalds 已提交
81

82
#define S1(x0, x1, x2, x3, x4) ({			\
L
Linus Torvalds 已提交
83
					x4  = x1;	\
84
	x1 ^= x0;	x0 ^= x3;	x3 = ~x3;	\
L
Linus Torvalds 已提交
85 86 87
	x4 &= x1;	x0 |= x1;	x3 ^= x2;	\
	x0 ^= x3;	x1 ^= x3;	x3 ^= x4;	\
	x1 |= x4;	x4 ^= x2;	x2 &= x0;	\
88 89 90
	x2 ^= x1;	x1 |= x0;	x0 = ~x0;	\
	x0 ^= x2;	x4 ^= x1;			\
	})
L
Linus Torvalds 已提交
91

92 93
#define S2(x0, x1, x2, x3, x4) ({			\
					x3 = ~x3;	\
L
Linus Torvalds 已提交
94 95 96
	x1 ^= x0;	x4  = x0;	x0 &= x2;	\
	x0 ^= x3;	x3 |= x4;	x2 ^= x1;	\
	x3 ^= x1;	x1 &= x0;	x0 ^= x2;	\
97
	x2 &= x3;	x3 |= x1;	x0 = ~x0;	\
L
Linus Torvalds 已提交
98
	x3 ^= x0;	x4 ^= x0;	x0 ^= x2;	\
99 100
	x1 |= x2;					\
	})
L
Linus Torvalds 已提交
101

102
#define S3(x0, x1, x2, x3, x4) ({			\
L
Linus Torvalds 已提交
103 104 105 106 107 108
					x4  = x1;	\
	x1 ^= x3;	x3 |= x0;	x4 &= x0;	\
	x0 ^= x2;	x2 ^= x1;	x1 &= x3;	\
	x2 ^= x3;	x0 |= x4;	x4 ^= x3;	\
	x1 ^= x0;	x0 &= x3;	x3 &= x4;	\
	x3 ^= x2;	x4 |= x1;	x2 &= x1;	\
109 110
	x4 ^= x3;	x0 ^= x3;	x3 ^= x2;	\
	})
L
Linus Torvalds 已提交
111

112
#define S4(x0, x1, x2, x3, x4) ({			\
L
Linus Torvalds 已提交
113 114 115 116 117 118
					x4  = x3;	\
	x3 &= x0;	x0 ^= x4;			\
	x3 ^= x2;	x2 |= x4;	x0 ^= x1;	\
	x4 ^= x3;	x2 |= x0;			\
	x2 ^= x1;	x1 &= x0;			\
	x1 ^= x4;	x4 &= x2;	x2 ^= x3;	\
119 120 121
	x4 ^= x0;	x3 |= x1;	x1 = ~x1;	\
	x3 ^= x0;					\
	})
L
Linus Torvalds 已提交
122

123
#define S5(x0, x1, x2, x3, x4) ({			\
L
Linus Torvalds 已提交
124
	x4  = x1;	x1 |= x0;			\
125
	x2 ^= x1;	x3 = ~x3;	x4 ^= x0;	\
L
Linus Torvalds 已提交
126 127 128 129
	x0 ^= x2;	x1 &= x4;	x4 |= x3;	\
	x4 ^= x0;	x0 &= x3;	x1 ^= x3;	\
	x3 ^= x2;	x0 ^= x1;	x2 &= x4;	\
	x1 ^= x2;	x2 &= x0;			\
130 131
	x3 ^= x2;					\
	})
L
Linus Torvalds 已提交
132

133
#define S6(x0, x1, x2, x3, x4) ({			\
L
Linus Torvalds 已提交
134 135
					x4  = x1;	\
	x3 ^= x0;	x1 ^= x2;	x2 ^= x0;	\
136
	x0 &= x3;	x1 |= x3;	x4 = ~x4;	\
L
Linus Torvalds 已提交
137 138 139
	x0 ^= x1;	x1 ^= x2;			\
	x3 ^= x4;	x4 ^= x0;	x2 &= x0;	\
	x4 ^= x1;	x2 ^= x3;	x3 &= x1;	\
140 141
	x3 ^= x0;	x1 ^= x2;			\
	})
L
Linus Torvalds 已提交
142

143 144 145
#define S7(x0, x1, x2, x3, x4) ({			\
					x1 = ~x1;	\
	x4  = x1;	x0 = ~x0;	x1 &= x2;	\
L
Linus Torvalds 已提交
146 147 148 149 150
	x1 ^= x3;	x3 |= x4;	x4 ^= x2;	\
	x2 ^= x3;	x3 ^= x0;	x0 |= x1;	\
	x2 &= x0;	x0 ^= x4;	x4 ^= x3;	\
	x3 &= x0;	x4 ^= x1;			\
	x2 ^= x4;	x3 ^= x1;	x4 |= x0;	\
151 152
	x4 ^= x1;					\
	})
L
Linus Torvalds 已提交
153

154
#define SI0(x0, x1, x2, x3, x4) ({			\
L
Linus Torvalds 已提交
155
			x4  = x3;	x1 ^= x0;	\
156
	x3 |= x1;	x4 ^= x1;	x0 = ~x0;	\
L
Linus Torvalds 已提交
157 158 159
	x2 ^= x3;	x3 ^= x0;	x0 &= x1;	\
	x0 ^= x2;	x2 &= x3;	x3 ^= x4;	\
	x2 ^= x3;	x1 ^= x3;	x3 &= x0;	\
160 161
	x1 ^= x0;	x0 ^= x2;	x4 ^= x3;	\
	})
L
Linus Torvalds 已提交
162

163
#define SI1(x0, x1, x2, x3, x4) ({			\
L
Linus Torvalds 已提交
164
	x1 ^= x3;	x4  = x0;			\
165
	x0 ^= x2;	x2 = ~x2;	x4 |= x1;	\
L
Linus Torvalds 已提交
166 167 168 169
	x4 ^= x3;	x3 &= x1;	x1 ^= x2;	\
	x2 &= x4;	x4 ^= x1;	x1 |= x3;	\
	x3 ^= x0;	x2 ^= x0;	x0 |= x4;	\
	x2 ^= x4;	x1 ^= x0;			\
170 171
	x4 ^= x1;					\
	})
L
Linus Torvalds 已提交
172

173 174
#define SI2(x0, x1, x2, x3, x4) ({			\
	x2 ^= x1;	x4  = x3;	x3 = ~x3;	\
L
Linus Torvalds 已提交
175 176 177 178
	x3 |= x2;	x2 ^= x4;	x4 ^= x0;	\
	x3 ^= x1;	x1 |= x2;	x2 ^= x0;	\
	x1 ^= x4;	x4 |= x3;	x2 ^= x3;	\
	x4 ^= x2;	x2 &= x1;			\
179 180
	x2 ^= x3;	x3 ^= x4;	x4 ^= x0;	\
	})
L
Linus Torvalds 已提交
181

182
#define SI3(x0, x1, x2, x3, x4) ({			\
L
Linus Torvalds 已提交
183 184 185 186 187 188
					x2 ^= x1;	\
	x4  = x1;	x1 &= x2;			\
	x1 ^= x0;	x0 |= x4;	x4 ^= x3;	\
	x0 ^= x3;	x3 |= x1;	x1 ^= x2;	\
	x1 ^= x3;	x0 ^= x2;	x2 ^= x3;	\
	x3 &= x1;	x1 ^= x0;	x0 &= x2;	\
189 190
	x4 ^= x3;	x3 ^= x0;	x0 ^= x1;	\
	})
L
Linus Torvalds 已提交
191

192
#define SI4(x0, x1, x2, x3, x4) ({			\
L
Linus Torvalds 已提交
193
	x2 ^= x3;	x4  = x0;	x0 &= x1;	\
194
	x0 ^= x2;	x2 |= x3;	x4 = ~x4;	\
L
Linus Torvalds 已提交
195 196 197 198
	x1 ^= x0;	x0 ^= x2;	x2 &= x4;	\
	x2 ^= x0;	x0 |= x4;			\
	x0 ^= x3;	x3 &= x2;			\
	x4 ^= x3;	x3 ^= x1;	x1 &= x0;	\
199 200
	x4 ^= x1;	x0 ^= x3;			\
	})
L
Linus Torvalds 已提交
201

202
#define SI5(x0, x1, x2, x3, x4) ({			\
L
Linus Torvalds 已提交
203 204
			x4  = x1;	x1 |= x2;	\
	x2 ^= x4;	x1 ^= x3;	x3 &= x4;	\
205
	x2 ^= x3;	x3 |= x0;	x0 = ~x0;	\
L
Linus Torvalds 已提交
206 207 208
	x3 ^= x2;	x2 |= x0;	x4 ^= x1;	\
	x2 ^= x4;	x4 &= x0;	x0 ^= x1;	\
	x1 ^= x3;	x0 &= x2;	x2 ^= x3;	\
209 210
	x0 ^= x2;	x2 ^= x4;	x4 ^= x3;	\
	})
L
Linus Torvalds 已提交
211

212
#define SI6(x0, x1, x2, x3, x4) ({			\
L
Linus Torvalds 已提交
213 214 215
			x0 ^= x2;			\
	x4  = x0;	x0 &= x3;	x2 ^= x3;	\
	x0 ^= x2;	x3 ^= x1;	x2 |= x4;	\
216
	x2 ^= x3;	x3 &= x0;	x0 = ~x0;	\
L
Linus Torvalds 已提交
217 218
	x3 ^= x1;	x1 &= x2;	x4 ^= x0;	\
	x3 ^= x4;	x4 ^= x2;	x0 ^= x1;	\
219 220
	x2 ^= x0;					\
	})
L
Linus Torvalds 已提交
221

222
#define SI7(x0, x1, x2, x3, x4) ({			\
L
Linus Torvalds 已提交
223
	x4  = x3;	x3 &= x0;	x0 ^= x2;	\
224
	x2 |= x4;	x4 ^= x1;	x0 = ~x0;	\
L
Linus Torvalds 已提交
225 226 227 228
	x1 |= x3;	x4 ^= x0;	x0 &= x2;	\
	x0 ^= x1;	x1 &= x2;	x3 ^= x2;	\
	x4 ^= x3;	x2 &= x3;	x3 |= x0;	\
	x1 ^= x4;	x3 ^= x4;	x4 &= x0;	\
229 230
	x4 ^= x2;					\
	})
L
Linus Torvalds 已提交
231

232 233
int __serpent_setkey(struct serpent_ctx *ctx, const u8 *key,
		     unsigned int keylen)
L
Linus Torvalds 已提交
234
{
235
	u32 *k = ctx->expkey;
L
Linus Torvalds 已提交
236
	u8  *k8 = (u8 *)k;
237
	u32 r0, r1, r2, r3, r4;
L
Linus Torvalds 已提交
238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256
	int i;

	/* Copy key, add padding */

	for (i = 0; i < keylen; ++i)
		k8[i] = key[i];
	if (i < SERPENT_MAX_KEY_SIZE)
		k8[i++] = 1;
	while (i < SERPENT_MAX_KEY_SIZE)
		k8[i++] = 0;

	/* Expand key using polynomial */

	r0 = le32_to_cpu(k[3]);
	r1 = le32_to_cpu(k[4]);
	r2 = le32_to_cpu(k[5]);
	r3 = le32_to_cpu(k[6]);
	r4 = le32_to_cpu(k[7]);

257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289
	keyiter(le32_to_cpu(k[0]), r0, r4, r2, 0, 0);
	keyiter(le32_to_cpu(k[1]), r1, r0, r3, 1, 1);
	keyiter(le32_to_cpu(k[2]), r2, r1, r4, 2, 2);
	keyiter(le32_to_cpu(k[3]), r3, r2, r0, 3, 3);
	keyiter(le32_to_cpu(k[4]), r4, r3, r1, 4, 4);
	keyiter(le32_to_cpu(k[5]), r0, r4, r2, 5, 5);
	keyiter(le32_to_cpu(k[6]), r1, r0, r3, 6, 6);
	keyiter(le32_to_cpu(k[7]), r2, r1, r4, 7, 7);

	keyiter(k[0], r3, r2, r0, 8, 8);
	keyiter(k[1], r4, r3, r1, 9, 9);
	keyiter(k[2], r0, r4, r2, 10, 10);
	keyiter(k[3], r1, r0, r3, 11, 11);
	keyiter(k[4], r2, r1, r4, 12, 12);
	keyiter(k[5], r3, r2, r0, 13, 13);
	keyiter(k[6], r4, r3, r1, 14, 14);
	keyiter(k[7], r0, r4, r2, 15, 15);
	keyiter(k[8], r1, r0, r3, 16, 16);
	keyiter(k[9], r2, r1, r4, 17, 17);
	keyiter(k[10], r3, r2, r0, 18, 18);
	keyiter(k[11], r4, r3, r1, 19, 19);
	keyiter(k[12], r0, r4, r2, 20, 20);
	keyiter(k[13], r1, r0, r3, 21, 21);
	keyiter(k[14], r2, r1, r4, 22, 22);
	keyiter(k[15], r3, r2, r0, 23, 23);
	keyiter(k[16], r4, r3, r1, 24, 24);
	keyiter(k[17], r0, r4, r2, 25, 25);
	keyiter(k[18], r1, r0, r3, 26, 26);
	keyiter(k[19], r2, r1, r4, 27, 27);
	keyiter(k[20], r3, r2, r0, 28, 28);
	keyiter(k[21], r4, r3, r1, 29, 29);
	keyiter(k[22], r0, r4, r2, 30, 30);
	keyiter(k[23], r1, r0, r3, 31, 31);
L
Linus Torvalds 已提交
290 291 292

	k += 50;

293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342
	keyiter(k[-26], r2, r1, r4, 32, -18);
	keyiter(k[-25], r3, r2, r0, 33, -17);
	keyiter(k[-24], r4, r3, r1, 34, -16);
	keyiter(k[-23], r0, r4, r2, 35, -15);
	keyiter(k[-22], r1, r0, r3, 36, -14);
	keyiter(k[-21], r2, r1, r4, 37, -13);
	keyiter(k[-20], r3, r2, r0, 38, -12);
	keyiter(k[-19], r4, r3, r1, 39, -11);
	keyiter(k[-18], r0, r4, r2, 40, -10);
	keyiter(k[-17], r1, r0, r3, 41, -9);
	keyiter(k[-16], r2, r1, r4, 42, -8);
	keyiter(k[-15], r3, r2, r0, 43, -7);
	keyiter(k[-14], r4, r3, r1, 44, -6);
	keyiter(k[-13], r0, r4, r2, 45, -5);
	keyiter(k[-12], r1, r0, r3, 46, -4);
	keyiter(k[-11], r2, r1, r4, 47, -3);
	keyiter(k[-10], r3, r2, r0, 48, -2);
	keyiter(k[-9], r4, r3, r1, 49, -1);
	keyiter(k[-8], r0, r4, r2, 50, 0);
	keyiter(k[-7], r1, r0, r3, 51, 1);
	keyiter(k[-6], r2, r1, r4, 52, 2);
	keyiter(k[-5], r3, r2, r0, 53, 3);
	keyiter(k[-4], r4, r3, r1, 54, 4);
	keyiter(k[-3], r0, r4, r2, 55, 5);
	keyiter(k[-2], r1, r0, r3, 56, 6);
	keyiter(k[-1], r2, r1, r4, 57, 7);
	keyiter(k[0], r3, r2, r0, 58, 8);
	keyiter(k[1], r4, r3, r1, 59, 9);
	keyiter(k[2], r0, r4, r2, 60, 10);
	keyiter(k[3], r1, r0, r3, 61, 11);
	keyiter(k[4], r2, r1, r4, 62, 12);
	keyiter(k[5], r3, r2, r0, 63, 13);
	keyiter(k[6], r4, r3, r1, 64, 14);
	keyiter(k[7], r0, r4, r2, 65, 15);
	keyiter(k[8], r1, r0, r3, 66, 16);
	keyiter(k[9], r2, r1, r4, 67, 17);
	keyiter(k[10], r3, r2, r0, 68, 18);
	keyiter(k[11], r4, r3, r1, 69, 19);
	keyiter(k[12], r0, r4, r2, 70, 20);
	keyiter(k[13], r1, r0, r3, 71, 21);
	keyiter(k[14], r2, r1, r4, 72, 22);
	keyiter(k[15], r3, r2, r0, 73, 23);
	keyiter(k[16], r4, r3, r1, 74, 24);
	keyiter(k[17], r0, r4, r2, 75, 25);
	keyiter(k[18], r1, r0, r3, 76, 26);
	keyiter(k[19], r2, r1, r4, 77, 27);
	keyiter(k[20], r3, r2, r0, 78, 28);
	keyiter(k[21], r4, r3, r1, 79, 29);
	keyiter(k[22], r0, r4, r2, 80, 30);
	keyiter(k[23], r1, r0, r3, 81, 31);
L
Linus Torvalds 已提交
343 344 345

	k += 50;

346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395
	keyiter(k[-26], r2, r1, r4, 82, -18);
	keyiter(k[-25], r3, r2, r0, 83, -17);
	keyiter(k[-24], r4, r3, r1, 84, -16);
	keyiter(k[-23], r0, r4, r2, 85, -15);
	keyiter(k[-22], r1, r0, r3, 86, -14);
	keyiter(k[-21], r2, r1, r4, 87, -13);
	keyiter(k[-20], r3, r2, r0, 88, -12);
	keyiter(k[-19], r4, r3, r1, 89, -11);
	keyiter(k[-18], r0, r4, r2, 90, -10);
	keyiter(k[-17], r1, r0, r3, 91, -9);
	keyiter(k[-16], r2, r1, r4, 92, -8);
	keyiter(k[-15], r3, r2, r0, 93, -7);
	keyiter(k[-14], r4, r3, r1, 94, -6);
	keyiter(k[-13], r0, r4, r2, 95, -5);
	keyiter(k[-12], r1, r0, r3, 96, -4);
	keyiter(k[-11], r2, r1, r4, 97, -3);
	keyiter(k[-10], r3, r2, r0, 98, -2);
	keyiter(k[-9], r4, r3, r1, 99, -1);
	keyiter(k[-8], r0, r4, r2, 100, 0);
	keyiter(k[-7], r1, r0, r3, 101, 1);
	keyiter(k[-6], r2, r1, r4, 102, 2);
	keyiter(k[-5], r3, r2, r0, 103, 3);
	keyiter(k[-4], r4, r3, r1, 104, 4);
	keyiter(k[-3], r0, r4, r2, 105, 5);
	keyiter(k[-2], r1, r0, r3, 106, 6);
	keyiter(k[-1], r2, r1, r4, 107, 7);
	keyiter(k[0], r3, r2, r0, 108, 8);
	keyiter(k[1], r4, r3, r1, 109, 9);
	keyiter(k[2], r0, r4, r2, 110, 10);
	keyiter(k[3], r1, r0, r3, 111, 11);
	keyiter(k[4], r2, r1, r4, 112, 12);
	keyiter(k[5], r3, r2, r0, 113, 13);
	keyiter(k[6], r4, r3, r1, 114, 14);
	keyiter(k[7], r0, r4, r2, 115, 15);
	keyiter(k[8], r1, r0, r3, 116, 16);
	keyiter(k[9], r2, r1, r4, 117, 17);
	keyiter(k[10], r3, r2, r0, 118, 18);
	keyiter(k[11], r4, r3, r1, 119, 19);
	keyiter(k[12], r0, r4, r2, 120, 20);
	keyiter(k[13], r1, r0, r3, 121, 21);
	keyiter(k[14], r2, r1, r4, 122, 22);
	keyiter(k[15], r3, r2, r0, 123, 23);
	keyiter(k[16], r4, r3, r1, 124, 24);
	keyiter(k[17], r0, r4, r2, 125, 25);
	keyiter(k[18], r1, r0, r3, 126, 26);
	keyiter(k[19], r2, r1, r4, 127, 27);
	keyiter(k[20], r3, r2, r0, 128, 28);
	keyiter(k[21], r4, r3, r1, 129, 29);
	keyiter(k[22], r0, r4, r2, 130, 30);
	keyiter(k[23], r1, r0, r3, 131, 31);
L
Linus Torvalds 已提交
396 397 398

	/* Apply S-boxes */

399 400 401 402 403 404 405 406 407 408 409 410 411 412
	S3(r3, r4, r0, r1, r2); store_and_load_keys(r1, r2, r4, r3, 28, 24);
	S4(r1, r2, r4, r3, r0); store_and_load_keys(r2, r4, r3, r0, 24, 20);
	S5(r2, r4, r3, r0, r1); store_and_load_keys(r1, r2, r4, r0, 20, 16);
	S6(r1, r2, r4, r0, r3); store_and_load_keys(r4, r3, r2, r0, 16, 12);
	S7(r4, r3, r2, r0, r1); store_and_load_keys(r1, r2, r0, r4, 12, 8);
	S0(r1, r2, r0, r4, r3); store_and_load_keys(r0, r2, r4, r1, 8, 4);
	S1(r0, r2, r4, r1, r3); store_and_load_keys(r3, r4, r1, r0, 4, 0);
	S2(r3, r4, r1, r0, r2); store_and_load_keys(r2, r4, r3, r0, 0, -4);
	S3(r2, r4, r3, r0, r1); store_and_load_keys(r0, r1, r4, r2, -4, -8);
	S4(r0, r1, r4, r2, r3); store_and_load_keys(r1, r4, r2, r3, -8, -12);
	S5(r1, r4, r2, r3, r0); store_and_load_keys(r0, r1, r4, r3, -12, -16);
	S6(r0, r1, r4, r3, r2); store_and_load_keys(r4, r2, r1, r3, -16, -20);
	S7(r4, r2, r1, r3, r0); store_and_load_keys(r0, r1, r3, r4, -20, -24);
	S0(r0, r1, r3, r4, r2); store_and_load_keys(r3, r1, r4, r0, -24, -28);
L
Linus Torvalds 已提交
413
	k -= 50;
414 415 416 417 418 419 420 421 422 423 424
	S1(r3, r1, r4, r0, r2); store_and_load_keys(r2, r4, r0, r3, 22, 18);
	S2(r2, r4, r0, r3, r1); store_and_load_keys(r1, r4, r2, r3, 18, 14);
	S3(r1, r4, r2, r3, r0); store_and_load_keys(r3, r0, r4, r1, 14, 10);
	S4(r3, r0, r4, r1, r2); store_and_load_keys(r0, r4, r1, r2, 10, 6);
	S5(r0, r4, r1, r2, r3); store_and_load_keys(r3, r0, r4, r2, 6, 2);
	S6(r3, r0, r4, r2, r1); store_and_load_keys(r4, r1, r0, r2, 2, -2);
	S7(r4, r1, r0, r2, r3); store_and_load_keys(r3, r0, r2, r4, -2, -6);
	S0(r3, r0, r2, r4, r1); store_and_load_keys(r2, r0, r4, r3, -6, -10);
	S1(r2, r0, r4, r3, r1); store_and_load_keys(r1, r4, r3, r2, -10, -14);
	S2(r1, r4, r3, r2, r0); store_and_load_keys(r0, r4, r1, r2, -14, -18);
	S3(r0, r4, r1, r2, r3); store_and_load_keys(r2, r3, r4, r0, -18, -22);
L
Linus Torvalds 已提交
425
	k -= 50;
426 427 428 429 430 431 432 433
	S4(r2, r3, r4, r0, r1); store_and_load_keys(r3, r4, r0, r1, 28, 24);
	S5(r3, r4, r0, r1, r2); store_and_load_keys(r2, r3, r4, r1, 24, 20);
	S6(r2, r3, r4, r1, r0); store_and_load_keys(r4, r0, r3, r1, 20, 16);
	S7(r4, r0, r3, r1, r2); store_and_load_keys(r2, r3, r1, r4, 16, 12);
	S0(r2, r3, r1, r4, r0); store_and_load_keys(r1, r3, r4, r2, 12, 8);
	S1(r1, r3, r4, r2, r0); store_and_load_keys(r0, r4, r2, r1, 8, 4);
	S2(r0, r4, r2, r1, r3); store_and_load_keys(r3, r4, r0, r1, 4, 0);
	S3(r3, r4, r0, r1, r2); storekeys(r1, r2, r4, r3, 0);
L
Linus Torvalds 已提交
434 435 436

	return 0;
}
437 438 439 440 441 442
EXPORT_SYMBOL_GPL(__serpent_setkey);

int serpent_setkey(struct crypto_tfm *tfm, const u8 *key, unsigned int keylen)
{
	return __serpent_setkey(crypto_tfm_ctx(tfm), key, keylen);
}
443
EXPORT_SYMBOL_GPL(serpent_setkey);
L
Linus Torvalds 已提交
444

445
void __serpent_encrypt(struct serpent_ctx *ctx, u8 *dst, const u8 *src)
L
Linus Torvalds 已提交
446
{
447
	const u32 *k = ctx->expkey;
448 449 450
	const __le32 *s = (const __le32 *)src;
	__le32	*d = (__le32 *)dst;
	u32	r0, r1, r2, r3, r4;
L
Linus Torvalds 已提交
451 452 453 454 455 456 457 458 459 460 461

/*
 * Note: The conversions between u8* and u32* might cause trouble
 * on architectures with stricter alignment rules than x86
 */

	r0 = le32_to_cpu(s[0]);
	r1 = le32_to_cpu(s[1]);
	r2 = le32_to_cpu(s[2]);
	r3 = le32_to_cpu(s[3]);

462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494
					K(r0, r1, r2, r3, 0);
	S0(r0, r1, r2, r3, r4);		LK(r2, r1, r3, r0, r4, 1);
	S1(r2, r1, r3, r0, r4);		LK(r4, r3, r0, r2, r1, 2);
	S2(r4, r3, r0, r2, r1);		LK(r1, r3, r4, r2, r0, 3);
	S3(r1, r3, r4, r2, r0);		LK(r2, r0, r3, r1, r4, 4);
	S4(r2, r0, r3, r1, r4);		LK(r0, r3, r1, r4, r2, 5);
	S5(r0, r3, r1, r4, r2);		LK(r2, r0, r3, r4, r1, 6);
	S6(r2, r0, r3, r4, r1);		LK(r3, r1, r0, r4, r2, 7);
	S7(r3, r1, r0, r4, r2);		LK(r2, r0, r4, r3, r1, 8);
	S0(r2, r0, r4, r3, r1);		LK(r4, r0, r3, r2, r1, 9);
	S1(r4, r0, r3, r2, r1);		LK(r1, r3, r2, r4, r0, 10);
	S2(r1, r3, r2, r4, r0);		LK(r0, r3, r1, r4, r2, 11);
	S3(r0, r3, r1, r4, r2);		LK(r4, r2, r3, r0, r1, 12);
	S4(r4, r2, r3, r0, r1);		LK(r2, r3, r0, r1, r4, 13);
	S5(r2, r3, r0, r1, r4);		LK(r4, r2, r3, r1, r0, 14);
	S6(r4, r2, r3, r1, r0);		LK(r3, r0, r2, r1, r4, 15);
	S7(r3, r0, r2, r1, r4);		LK(r4, r2, r1, r3, r0, 16);
	S0(r4, r2, r1, r3, r0);		LK(r1, r2, r3, r4, r0, 17);
	S1(r1, r2, r3, r4, r0);		LK(r0, r3, r4, r1, r2, 18);
	S2(r0, r3, r4, r1, r2);		LK(r2, r3, r0, r1, r4, 19);
	S3(r2, r3, r0, r1, r4);		LK(r1, r4, r3, r2, r0, 20);
	S4(r1, r4, r3, r2, r0);		LK(r4, r3, r2, r0, r1, 21);
	S5(r4, r3, r2, r0, r1);		LK(r1, r4, r3, r0, r2, 22);
	S6(r1, r4, r3, r0, r2);		LK(r3, r2, r4, r0, r1, 23);
	S7(r3, r2, r4, r0, r1);		LK(r1, r4, r0, r3, r2, 24);
	S0(r1, r4, r0, r3, r2);		LK(r0, r4, r3, r1, r2, 25);
	S1(r0, r4, r3, r1, r2);		LK(r2, r3, r1, r0, r4, 26);
	S2(r2, r3, r1, r0, r4);		LK(r4, r3, r2, r0, r1, 27);
	S3(r4, r3, r2, r0, r1);		LK(r0, r1, r3, r4, r2, 28);
	S4(r0, r1, r3, r4, r2);		LK(r1, r3, r4, r2, r0, 29);
	S5(r1, r3, r4, r2, r0);		LK(r0, r1, r3, r2, r4, 30);
	S6(r0, r1, r3, r2, r4);		LK(r3, r4, r1, r2, r0, 31);
	S7(r3, r4, r1, r2, r0);		K(r0, r1, r2, r3, 32);
L
Linus Torvalds 已提交
495 496 497 498 499 500

	d[0] = cpu_to_le32(r0);
	d[1] = cpu_to_le32(r1);
	d[2] = cpu_to_le32(r2);
	d[3] = cpu_to_le32(r3);
}
501
EXPORT_SYMBOL_GPL(__serpent_encrypt);
L
Linus Torvalds 已提交
502

503
static void serpent_encrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
L
Linus Torvalds 已提交
504
{
505
	struct serpent_ctx *ctx = crypto_tfm_ctx(tfm);
506 507 508 509 510 511 512

	__serpent_encrypt(ctx, dst, src);
}

void __serpent_decrypt(struct serpent_ctx *ctx, u8 *dst, const u8 *src)
{
	const u32 *k = ctx->expkey;
513 514 515
	const __le32 *s = (const __le32 *)src;
	__le32	*d = (__le32 *)dst;
	u32	r0, r1, r2, r3, r4;
L
Linus Torvalds 已提交
516 517 518 519 520 521

	r0 = le32_to_cpu(s[0]);
	r1 = le32_to_cpu(s[1]);
	r2 = le32_to_cpu(s[2]);
	r3 = le32_to_cpu(s[3]);

522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554
					K(r0, r1, r2, r3, 32);
	SI7(r0, r1, r2, r3, r4);	KL(r1, r3, r0, r4, r2, 31);
	SI6(r1, r3, r0, r4, r2);	KL(r0, r2, r4, r1, r3, 30);
	SI5(r0, r2, r4, r1, r3);	KL(r2, r3, r0, r4, r1, 29);
	SI4(r2, r3, r0, r4, r1);	KL(r2, r0, r1, r4, r3, 28);
	SI3(r2, r0, r1, r4, r3);	KL(r1, r2, r3, r4, r0, 27);
	SI2(r1, r2, r3, r4, r0);	KL(r2, r0, r4, r3, r1, 26);
	SI1(r2, r0, r4, r3, r1);	KL(r1, r0, r4, r3, r2, 25);
	SI0(r1, r0, r4, r3, r2);	KL(r4, r2, r0, r1, r3, 24);
	SI7(r4, r2, r0, r1, r3);	KL(r2, r1, r4, r3, r0, 23);
	SI6(r2, r1, r4, r3, r0);	KL(r4, r0, r3, r2, r1, 22);
	SI5(r4, r0, r3, r2, r1);	KL(r0, r1, r4, r3, r2, 21);
	SI4(r0, r1, r4, r3, r2);	KL(r0, r4, r2, r3, r1, 20);
	SI3(r0, r4, r2, r3, r1);	KL(r2, r0, r1, r3, r4, 19);
	SI2(r2, r0, r1, r3, r4);	KL(r0, r4, r3, r1, r2, 18);
	SI1(r0, r4, r3, r1, r2);	KL(r2, r4, r3, r1, r0, 17);
	SI0(r2, r4, r3, r1, r0);	KL(r3, r0, r4, r2, r1, 16);
	SI7(r3, r0, r4, r2, r1);	KL(r0, r2, r3, r1, r4, 15);
	SI6(r0, r2, r3, r1, r4);	KL(r3, r4, r1, r0, r2, 14);
	SI5(r3, r4, r1, r0, r2);	KL(r4, r2, r3, r1, r0, 13);
	SI4(r4, r2, r3, r1, r0);	KL(r4, r3, r0, r1, r2, 12);
	SI3(r4, r3, r0, r1, r2);	KL(r0, r4, r2, r1, r3, 11);
	SI2(r0, r4, r2, r1, r3);	KL(r4, r3, r1, r2, r0, 10);
	SI1(r4, r3, r1, r2, r0);	KL(r0, r3, r1, r2, r4, 9);
	SI0(r0, r3, r1, r2, r4);	KL(r1, r4, r3, r0, r2, 8);
	SI7(r1, r4, r3, r0, r2);	KL(r4, r0, r1, r2, r3, 7);
	SI6(r4, r0, r1, r2, r3);	KL(r1, r3, r2, r4, r0, 6);
	SI5(r1, r3, r2, r4, r0);	KL(r3, r0, r1, r2, r4, 5);
	SI4(r3, r0, r1, r2, r4);	KL(r3, r1, r4, r2, r0, 4);
	SI3(r3, r1, r4, r2, r0);	KL(r4, r3, r0, r2, r1, 3);
	SI2(r4, r3, r0, r2, r1);	KL(r3, r1, r2, r0, r4, 2);
	SI1(r3, r1, r2, r0, r4);	KL(r4, r1, r2, r0, r3, 1);
	SI0(r4, r1, r2, r0, r3);	K(r2, r3, r1, r4, 0);
L
Linus Torvalds 已提交
555 556 557 558 559 560

	d[0] = cpu_to_le32(r2);
	d[1] = cpu_to_le32(r3);
	d[2] = cpu_to_le32(r1);
	d[3] = cpu_to_le32(r4);
}
561 562 563 564 565 566 567 568
EXPORT_SYMBOL_GPL(__serpent_decrypt);

static void serpent_decrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
{
	struct serpent_ctx *ctx = crypto_tfm_ctx(tfm);

	__serpent_decrypt(ctx, dst, src);
}
L
Linus Torvalds 已提交
569

570
static int tnepres_setkey(struct crypto_tfm *tfm, const u8 *key,
571
			  unsigned int keylen)
L
Linus Torvalds 已提交
572 573 574 575 576 577
{
	u8 rev_key[SERPENT_MAX_KEY_SIZE];
	int i;

	for (i = 0; i < keylen; ++i)
		rev_key[keylen - i - 1] = key[i];
578

579
	return serpent_setkey(tfm, rev_key, keylen);
L
Linus Torvalds 已提交
580 581
}

582
static void tnepres_encrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
L
Linus Torvalds 已提交
583 584 585 586 587 588 589 590 591 592 593
{
	const u32 * const s = (const u32 * const)src;
	u32 * const d = (u32 * const)dst;

	u32 rs[4], rd[4];

	rs[0] = swab32(s[3]);
	rs[1] = swab32(s[2]);
	rs[2] = swab32(s[1]);
	rs[3] = swab32(s[0]);

594
	serpent_encrypt(tfm, (u8 *)rd, (u8 *)rs);
L
Linus Torvalds 已提交
595 596 597 598 599 600 601

	d[0] = swab32(rd[3]);
	d[1] = swab32(rd[2]);
	d[2] = swab32(rd[1]);
	d[3] = swab32(rd[0]);
}

602
static void tnepres_decrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
L
Linus Torvalds 已提交
603 604 605 606 607 608 609 610 611 612 613
{
	const u32 * const s = (const u32 * const)src;
	u32 * const d = (u32 * const)dst;

	u32 rs[4], rd[4];

	rs[0] = swab32(s[3]);
	rs[1] = swab32(s[2]);
	rs[2] = swab32(s[1]);
	rs[3] = swab32(s[0]);

614
	serpent_decrypt(tfm, (u8 *)rd, (u8 *)rs);
L
Linus Torvalds 已提交
615 616 617 618 619 620 621

	d[0] = swab32(rd[3]);
	d[1] = swab32(rd[2]);
	d[2] = swab32(rd[1]);
	d[3] = swab32(rd[0]);
}

622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637
static struct crypto_alg srp_algs[2] = { {
	.cra_name		=	"serpent",
	.cra_driver_name	=	"serpent-generic",
	.cra_priority		=	100,
	.cra_flags		=	CRYPTO_ALG_TYPE_CIPHER,
	.cra_blocksize		=	SERPENT_BLOCK_SIZE,
	.cra_ctxsize		=	sizeof(struct serpent_ctx),
	.cra_alignmask		=	3,
	.cra_module		=	THIS_MODULE,
	.cra_u			=	{ .cipher = {
	.cia_min_keysize	=	SERPENT_MIN_KEY_SIZE,
	.cia_max_keysize	=	SERPENT_MAX_KEY_SIZE,
	.cia_setkey		=	serpent_setkey,
	.cia_encrypt		=	serpent_encrypt,
	.cia_decrypt		=	serpent_decrypt } }
}, {
L
Linus Torvalds 已提交
638 639 640 641
	.cra_name		=	"tnepres",
	.cra_flags		=	CRYPTO_ALG_TYPE_CIPHER,
	.cra_blocksize		=	SERPENT_BLOCK_SIZE,
	.cra_ctxsize		=	sizeof(struct serpent_ctx),
642
	.cra_alignmask		=	3,
L
Linus Torvalds 已提交
643 644 645 646
	.cra_module		=	THIS_MODULE,
	.cra_u			=	{ .cipher = {
	.cia_min_keysize	=	SERPENT_MIN_KEY_SIZE,
	.cia_max_keysize	=	SERPENT_MAX_KEY_SIZE,
647 648 649
	.cia_setkey		=	tnepres_setkey,
	.cia_encrypt		=	tnepres_encrypt,
	.cia_decrypt		=	tnepres_decrypt } }
650
} };
L
Linus Torvalds 已提交
651

652
static int __init serpent_mod_init(void)
L
Linus Torvalds 已提交
653
{
654
	return crypto_register_algs(srp_algs, ARRAY_SIZE(srp_algs));
L
Linus Torvalds 已提交
655 656
}

657
static void __exit serpent_mod_fini(void)
L
Linus Torvalds 已提交
658
{
659
	crypto_unregister_algs(srp_algs, ARRAY_SIZE(srp_algs));
L
Linus Torvalds 已提交
660 661
}

662 663
module_init(serpent_mod_init);
module_exit(serpent_mod_fini);
L
Linus Torvalds 已提交
664 665 666 667

MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("Serpent and tnepres (kerneli compatible serpent reversed) Cipher Algorithm");
MODULE_AUTHOR("Dag Arne Osvik <osvik@ii.uib.no>");
668 669
MODULE_ALIAS_CRYPTO("tnepres");
MODULE_ALIAS_CRYPTO("serpent");
670
MODULE_ALIAS_CRYPTO("serpent-generic");