xref: /freebsd/sys/crypto/aesni/aesni_wrap.c (revision 59c8e88e72633afbc47a4ace0d2170d00d51f7dc)
1 /*-
2  * Copyright (C) 2008 Damien Miller <djm@mindrot.org>
3  * Copyright (c) 2010 Konstantin Belousov <kib@FreeBSD.org>
4  * Copyright (c) 2010-2011 Pawel Jakub Dawidek <pawel@dawidek.net>
5  * Copyright 2012-2013 John-Mark Gurney <jmg@FreeBSD.org>
6  * Copyright (c) 2014 The FreeBSD Foundation
7  * All rights reserved.
8  *
9  * Portions of this software were developed by John-Mark Gurney
10  * under sponsorship of the FreeBSD Foundation and
11  * Rubicon Communications, LLC (Netgate).
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 #include <sys/param.h>
36 #include <sys/libkern.h>
37 #include <sys/malloc.h>
38 #include <sys/proc.h>
39 #include <sys/systm.h>
40 #include <crypto/aesni/aesni.h>
41 
42 #include <opencrypto/gmac.h>
43 
44 #include "aesencdec.h"
45 #include <smmintrin.h>
46 
47 MALLOC_DECLARE(M_AESNI);
48 
49 struct blocks8 {
50 	__m128i	blk[8];
51 } __packed;
52 
53 void
54 aesni_encrypt_cbc(int rounds, const void *key_schedule, size_t len,
55     const uint8_t *from, uint8_t *to, const uint8_t iv[static AES_BLOCK_LEN])
56 {
57 	__m128i tot, ivreg;
58 	size_t i;
59 
60 	len /= AES_BLOCK_LEN;
61 	ivreg = _mm_loadu_si128((const __m128i *)iv);
62 	for (i = 0; i < len; i++) {
63 		tot = aesni_enc(rounds - 1, key_schedule,
64 		    _mm_loadu_si128((const __m128i *)from) ^ ivreg);
65 		ivreg = tot;
66 		_mm_storeu_si128((__m128i *)to, tot);
67 		from += AES_BLOCK_LEN;
68 		to += AES_BLOCK_LEN;
69 	}
70 }
71 
72 void
73 aesni_decrypt_cbc(int rounds, const void *key_schedule, size_t len,
74     uint8_t *buf, const uint8_t iv[static AES_BLOCK_LEN])
75 {
76 	__m128i blocks[8];
77 	struct blocks8 *blks;
78 	__m128i ivreg, nextiv;
79 	size_t i, j, cnt;
80 
81 	ivreg = _mm_loadu_si128((const __m128i *)iv);
82 	cnt = len / AES_BLOCK_LEN / 8;
83 	for (i = 0; i < cnt; i++) {
84 		blks = (struct blocks8 *)buf;
85 		aesni_dec8(rounds - 1, key_schedule, blks->blk[0], blks->blk[1],
86 		    blks->blk[2], blks->blk[3], blks->blk[4], blks->blk[5],
87 		    blks->blk[6], blks->blk[7], &blocks[0]);
88 		for (j = 0; j < 8; j++) {
89 			nextiv = blks->blk[j];
90 			blks->blk[j] = blocks[j] ^ ivreg;
91 			ivreg = nextiv;
92 		}
93 		buf += AES_BLOCK_LEN * 8;
94 	}
95 	i *= 8;
96 	cnt = len / AES_BLOCK_LEN;
97 	for (; i < cnt; i++) {
98 		nextiv = _mm_loadu_si128((void *)buf);
99 		_mm_storeu_si128((void *)buf,
100 		    aesni_dec(rounds - 1, key_schedule, nextiv) ^ ivreg);
101 		ivreg = nextiv;
102 		buf += AES_BLOCK_LEN;
103 	}
104 }
105 
106 void
107 aesni_encrypt_ecb(int rounds, const void *key_schedule, size_t len,
108     const uint8_t *from, uint8_t *to)
109 {
110 	__m128i tot;
111 	__m128i tout[8];
112 	struct blocks8 *top;
113 	const struct blocks8 *blks;
114 	size_t i, cnt;
115 
116 	cnt = len / AES_BLOCK_LEN / 8;
117 	for (i = 0; i < cnt; i++) {
118 		blks = (const struct blocks8 *)from;
119 		top = (struct blocks8 *)to;
120 		aesni_enc8(rounds - 1, key_schedule, blks->blk[0], blks->blk[1],
121 		    blks->blk[2], blks->blk[3], blks->blk[4], blks->blk[5],
122 		    blks->blk[6], blks->blk[7], tout);
123 		top->blk[0] = tout[0];
124 		top->blk[1] = tout[1];
125 		top->blk[2] = tout[2];
126 		top->blk[3] = tout[3];
127 		top->blk[4] = tout[4];
128 		top->blk[5] = tout[5];
129 		top->blk[6] = tout[6];
130 		top->blk[7] = tout[7];
131 		from += AES_BLOCK_LEN * 8;
132 		to += AES_BLOCK_LEN * 8;
133 	}
134 	i *= 8;
135 	cnt = len / AES_BLOCK_LEN;
136 	for (; i < cnt; i++) {
137 		tot = aesni_enc(rounds - 1, key_schedule,
138 		    _mm_loadu_si128((const __m128i *)from));
139 		_mm_storeu_si128((__m128i *)to, tot);
140 		from += AES_BLOCK_LEN;
141 		to += AES_BLOCK_LEN;
142 	}
143 }
144 
145 void
146 aesni_decrypt_ecb(int rounds, const void *key_schedule, size_t len,
147     const uint8_t *from, uint8_t *to)
148 {
149 	__m128i tot;
150 	__m128i tout[8];
151 	const struct blocks8 *blks;
152 	struct blocks8 *top;
153 	size_t i, cnt;
154 
155 	cnt = len / AES_BLOCK_LEN / 8;
156 	for (i = 0; i < cnt; i++) {
157 		blks = (const struct blocks8 *)from;
158 		top = (struct blocks8 *)to;
159 		aesni_dec8(rounds - 1, key_schedule, blks->blk[0], blks->blk[1],
160 		    blks->blk[2], blks->blk[3], blks->blk[4], blks->blk[5],
161 		    blks->blk[6], blks->blk[7], tout);
162 		top->blk[0] = tout[0];
163 		top->blk[1] = tout[1];
164 		top->blk[2] = tout[2];
165 		top->blk[3] = tout[3];
166 		top->blk[4] = tout[4];
167 		top->blk[5] = tout[5];
168 		top->blk[6] = tout[6];
169 		top->blk[7] = tout[7];
170 		from += AES_BLOCK_LEN * 8;
171 		to += AES_BLOCK_LEN * 8;
172 	}
173 	i *= 8;
174 	cnt = len / AES_BLOCK_LEN;
175 	for (; i < cnt; i++) {
176 		tot = aesni_dec(rounds - 1, key_schedule,
177 		    _mm_loadu_si128((const __m128i *)from));
178 		_mm_storeu_si128((__m128i *)to, tot);
179 		from += AES_BLOCK_LEN;
180 		to += AES_BLOCK_LEN;
181 	}
182 }
183 
184 /*
185  * mixed endian increment, low 64bits stored in hi word to be compatible
186  * with _icm's BSWAP.
187  */
188 static inline __m128i
189 nextc(__m128i x)
190 {
191 	const __m128i ONE = _mm_setr_epi32(0, 0, 1, 0);
192 	const __m128i ZERO = _mm_setzero_si128();
193 
194 	x = _mm_add_epi64(x, ONE);
195 	__m128i t = _mm_cmpeq_epi64(x, ZERO);
196 	t = _mm_unpackhi_epi64(t, ZERO);
197 	x = _mm_sub_epi64(x, t);
198 
199 	return x;
200 }
201 
202 void
203 aesni_encrypt_icm(int rounds, const void *key_schedule, size_t len,
204     const uint8_t *from, uint8_t *to, const uint8_t iv[static AES_BLOCK_LEN])
205 {
206 	__m128i tot;
207 	__m128i tmp1, tmp2, tmp3, tmp4;
208 	__m128i tmp5, tmp6, tmp7, tmp8;
209 	__m128i ctr1, ctr2, ctr3, ctr4;
210 	__m128i ctr5, ctr6, ctr7, ctr8;
211 	__m128i BSWAP_EPI64;
212 	__m128i tout[8];
213 	__m128i block;
214 	struct blocks8 *top;
215 	const struct blocks8 *blks;
216 	size_t i, cnt, resid;
217 
218 	BSWAP_EPI64 = _mm_set_epi8(8,9,10,11,12,13,14,15,0,1,2,3,4,5,6,7);
219 
220 	ctr1 = _mm_loadu_si128((const __m128i *)iv);
221 	ctr1 = _mm_shuffle_epi8(ctr1, BSWAP_EPI64);
222 
223 	cnt = len / AES_BLOCK_LEN / 8;
224 	for (i = 0; i < cnt; i++) {
225 		tmp1 = _mm_shuffle_epi8(ctr1, BSWAP_EPI64);
226 		ctr2 = nextc(ctr1);
227 		tmp2 = _mm_shuffle_epi8(ctr2, BSWAP_EPI64);
228 		ctr3 = nextc(ctr2);
229 		tmp3 = _mm_shuffle_epi8(ctr3, BSWAP_EPI64);
230 		ctr4 = nextc(ctr3);
231 		tmp4 = _mm_shuffle_epi8(ctr4, BSWAP_EPI64);
232 		ctr5 = nextc(ctr4);
233 		tmp5 = _mm_shuffle_epi8(ctr5, BSWAP_EPI64);
234 		ctr6 = nextc(ctr5);
235 		tmp6 = _mm_shuffle_epi8(ctr6, BSWAP_EPI64);
236 		ctr7 = nextc(ctr6);
237 		tmp7 = _mm_shuffle_epi8(ctr7, BSWAP_EPI64);
238 		ctr8 = nextc(ctr7);
239 		tmp8 = _mm_shuffle_epi8(ctr8, BSWAP_EPI64);
240 		ctr1 = nextc(ctr8);
241 
242 		blks = (const struct blocks8 *)from;
243 		top = (struct blocks8 *)to;
244 		aesni_enc8(rounds - 1, key_schedule, tmp1, tmp2, tmp3, tmp4,
245 		    tmp5, tmp6, tmp7, tmp8, tout);
246 
247 		top->blk[0] = blks->blk[0] ^ tout[0];
248 		top->blk[1] = blks->blk[1] ^ tout[1];
249 		top->blk[2] = blks->blk[2] ^ tout[2];
250 		top->blk[3] = blks->blk[3] ^ tout[3];
251 		top->blk[4] = blks->blk[4] ^ tout[4];
252 		top->blk[5] = blks->blk[5] ^ tout[5];
253 		top->blk[6] = blks->blk[6] ^ tout[6];
254 		top->blk[7] = blks->blk[7] ^ tout[7];
255 
256 		from += AES_BLOCK_LEN * 8;
257 		to += AES_BLOCK_LEN * 8;
258 	}
259 	i *= 8;
260 	cnt = len / AES_BLOCK_LEN;
261 	for (; i < cnt; i++) {
262 		tmp1 = _mm_shuffle_epi8(ctr1, BSWAP_EPI64);
263 		ctr1 = nextc(ctr1);
264 
265 		tot = aesni_enc(rounds - 1, key_schedule, tmp1);
266 
267 		tot = tot ^ _mm_loadu_si128((const __m128i *)from);
268 		_mm_storeu_si128((__m128i *)to, tot);
269 
270 		from += AES_BLOCK_LEN;
271 		to += AES_BLOCK_LEN;
272 	}
273 
274 	/*
275 	 * Handle remaining partial round.  Copy the remaining payload onto the
276 	 * stack to ensure that the full block can be loaded safely.
277 	 */
278 	resid = len % AES_BLOCK_LEN;
279 	if (resid != 0) {
280 		tmp1 = _mm_shuffle_epi8(ctr1, BSWAP_EPI64);
281 		tot = aesni_enc(rounds - 1, key_schedule, tmp1);
282 		block = _mm_setzero_si128();
283 		memcpy(&block, from, resid);
284 		tot = tot ^ _mm_loadu_si128(&block);
285 		memcpy(to, &tot, resid);
286 		explicit_bzero(&block, sizeof(block));
287 	}
288 }
289 
290 #define	AES_XTS_BLOCKSIZE	16
291 #define	AES_XTS_IVSIZE		8
292 #define	AES_XTS_ALPHA		0x87	/* GF(2^128) generator polynomial */
293 
294 static inline __m128i
295 xts_crank_lfsr(__m128i inp)
296 {
297 	const __m128i alphamask = _mm_set_epi32(1, 1, 1, AES_XTS_ALPHA);
298 	__m128i xtweak, ret;
299 
300 	/* set up xor mask */
301 	xtweak = _mm_shuffle_epi32(inp, 0x93);
302 	xtweak = _mm_srai_epi32(xtweak, 31);
303 	xtweak &= alphamask;
304 
305 	/* next term */
306 	ret = _mm_slli_epi32(inp, 1);
307 	ret ^= xtweak;
308 
309 	return ret;
310 }
311 
312 static void
313 aesni_crypt_xts_block(int rounds, const __m128i *key_schedule, __m128i *tweak,
314     const uint8_t *from, uint8_t *to, int do_encrypt)
315 {
316 	__m128i block;
317 
318 	block = _mm_loadu_si128((const __m128i *)from) ^ *tweak;
319 
320 	if (do_encrypt)
321 		block = aesni_enc(rounds - 1, key_schedule, block);
322 	else
323 		block = aesni_dec(rounds - 1, key_schedule, block);
324 
325 	_mm_storeu_si128((__m128i *)to, block ^ *tweak);
326 
327 	*tweak = xts_crank_lfsr(*tweak);
328 }
329 
330 static void
331 aesni_crypt_xts_block8(int rounds, const __m128i *key_schedule, __m128i *tweak,
332     const uint8_t *from, uint8_t *to, int do_encrypt)
333 {
334 	__m128i tmptweak;
335 	__m128i a, b, c, d, e, f, g, h;
336 	__m128i tweaks[8];
337 	__m128i tmp[8];
338 	__m128i *top;
339 	const __m128i *fromp;
340 
341 	tmptweak = *tweak;
342 
343 	/*
344 	 * unroll the loop.  This lets gcc put values directly in the
345 	 * register and saves memory accesses.
346 	 */
347 	fromp = (const __m128i *)from;
348 #define PREPINP(v, pos) 					\
349 		do {						\
350 			tweaks[(pos)] = tmptweak;		\
351 			(v) = _mm_loadu_si128(&fromp[pos]) ^	\
352 			    tmptweak;				\
353 			tmptweak = xts_crank_lfsr(tmptweak);	\
354 		} while (0)
355 	PREPINP(a, 0);
356 	PREPINP(b, 1);
357 	PREPINP(c, 2);
358 	PREPINP(d, 3);
359 	PREPINP(e, 4);
360 	PREPINP(f, 5);
361 	PREPINP(g, 6);
362 	PREPINP(h, 7);
363 	*tweak = tmptweak;
364 
365 	if (do_encrypt)
366 		aesni_enc8(rounds - 1, key_schedule, a, b, c, d, e, f, g, h,
367 		    tmp);
368 	else
369 		aesni_dec8(rounds - 1, key_schedule, a, b, c, d, e, f, g, h,
370 		    tmp);
371 
372 	top = (__m128i *)to;
373 	_mm_storeu_si128(&top[0], tmp[0] ^ tweaks[0]);
374 	_mm_storeu_si128(&top[1], tmp[1] ^ tweaks[1]);
375 	_mm_storeu_si128(&top[2], tmp[2] ^ tweaks[2]);
376 	_mm_storeu_si128(&top[3], tmp[3] ^ tweaks[3]);
377 	_mm_storeu_si128(&top[4], tmp[4] ^ tweaks[4]);
378 	_mm_storeu_si128(&top[5], tmp[5] ^ tweaks[5]);
379 	_mm_storeu_si128(&top[6], tmp[6] ^ tweaks[6]);
380 	_mm_storeu_si128(&top[7], tmp[7] ^ tweaks[7]);
381 }
382 
383 static void
384 aesni_crypt_xts(int rounds, const __m128i *data_schedule,
385     const __m128i *tweak_schedule, size_t len, const uint8_t *from,
386     uint8_t *to, const uint8_t iv[static AES_BLOCK_LEN], int do_encrypt)
387 {
388 	__m128i tweakreg;
389 	uint8_t tweak[AES_XTS_BLOCKSIZE] __aligned(16);
390 	size_t i, cnt;
391 
392 	/*
393 	 * Prepare tweak as E_k2(IV). IV is specified as LE representation
394 	 * of a 64-bit block number which we allow to be passed in directly.
395 	 */
396 #if BYTE_ORDER == LITTLE_ENDIAN
397 	bcopy(iv, tweak, AES_XTS_IVSIZE);
398 	/* Last 64 bits of IV are always zero. */
399 	bzero(tweak + AES_XTS_IVSIZE, AES_XTS_IVSIZE);
400 #else
401 #error Only LITTLE_ENDIAN architectures are supported.
402 #endif
403 	tweakreg = _mm_loadu_si128((__m128i *)&tweak[0]);
404 	tweakreg = aesni_enc(rounds - 1, tweak_schedule, tweakreg);
405 
406 	cnt = len / AES_XTS_BLOCKSIZE / 8;
407 	for (i = 0; i < cnt; i++) {
408 		aesni_crypt_xts_block8(rounds, data_schedule, &tweakreg,
409 		    from, to, do_encrypt);
410 		from += AES_XTS_BLOCKSIZE * 8;
411 		to += AES_XTS_BLOCKSIZE * 8;
412 	}
413 	i *= 8;
414 	cnt = len / AES_XTS_BLOCKSIZE;
415 	for (; i < cnt; i++) {
416 		aesni_crypt_xts_block(rounds, data_schedule, &tweakreg,
417 		    from, to, do_encrypt);
418 		from += AES_XTS_BLOCKSIZE;
419 		to += AES_XTS_BLOCKSIZE;
420 	}
421 }
422 
423 void
424 aesni_encrypt_xts(int rounds, const void *data_schedule,
425     const void *tweak_schedule, size_t len, const uint8_t *from, uint8_t *to,
426     const uint8_t iv[static AES_BLOCK_LEN])
427 {
428 
429 	aesni_crypt_xts(rounds, data_schedule, tweak_schedule, len, from, to,
430 	    iv, 1);
431 }
432 
433 void
434 aesni_decrypt_xts(int rounds, const void *data_schedule,
435     const void *tweak_schedule, size_t len, const uint8_t *from, uint8_t *to,
436     const uint8_t iv[static AES_BLOCK_LEN])
437 {
438 
439 	aesni_crypt_xts(rounds, data_schedule, tweak_schedule, len, from, to,
440 	    iv, 0);
441 }
442 
443 void
444 aesni_cipher_setup_common(struct aesni_session *ses,
445     const struct crypto_session_params *csp, const uint8_t *key, int keylen)
446 {
447 	int decsched;
448 
449 	decsched = 1;
450 
451 	switch (csp->csp_cipher_alg) {
452 	case CRYPTO_AES_ICM:
453 	case CRYPTO_AES_NIST_GCM_16:
454 	case CRYPTO_AES_CCM_16:
455 		decsched = 0;
456 		break;
457 	}
458 
459 	if (csp->csp_cipher_alg == CRYPTO_AES_XTS)
460 		keylen /= 2;
461 
462 	switch (keylen * 8) {
463 	case 128:
464 		ses->rounds = AES128_ROUNDS;
465 		break;
466 	case 192:
467 		ses->rounds = AES192_ROUNDS;
468 		break;
469 	case 256:
470 		ses->rounds = AES256_ROUNDS;
471 		break;
472 	default:
473 		panic("shouldn't happen");
474 	}
475 
476 	aesni_set_enckey(key, ses->enc_schedule, ses->rounds);
477 	if (decsched)
478 		aesni_set_deckey(ses->enc_schedule, ses->dec_schedule,
479 		    ses->rounds);
480 
481 	if (csp->csp_cipher_alg == CRYPTO_AES_XTS)
482 		aesni_set_enckey(key + keylen, ses->xts_schedule,
483 		    ses->rounds);
484 }
485