1
2
3
4
5
6
7
8
9
10
11
12
13#include <linux/module.h>
14#include <linux/types.h>
15#include <linux/crypto.h>
16#include <linux/err.h>
17#include <crypto/algapi.h>
18#include <crypto/ctr.h>
19#include <crypto/twofish.h>
20#include <crypto/lrw.h>
21#include <crypto/xts.h>
22#include <asm/xcr.h>
23#include <asm/xsave.h>
24#include <asm/crypto/twofish.h>
25#include <asm/crypto/ablk_helper.h>
26#include <asm/crypto/glue_helper.h>
27#include <crypto/scatterwalk.h>
28
29#define TF_AVX2_PARALLEL_BLOCKS 16
30
31
32asmlinkage void twofish_ecb_enc_16way(struct twofish_ctx *ctx, u8 *dst,
33 const u8 *src);
34asmlinkage void twofish_ecb_dec_16way(struct twofish_ctx *ctx, u8 *dst,
35 const u8 *src);
36asmlinkage void twofish_cbc_dec_16way(void *ctx, u128 *dst, const u128 *src);
37
38asmlinkage void twofish_ctr_16way(void *ctx, u128 *dst, const u128 *src,
39 le128 *iv);
40
41asmlinkage void twofish_xts_enc_16way(struct twofish_ctx *ctx, u8 *dst,
42 const u8 *src, le128 *iv);
43asmlinkage void twofish_xts_dec_16way(struct twofish_ctx *ctx, u8 *dst,
44 const u8 *src, le128 *iv);
45
46static inline void twofish_enc_blk_3way(struct twofish_ctx *ctx, u8 *dst,
47 const u8 *src)
48{
49 __twofish_enc_blk_3way(ctx, dst, src, false);
50}
51
52static const struct common_glue_ctx twofish_enc = {
53 .num_funcs = 4,
54 .fpu_blocks_limit = 8,
55
56 .funcs = { {
57 .num_blocks = 16,
58 .fn_u = { .ecb = GLUE_FUNC_CAST(twofish_ecb_enc_16way) }
59 }, {
60 .num_blocks = 8,
61 .fn_u = { .ecb = GLUE_FUNC_CAST(twofish_ecb_enc_8way) }
62 }, {
63 .num_blocks = 3,
64 .fn_u = { .ecb = GLUE_FUNC_CAST(twofish_enc_blk_3way) }
65 }, {
66 .num_blocks = 1,
67 .fn_u = { .ecb = GLUE_FUNC_CAST(twofish_enc_blk) }
68 } }
69};
70
71static const struct common_glue_ctx twofish_ctr = {
72 .num_funcs = 4,
73 .fpu_blocks_limit = 8,
74
75 .funcs = { {
76 .num_blocks = 16,
77 .fn_u = { .ctr = GLUE_CTR_FUNC_CAST(twofish_ctr_16way) }
78 }, {
79 .num_blocks = 8,
80 .fn_u = { .ctr = GLUE_CTR_FUNC_CAST(twofish_ctr_8way) }
81 }, {
82 .num_blocks = 3,
83 .fn_u = { .ctr = GLUE_CTR_FUNC_CAST(twofish_enc_blk_ctr_3way) }
84 }, {
85 .num_blocks = 1,
86 .fn_u = { .ctr = GLUE_CTR_FUNC_CAST(twofish_enc_blk_ctr) }
87 } }
88};
89
90static const struct common_glue_ctx twofish_enc_xts = {
91 .num_funcs = 3,
92 .fpu_blocks_limit = 8,
93
94 .funcs = { {
95 .num_blocks = 16,
96 .fn_u = { .xts = GLUE_XTS_FUNC_CAST(twofish_xts_enc_16way) }
97 }, {
98 .num_blocks = 8,
99 .fn_u = { .xts = GLUE_XTS_FUNC_CAST(twofish_xts_enc_8way) }
100 }, {
101 .num_blocks = 1,
102 .fn_u = { .xts = GLUE_XTS_FUNC_CAST(twofish_xts_enc) }
103 } }
104};
105
106static const struct common_glue_ctx twofish_dec = {
107 .num_funcs = 4,
108 .fpu_blocks_limit = 8,
109
110 .funcs = { {
111 .num_blocks = 16,
112 .fn_u = { .ecb = GLUE_FUNC_CAST(twofish_ecb_dec_16way) }
113 }, {
114 .num_blocks = 8,
115 .fn_u = { .ecb = GLUE_FUNC_CAST(twofish_ecb_dec_8way) }
116 }, {
117 .num_blocks = 3,
118 .fn_u = { .ecb = GLUE_FUNC_CAST(twofish_dec_blk_3way) }
119 }, {
120 .num_blocks = 1,
121 .fn_u = { .ecb = GLUE_FUNC_CAST(twofish_dec_blk) }
122 } }
123};
124
125static const struct common_glue_ctx twofish_dec_cbc = {
126 .num_funcs = 4,
127 .fpu_blocks_limit = 8,
128
129 .funcs = { {
130 .num_blocks = 16,
131 .fn_u = { .cbc = GLUE_CBC_FUNC_CAST(twofish_cbc_dec_16way) }
132 }, {
133 .num_blocks = 8,
134 .fn_u = { .cbc = GLUE_CBC_FUNC_CAST(twofish_cbc_dec_8way) }
135 }, {
136 .num_blocks = 3,
137 .fn_u = { .cbc = GLUE_CBC_FUNC_CAST(twofish_dec_blk_cbc_3way) }
138 }, {
139 .num_blocks = 1,
140 .fn_u = { .cbc = GLUE_CBC_FUNC_CAST(twofish_dec_blk) }
141 } }
142};
143
144static const struct common_glue_ctx twofish_dec_xts = {
145 .num_funcs = 3,
146 .fpu_blocks_limit = 8,
147
148 .funcs = { {
149 .num_blocks = 16,
150 .fn_u = { .xts = GLUE_XTS_FUNC_CAST(twofish_xts_dec_16way) }
151 }, {
152 .num_blocks = 8,
153 .fn_u = { .xts = GLUE_XTS_FUNC_CAST(twofish_xts_dec_8way) }
154 }, {
155 .num_blocks = 1,
156 .fn_u = { .xts = GLUE_XTS_FUNC_CAST(twofish_xts_dec) }
157 } }
158};
159
160static int ecb_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
161 struct scatterlist *src, unsigned int nbytes)
162{
163 return glue_ecb_crypt_128bit(&twofish_enc, desc, dst, src, nbytes);
164}
165
166static int ecb_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
167 struct scatterlist *src, unsigned int nbytes)
168{
169 return glue_ecb_crypt_128bit(&twofish_dec, desc, dst, src, nbytes);
170}
171
172static int cbc_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
173 struct scatterlist *src, unsigned int nbytes)
174{
175 return glue_cbc_encrypt_128bit(GLUE_FUNC_CAST(twofish_enc_blk), desc,
176 dst, src, nbytes);
177}
178
179static int cbc_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
180 struct scatterlist *src, unsigned int nbytes)
181{
182 return glue_cbc_decrypt_128bit(&twofish_dec_cbc, desc, dst, src,
183 nbytes);
184}
185
186static int ctr_crypt(struct blkcipher_desc *desc, struct scatterlist *dst,
187 struct scatterlist *src, unsigned int nbytes)
188{
189 return glue_ctr_crypt_128bit(&twofish_ctr, desc, dst, src, nbytes);
190}
191
192static inline bool twofish_fpu_begin(bool fpu_enabled, unsigned int nbytes)
193{
194
195 return glue_fpu_begin(TF_BLOCK_SIZE, 8, NULL, fpu_enabled, nbytes);
196}
197
198static inline void twofish_fpu_end(bool fpu_enabled)
199{
200 glue_fpu_end(fpu_enabled);
201}
202
203struct crypt_priv {
204 struct twofish_ctx *ctx;
205 bool fpu_enabled;
206};
207
208static void encrypt_callback(void *priv, u8 *srcdst, unsigned int nbytes)
209{
210 const unsigned int bsize = TF_BLOCK_SIZE;
211 struct crypt_priv *ctx = priv;
212 int i;
213
214 ctx->fpu_enabled = twofish_fpu_begin(ctx->fpu_enabled, nbytes);
215
216 while (nbytes >= TF_AVX2_PARALLEL_BLOCKS * bsize) {
217 twofish_ecb_enc_16way(ctx->ctx, srcdst, srcdst);
218 srcdst += bsize * TF_AVX2_PARALLEL_BLOCKS;
219 nbytes -= bsize * TF_AVX2_PARALLEL_BLOCKS;
220 }
221
222 while (nbytes >= 8 * bsize) {
223 twofish_ecb_enc_8way(ctx->ctx, srcdst, srcdst);
224 srcdst += bsize * 8;
225 nbytes -= bsize * 8;
226 }
227
228 while (nbytes >= 3 * bsize) {
229 twofish_enc_blk_3way(ctx->ctx, srcdst, srcdst);
230 srcdst += bsize * 3;
231 nbytes -= bsize * 3;
232 }
233
234 for (i = 0; i < nbytes / bsize; i++, srcdst += bsize)
235 twofish_enc_blk(ctx->ctx, srcdst, srcdst);
236}
237
238static void decrypt_callback(void *priv, u8 *srcdst, unsigned int nbytes)
239{
240 const unsigned int bsize = TF_BLOCK_SIZE;
241 struct crypt_priv *ctx = priv;
242 int i;
243
244 ctx->fpu_enabled = twofish_fpu_begin(ctx->fpu_enabled, nbytes);
245
246 while (nbytes >= TF_AVX2_PARALLEL_BLOCKS * bsize) {
247 twofish_ecb_dec_16way(ctx->ctx, srcdst, srcdst);
248 srcdst += bsize * TF_AVX2_PARALLEL_BLOCKS;
249 nbytes -= bsize * TF_AVX2_PARALLEL_BLOCKS;
250 }
251
252 while (nbytes >= 8 * bsize) {
253 twofish_ecb_dec_8way(ctx->ctx, srcdst, srcdst);
254 srcdst += bsize * 8;
255 nbytes -= bsize * 8;
256 }
257
258 while (nbytes >= 3 * bsize) {
259 twofish_dec_blk_3way(ctx->ctx, srcdst, srcdst);
260 srcdst += bsize * 3;
261 nbytes -= bsize * 3;
262 }
263
264 for (i = 0; i < nbytes / bsize; i++, srcdst += bsize)
265 twofish_dec_blk(ctx->ctx, srcdst, srcdst);
266}
267
268static int lrw_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
269 struct scatterlist *src, unsigned int nbytes)
270{
271 struct twofish_lrw_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
272 be128 buf[TF_AVX2_PARALLEL_BLOCKS];
273 struct crypt_priv crypt_ctx = {
274 .ctx = &ctx->twofish_ctx,
275 .fpu_enabled = false,
276 };
277 struct lrw_crypt_req req = {
278 .tbuf = buf,
279 .tbuflen = sizeof(buf),
280
281 .table_ctx = &ctx->lrw_table,
282 .crypt_ctx = &crypt_ctx,
283 .crypt_fn = encrypt_callback,
284 };
285 int ret;
286
287 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
288 ret = lrw_crypt(desc, dst, src, nbytes, &req);
289 twofish_fpu_end(crypt_ctx.fpu_enabled);
290
291 return ret;
292}
293
294static int lrw_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
295 struct scatterlist *src, unsigned int nbytes)
296{
297 struct twofish_lrw_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
298 be128 buf[TF_AVX2_PARALLEL_BLOCKS];
299 struct crypt_priv crypt_ctx = {
300 .ctx = &ctx->twofish_ctx,
301 .fpu_enabled = false,
302 };
303 struct lrw_crypt_req req = {
304 .tbuf = buf,
305 .tbuflen = sizeof(buf),
306
307 .table_ctx = &ctx->lrw_table,
308 .crypt_ctx = &crypt_ctx,
309 .crypt_fn = decrypt_callback,
310 };
311 int ret;
312
313 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
314 ret = lrw_crypt(desc, dst, src, nbytes, &req);
315 twofish_fpu_end(crypt_ctx.fpu_enabled);
316
317 return ret;
318}
319
320static int xts_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
321 struct scatterlist *src, unsigned int nbytes)
322{
323 struct twofish_xts_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
324
325 return glue_xts_crypt_128bit(&twofish_enc_xts, desc, dst, src, nbytes,
326 XTS_TWEAK_CAST(twofish_enc_blk),
327 &ctx->tweak_ctx, &ctx->crypt_ctx);
328}
329
330static int xts_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
331 struct scatterlist *src, unsigned int nbytes)
332{
333 struct twofish_xts_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
334
335 return glue_xts_crypt_128bit(&twofish_dec_xts, desc, dst, src, nbytes,
336 XTS_TWEAK_CAST(twofish_enc_blk),
337 &ctx->tweak_ctx, &ctx->crypt_ctx);
338}
339
340static struct crypto_alg tf_algs[10] = { {
341 .cra_name = "__ecb-twofish-avx2",
342 .cra_driver_name = "__driver-ecb-twofish-avx2",
343 .cra_priority = 0,
344 .cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
345 .cra_blocksize = TF_BLOCK_SIZE,
346 .cra_ctxsize = sizeof(struct twofish_ctx),
347 .cra_alignmask = 0,
348 .cra_type = &crypto_blkcipher_type,
349 .cra_module = THIS_MODULE,
350 .cra_u = {
351 .blkcipher = {
352 .min_keysize = TF_MIN_KEY_SIZE,
353 .max_keysize = TF_MAX_KEY_SIZE,
354 .setkey = twofish_setkey,
355 .encrypt = ecb_encrypt,
356 .decrypt = ecb_decrypt,
357 },
358 },
359}, {
360 .cra_name = "__cbc-twofish-avx2",
361 .cra_driver_name = "__driver-cbc-twofish-avx2",
362 .cra_priority = 0,
363 .cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
364 .cra_blocksize = TF_BLOCK_SIZE,
365 .cra_ctxsize = sizeof(struct twofish_ctx),
366 .cra_alignmask = 0,
367 .cra_type = &crypto_blkcipher_type,
368 .cra_module = THIS_MODULE,
369 .cra_u = {
370 .blkcipher = {
371 .min_keysize = TF_MIN_KEY_SIZE,
372 .max_keysize = TF_MAX_KEY_SIZE,
373 .setkey = twofish_setkey,
374 .encrypt = cbc_encrypt,
375 .decrypt = cbc_decrypt,
376 },
377 },
378}, {
379 .cra_name = "__ctr-twofish-avx2",
380 .cra_driver_name = "__driver-ctr-twofish-avx2",
381 .cra_priority = 0,
382 .cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
383 .cra_blocksize = 1,
384 .cra_ctxsize = sizeof(struct twofish_ctx),
385 .cra_alignmask = 0,
386 .cra_type = &crypto_blkcipher_type,
387 .cra_module = THIS_MODULE,
388 .cra_u = {
389 .blkcipher = {
390 .min_keysize = TF_MIN_KEY_SIZE,
391 .max_keysize = TF_MAX_KEY_SIZE,
392 .ivsize = TF_BLOCK_SIZE,
393 .setkey = twofish_setkey,
394 .encrypt = ctr_crypt,
395 .decrypt = ctr_crypt,
396 },
397 },
398}, {
399 .cra_name = "__lrw-twofish-avx2",
400 .cra_driver_name = "__driver-lrw-twofish-avx2",
401 .cra_priority = 0,
402 .cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
403 .cra_blocksize = TF_BLOCK_SIZE,
404 .cra_ctxsize = sizeof(struct twofish_lrw_ctx),
405 .cra_alignmask = 0,
406 .cra_type = &crypto_blkcipher_type,
407 .cra_module = THIS_MODULE,
408 .cra_exit = lrw_twofish_exit_tfm,
409 .cra_u = {
410 .blkcipher = {
411 .min_keysize = TF_MIN_KEY_SIZE +
412 TF_BLOCK_SIZE,
413 .max_keysize = TF_MAX_KEY_SIZE +
414 TF_BLOCK_SIZE,
415 .ivsize = TF_BLOCK_SIZE,
416 .setkey = lrw_twofish_setkey,
417 .encrypt = lrw_encrypt,
418 .decrypt = lrw_decrypt,
419 },
420 },
421}, {
422 .cra_name = "__xts-twofish-avx2",
423 .cra_driver_name = "__driver-xts-twofish-avx2",
424 .cra_priority = 0,
425 .cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
426 .cra_blocksize = TF_BLOCK_SIZE,
427 .cra_ctxsize = sizeof(struct twofish_xts_ctx),
428 .cra_alignmask = 0,
429 .cra_type = &crypto_blkcipher_type,
430 .cra_module = THIS_MODULE,
431 .cra_u = {
432 .blkcipher = {
433 .min_keysize = TF_MIN_KEY_SIZE * 2,
434 .max_keysize = TF_MAX_KEY_SIZE * 2,
435 .ivsize = TF_BLOCK_SIZE,
436 .setkey = xts_twofish_setkey,
437 .encrypt = xts_encrypt,
438 .decrypt = xts_decrypt,
439 },
440 },
441}, {
442 .cra_name = "ecb(twofish)",
443 .cra_driver_name = "ecb-twofish-avx2",
444 .cra_priority = 500,
445 .cra_flags = CRYPTO_ALG_TYPE_ABLKCIPHER | CRYPTO_ALG_ASYNC,
446 .cra_blocksize = TF_BLOCK_SIZE,
447 .cra_ctxsize = sizeof(struct async_helper_ctx),
448 .cra_alignmask = 0,
449 .cra_type = &crypto_ablkcipher_type,
450 .cra_module = THIS_MODULE,
451 .cra_init = ablk_init,
452 .cra_exit = ablk_exit,
453 .cra_u = {
454 .ablkcipher = {
455 .min_keysize = TF_MIN_KEY_SIZE,
456 .max_keysize = TF_MAX_KEY_SIZE,
457 .setkey = ablk_set_key,
458 .encrypt = ablk_encrypt,
459 .decrypt = ablk_decrypt,
460 },
461 },
462}, {
463 .cra_name = "cbc(twofish)",
464 .cra_driver_name = "cbc-twofish-avx2",
465 .cra_priority = 500,
466 .cra_flags = CRYPTO_ALG_TYPE_ABLKCIPHER | CRYPTO_ALG_ASYNC,
467 .cra_blocksize = TF_BLOCK_SIZE,
468 .cra_ctxsize = sizeof(struct async_helper_ctx),
469 .cra_alignmask = 0,
470 .cra_type = &crypto_ablkcipher_type,
471 .cra_module = THIS_MODULE,
472 .cra_init = ablk_init,
473 .cra_exit = ablk_exit,
474 .cra_u = {
475 .ablkcipher = {
476 .min_keysize = TF_MIN_KEY_SIZE,
477 .max_keysize = TF_MAX_KEY_SIZE,
478 .ivsize = TF_BLOCK_SIZE,
479 .setkey = ablk_set_key,
480 .encrypt = __ablk_encrypt,
481 .decrypt = ablk_decrypt,
482 },
483 },
484}, {
485 .cra_name = "ctr(twofish)",
486 .cra_driver_name = "ctr-twofish-avx2",
487 .cra_priority = 500,
488 .cra_flags = CRYPTO_ALG_TYPE_ABLKCIPHER | CRYPTO_ALG_ASYNC,
489 .cra_blocksize = 1,
490 .cra_ctxsize = sizeof(struct async_helper_ctx),
491 .cra_alignmask = 0,
492 .cra_type = &crypto_ablkcipher_type,
493 .cra_module = THIS_MODULE,
494 .cra_init = ablk_init,
495 .cra_exit = ablk_exit,
496 .cra_u = {
497 .ablkcipher = {
498 .min_keysize = TF_MIN_KEY_SIZE,
499 .max_keysize = TF_MAX_KEY_SIZE,
500 .ivsize = TF_BLOCK_SIZE,
501 .setkey = ablk_set_key,
502 .encrypt = ablk_encrypt,
503 .decrypt = ablk_encrypt,
504 .geniv = "chainiv",
505 },
506 },
507}, {
508 .cra_name = "lrw(twofish)",
509 .cra_driver_name = "lrw-twofish-avx2",
510 .cra_priority = 500,
511 .cra_flags = CRYPTO_ALG_TYPE_ABLKCIPHER | CRYPTO_ALG_ASYNC,
512 .cra_blocksize = TF_BLOCK_SIZE,
513 .cra_ctxsize = sizeof(struct async_helper_ctx),
514 .cra_alignmask = 0,
515 .cra_type = &crypto_ablkcipher_type,
516 .cra_module = THIS_MODULE,
517 .cra_init = ablk_init,
518 .cra_exit = ablk_exit,
519 .cra_u = {
520 .ablkcipher = {
521 .min_keysize = TF_MIN_KEY_SIZE +
522 TF_BLOCK_SIZE,
523 .max_keysize = TF_MAX_KEY_SIZE +
524 TF_BLOCK_SIZE,
525 .ivsize = TF_BLOCK_SIZE,
526 .setkey = ablk_set_key,
527 .encrypt = ablk_encrypt,
528 .decrypt = ablk_decrypt,
529 },
530 },
531}, {
532 .cra_name = "xts(twofish)",
533 .cra_driver_name = "xts-twofish-avx2",
534 .cra_priority = 500,
535 .cra_flags = CRYPTO_ALG_TYPE_ABLKCIPHER | CRYPTO_ALG_ASYNC,
536 .cra_blocksize = TF_BLOCK_SIZE,
537 .cra_ctxsize = sizeof(struct async_helper_ctx),
538 .cra_alignmask = 0,
539 .cra_type = &crypto_ablkcipher_type,
540 .cra_module = THIS_MODULE,
541 .cra_init = ablk_init,
542 .cra_exit = ablk_exit,
543 .cra_u = {
544 .ablkcipher = {
545 .min_keysize = TF_MIN_KEY_SIZE * 2,
546 .max_keysize = TF_MAX_KEY_SIZE * 2,
547 .ivsize = TF_BLOCK_SIZE,
548 .setkey = ablk_set_key,
549 .encrypt = ablk_encrypt,
550 .decrypt = ablk_decrypt,
551 },
552 },
553} };
554
555static int __init init(void)
556{
557 u64 xcr0;
558
559 if (!cpu_has_avx2 || !cpu_has_osxsave) {
560 pr_info("AVX2 instructions are not detected.\n");
561 return -ENODEV;
562 }
563
564 xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
565 if ((xcr0 & (XSTATE_SSE | XSTATE_YMM)) != (XSTATE_SSE | XSTATE_YMM)) {
566 pr_info("AVX2 detected but unusable.\n");
567 return -ENODEV;
568 }
569
570 return crypto_register_algs(tf_algs, ARRAY_SIZE(tf_algs));
571}
572
573static void __exit fini(void)
574{
575 crypto_unregister_algs(tf_algs, ARRAY_SIZE(tf_algs));
576}
577
578module_init(init);
579module_exit(fini);
580
581MODULE_LICENSE("GPL");
582MODULE_DESCRIPTION("Twofish Cipher Algorithm, AVX2 optimized");
583MODULE_ALIAS("twofish");
584MODULE_ALIAS("twofish-asm");
585
