1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Bit sliced AES using NEON instructions
4 *
5 * Copyright (C) 2016 - 2017 Linaro Ltd <ard.biesheuvel@linaro.org>
6 */
7
8 #include <asm/neon.h>
9 #include <asm/simd.h>
10 #include <crypto/aes.h>
11 #include <crypto/ctr.h>
12 #include <crypto/internal/simd.h>
13 #include <crypto/internal/skcipher.h>
14 #include <crypto/scatterwalk.h>
15 #include <crypto/xts.h>
16 #include <linux/module.h>
17
18 MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>");
19 MODULE_DESCRIPTION("Bit sliced AES using NEON instructions");
20 MODULE_LICENSE("GPL v2");
21
22 MODULE_ALIAS_CRYPTO("ecb(aes)");
23 MODULE_ALIAS_CRYPTO("cbc(aes)");
24 MODULE_ALIAS_CRYPTO("ctr(aes)");
25 MODULE_ALIAS_CRYPTO("xts(aes)");
26
27 asmlinkage void aesbs_convert_key(u8 out[], u32 const rk[], int rounds);
28
29 asmlinkage void aesbs_ecb_encrypt(u8 out[], u8 const in[], u8 const rk[],
30 int rounds, int blocks);
31 asmlinkage void aesbs_ecb_decrypt(u8 out[], u8 const in[], u8 const rk[],
32 int rounds, int blocks);
33
34 asmlinkage void aesbs_cbc_decrypt(u8 out[], u8 const in[], u8 const rk[],
35 int rounds, int blocks, u8 iv[]);
36
37 asmlinkage void aesbs_ctr_encrypt(u8 out[], u8 const in[], u8 const rk[],
38 int rounds, int blocks, u8 iv[]);
39
40 asmlinkage void aesbs_xts_encrypt(u8 out[], u8 const in[], u8 const rk[],
41 int rounds, int blocks, u8 iv[]);
42 asmlinkage void aesbs_xts_decrypt(u8 out[], u8 const in[], u8 const rk[],
43 int rounds, int blocks, u8 iv[]);
44
45 /* borrowed from aes-neon-blk.ko */
46 asmlinkage void neon_aes_ecb_encrypt(u8 out[], u8 const in[], u32 const rk[],
47 int rounds, int blocks);
48 asmlinkage void neon_aes_cbc_encrypt(u8 out[], u8 const in[], u32 const rk[],
49 int rounds, int blocks, u8 iv[]);
50 asmlinkage void neon_aes_ctr_encrypt(u8 out[], u8 const in[], u32 const rk[],
51 int rounds, int bytes, u8 ctr[]);
52 asmlinkage void neon_aes_xts_encrypt(u8 out[], u8 const in[],
53 u32 const rk1[], int rounds, int bytes,
54 u32 const rk2[], u8 iv[], int first);
55 asmlinkage void neon_aes_xts_decrypt(u8 out[], u8 const in[],
56 u32 const rk1[], int rounds, int bytes,
57 u32 const rk2[], u8 iv[], int first);
58
59 struct aesbs_ctx {
60 u8 rk[13 * (8 * AES_BLOCK_SIZE) + 32];
61 int rounds;
62 } __aligned(AES_BLOCK_SIZE);
63
64 struct aesbs_cbc_ctr_ctx {
65 struct aesbs_ctx key;
66 u32 enc[AES_MAX_KEYLENGTH_U32];
67 };
68
69 struct aesbs_xts_ctx {
70 struct aesbs_ctx key;
71 u32 twkey[AES_MAX_KEYLENGTH_U32];
72 struct crypto_aes_ctx cts;
73 };
74
75 static int aesbs_setkey(struct crypto_skcipher *tfm, const u8 *in_key,
76 unsigned int key_len)
77 {
78 struct aesbs_ctx *ctx = crypto_skcipher_ctx(tfm);
79 struct crypto_aes_ctx rk;
80 int err;
81
82 err = aes_expandkey(&rk, in_key, key_len);
83 if (err)
84 return err;
85
86 ctx->rounds = 6 + key_len / 4;
87
88 kernel_neon_begin();
89 aesbs_convert_key(ctx->rk, rk.key_enc, ctx->rounds);
90 kernel_neon_end();
91
92 return 0;
93 }
94
95 static int __ecb_crypt(struct skcipher_request *req,
96 void (*fn)(u8 out[], u8 const in[], u8 const rk[],
97 int rounds, int blocks))
98 {
99 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
100 struct aesbs_ctx *ctx = crypto_skcipher_ctx(tfm);
101 struct skcipher_walk walk;
102 int err;
103
104 err = skcipher_walk_virt(&walk, req, false);
105
106 while (walk.nbytes >= AES_BLOCK_SIZE) {
107 unsigned int blocks = walk.nbytes / AES_BLOCK_SIZE;
108
109 if (walk.nbytes < walk.total)
110 blocks = round_down(blocks,
111 walk.stride / AES_BLOCK_SIZE);
112
113 kernel_neon_begin();
114 fn(walk.dst.virt.addr, walk.src.virt.addr, ctx->rk,
115 ctx->rounds, blocks);
116 kernel_neon_end();
117 err = skcipher_walk_done(&walk,
118 walk.nbytes - blocks * AES_BLOCK_SIZE);
119 }
120
121 return err;
122 }
123
124 static int ecb_encrypt(struct skcipher_request *req)
125 {
126 return __ecb_crypt(req, aesbs_ecb_encrypt);
127 }
128
129 static int ecb_decrypt(struct skcipher_request *req)
130 {
131 return __ecb_crypt(req, aesbs_ecb_decrypt);
132 }
133
134 static int aesbs_cbc_ctr_setkey(struct crypto_skcipher *tfm, const u8 *in_key,
135 unsigned int key_len)
136 {
137 struct aesbs_cbc_ctr_ctx *ctx = crypto_skcipher_ctx(tfm);
138 struct crypto_aes_ctx rk;
139 int err;
140
141 err = aes_expandkey(&rk, in_key, key_len);
142 if (err)
143 return err;
144
145 ctx->key.rounds = 6 + key_len / 4;
146
147 memcpy(ctx->enc, rk.key_enc, sizeof(ctx->enc));
148
149 kernel_neon_begin();
150 aesbs_convert_key(ctx->key.rk, rk.key_enc, ctx->key.rounds);
151 kernel_neon_end();
152 memzero_explicit(&rk, sizeof(rk));
153
154 return 0;
155 }
156
157 static int cbc_encrypt(struct skcipher_request *req)
158 {
159 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
160 struct aesbs_cbc_ctr_ctx *ctx = crypto_skcipher_ctx(tfm);
161 struct skcipher_walk walk;
162 int err;
163
164 err = skcipher_walk_virt(&walk, req, false);
165
166 while (walk.nbytes >= AES_BLOCK_SIZE) {
167 unsigned int blocks = walk.nbytes / AES_BLOCK_SIZE;
168
169 /* fall back to the non-bitsliced NEON implementation */
170 kernel_neon_begin();
171 neon_aes_cbc_encrypt(walk.dst.virt.addr, walk.src.virt.addr,
172 ctx->enc, ctx->key.rounds, blocks,
173 walk.iv);
174 kernel_neon_end();
175 err = skcipher_walk_done(&walk, walk.nbytes % AES_BLOCK_SIZE);
176 }
177 return err;
178 }
179
180 static int cbc_decrypt(struct skcipher_request *req)
181 {
182 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
183 struct aesbs_cbc_ctr_ctx *ctx = crypto_skcipher_ctx(tfm);
184 struct skcipher_walk walk;
185 int err;
186
187 err = skcipher_walk_virt(&walk, req, false);
188
189 while (walk.nbytes >= AES_BLOCK_SIZE) {
190 unsigned int blocks = walk.nbytes / AES_BLOCK_SIZE;
191
192 if (walk.nbytes < walk.total)
193 blocks = round_down(blocks,
194 walk.stride / AES_BLOCK_SIZE);
195
196 kernel_neon_begin();
197 aesbs_cbc_decrypt(walk.dst.virt.addr, walk.src.virt.addr,
198 ctx->key.rk, ctx->key.rounds, blocks,
199 walk.iv);
200 kernel_neon_end();
201 err = skcipher_walk_done(&walk,
202 walk.nbytes - blocks * AES_BLOCK_SIZE);
203 }
204
205 return err;
206 }
207
208 static int ctr_encrypt(struct skcipher_request *req)
209 {
210 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
211 struct aesbs_cbc_ctr_ctx *ctx = crypto_skcipher_ctx(tfm);
212 struct skcipher_walk walk;
213 int err;
214
215 err = skcipher_walk_virt(&walk, req, false);
216
217 while (walk.nbytes > 0) {
218 int blocks = (walk.nbytes / AES_BLOCK_SIZE) & ~7;
219 int nbytes = walk.nbytes % (8 * AES_BLOCK_SIZE);
220 const u8 *src = walk.src.virt.addr;
221 u8 *dst = walk.dst.virt.addr;
222
223 kernel_neon_begin();
224 if (blocks >= 8) {
225 aesbs_ctr_encrypt(dst, src, ctx->key.rk, ctx->key.rounds,
226 blocks, walk.iv);
227 dst += blocks * AES_BLOCK_SIZE;
228 src += blocks * AES_BLOCK_SIZE;
229 }
230 if (nbytes && walk.nbytes == walk.total) {
231 u8 buf[AES_BLOCK_SIZE];
232 u8 *d = dst;
233
234 if (unlikely(nbytes < AES_BLOCK_SIZE))
235 src = dst = memcpy(buf + sizeof(buf) - nbytes,
236 src, nbytes);
237
238 neon_aes_ctr_encrypt(dst, src, ctx->enc, ctx->key.rounds,
239 nbytes, walk.iv);
240
241 if (unlikely(nbytes < AES_BLOCK_SIZE))
242 memcpy(d, dst, nbytes);
243
244 nbytes = 0;
245 }
246 kernel_neon_end();
247 err = skcipher_walk_done(&walk, nbytes);
248 }
249 return err;
250 }
251
252 static int aesbs_xts_setkey(struct crypto_skcipher *tfm, const u8 *in_key,
253 unsigned int key_len)
254 {
255 struct aesbs_xts_ctx *ctx = crypto_skcipher_ctx(tfm);
256 struct crypto_aes_ctx rk;
257 int err;
258
259 err = xts_verify_key(tfm, in_key, key_len);
260 if (err)
261 return err;
262
263 key_len /= 2;
264 err = aes_expandkey(&ctx->cts, in_key, key_len);
265 if (err)
266 return err;
267
268 err = aes_expandkey(&rk, in_key + key_len, key_len);
269 if (err)
270 return err;
271
272 memcpy(ctx->twkey, rk.key_enc, sizeof(ctx->twkey));
273
274 return aesbs_setkey(tfm, in_key, key_len);
275 }
276
277 static int __xts_crypt(struct skcipher_request *req, bool encrypt,
278 void (*fn)(u8 out[], u8 const in[], u8 const rk[],
279 int rounds, int blocks, u8 iv[]))
280 {
281 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
282 struct aesbs_xts_ctx *ctx = crypto_skcipher_ctx(tfm);
283 int tail = req->cryptlen % (8 * AES_BLOCK_SIZE);
284 struct scatterlist sg_src[2], sg_dst[2];
285 struct skcipher_request subreq;
286 struct scatterlist *src, *dst;
287 struct skcipher_walk walk;
288 int nbytes, err;
289 int first = 1;
290 u8 *out, *in;
291
292 if (req->cryptlen < AES_BLOCK_SIZE)
293 return -EINVAL;
294
295 /* ensure that the cts tail is covered by a single step */
296 if (unlikely(tail > 0 && tail < AES_BLOCK_SIZE)) {
297 int xts_blocks = DIV_ROUND_UP(req->cryptlen,
298 AES_BLOCK_SIZE) - 2;
299
300 skcipher_request_set_tfm(&subreq, tfm);
301 skcipher_request_set_callback(&subreq,
302 skcipher_request_flags(req),
303 NULL, NULL);
304 skcipher_request_set_crypt(&subreq, req->src, req->dst,
305 xts_blocks * AES_BLOCK_SIZE,
306 req->iv);
307 req = &subreq;
308 } else {
309 tail = 0;
310 }
311
312 err = skcipher_walk_virt(&walk, req, false);
313 if (err)
314 return err;
315
316 while (walk.nbytes >= AES_BLOCK_SIZE) {
317 int blocks = (walk.nbytes / AES_BLOCK_SIZE) & ~7;
318 out = walk.dst.virt.addr;
319 in = walk.src.virt.addr;
320 nbytes = walk.nbytes;
321
322 kernel_neon_begin();
323 if (blocks >= 8) {
324 if (first == 1)
325 neon_aes_ecb_encrypt(walk.iv, walk.iv,
326 ctx->twkey,
327 ctx->key.rounds, 1);
328 first = 2;
329
330 fn(out, in, ctx->key.rk, ctx->key.rounds, blocks,
331 walk.iv);
332
333 out += blocks * AES_BLOCK_SIZE;
334 in += blocks * AES_BLOCK_SIZE;
335 nbytes -= blocks * AES_BLOCK_SIZE;
336 }
337 if (walk.nbytes == walk.total && nbytes > 0) {
338 if (encrypt)
339 neon_aes_xts_encrypt(out, in, ctx->cts.key_enc,
340 ctx->key.rounds, nbytes,
341 ctx->twkey, walk.iv, first);
342 else
343 neon_aes_xts_decrypt(out, in, ctx->cts.key_dec,
344 ctx->key.rounds, nbytes,
345 ctx->twkey, walk.iv, first);
346 nbytes = first = 0;
347 }
348 kernel_neon_end();
349 err = skcipher_walk_done(&walk, nbytes);
350 }
351
352 if (err || likely(!tail))
353 return err;
354
355 /* handle ciphertext stealing */
356 dst = src = scatterwalk_ffwd(sg_src, req->src, req->cryptlen);
357 if (req->dst != req->src)
358 dst = scatterwalk_ffwd(sg_dst, req->dst, req->cryptlen);
359
360 skcipher_request_set_crypt(req, src, dst, AES_BLOCK_SIZE + tail,
361 req->iv);
362
363 err = skcipher_walk_virt(&walk, req, false);
364 if (err)
365 return err;
366
367 out = walk.dst.virt.addr;
368 in = walk.src.virt.addr;
369 nbytes = walk.nbytes;
370
371 kernel_neon_begin();
372 if (encrypt)
373 neon_aes_xts_encrypt(out, in, ctx->cts.key_enc, ctx->key.rounds,
374 nbytes, ctx->twkey, walk.iv, first);
375 else
376 neon_aes_xts_decrypt(out, in, ctx->cts.key_dec, ctx->key.rounds,
377 nbytes, ctx->twkey, walk.iv, first);
378 kernel_neon_end();
379
380 return skcipher_walk_done(&walk, 0);
381 }
382
383 static int xts_encrypt(struct skcipher_request *req)
384 {
385 return __xts_crypt(req, true, aesbs_xts_encrypt);
386 }
387
388 static int xts_decrypt(struct skcipher_request *req)
389 {
390 return __xts_crypt(req, false, aesbs_xts_decrypt);
391 }
392
393 static struct skcipher_alg aes_algs[] = { {
394 .base.cra_name = "ecb(aes)",
395 .base.cra_driver_name = "ecb-aes-neonbs",
396 .base.cra_priority = 250,
397 .base.cra_blocksize = AES_BLOCK_SIZE,
398 .base.cra_ctxsize = sizeof(struct aesbs_ctx),
399 .base.cra_module = THIS_MODULE,
400
401 .min_keysize = AES_MIN_KEY_SIZE,
402 .max_keysize = AES_MAX_KEY_SIZE,
403 .walksize = 8 * AES_BLOCK_SIZE,
404 .setkey = aesbs_setkey,
405 .encrypt = ecb_encrypt,
406 .decrypt = ecb_decrypt,
407 }, {
408 .base.cra_name = "cbc(aes)",
409 .base.cra_driver_name = "cbc-aes-neonbs",
410 .base.cra_priority = 250,
411 .base.cra_blocksize = AES_BLOCK_SIZE,
412 .base.cra_ctxsize = sizeof(struct aesbs_cbc_ctr_ctx),
413 .base.cra_module = THIS_MODULE,
414
415 .min_keysize = AES_MIN_KEY_SIZE,
416 .max_keysize = AES_MAX_KEY_SIZE,
417 .walksize = 8 * AES_BLOCK_SIZE,
418 .ivsize = AES_BLOCK_SIZE,
419 .setkey = aesbs_cbc_ctr_setkey,
420 .encrypt = cbc_encrypt,
421 .decrypt = cbc_decrypt,
422 }, {
423 .base.cra_name = "ctr(aes)",
424 .base.cra_driver_name = "ctr-aes-neonbs",
425 .base.cra_priority = 250,
426 .base.cra_blocksize = 1,
427 .base.cra_ctxsize = sizeof(struct aesbs_cbc_ctr_ctx),
428 .base.cra_module = THIS_MODULE,
429
430 .min_keysize = AES_MIN_KEY_SIZE,
431 .max_keysize = AES_MAX_KEY_SIZE,
432 .chunksize = AES_BLOCK_SIZE,
433 .walksize = 8 * AES_BLOCK_SIZE,
434 .ivsize = AES_BLOCK_SIZE,
435 .setkey = aesbs_cbc_ctr_setkey,
436 .encrypt = ctr_encrypt,
437 .decrypt = ctr_encrypt,
438 }, {
439 .base.cra_name = "xts(aes)",
440 .base.cra_driver_name = "xts-aes-neonbs",
441 .base.cra_priority = 250,
442 .base.cra_blocksize = AES_BLOCK_SIZE,
443 .base.cra_ctxsize = sizeof(struct aesbs_xts_ctx),
444 .base.cra_module = THIS_MODULE,
445
446 .min_keysize = 2 * AES_MIN_KEY_SIZE,
447 .max_keysize = 2 * AES_MAX_KEY_SIZE,
448 .walksize = 8 * AES_BLOCK_SIZE,
449 .ivsize = AES_BLOCK_SIZE,
450 .setkey = aesbs_xts_setkey,
451 .encrypt = xts_encrypt,
452 .decrypt = xts_decrypt,
453 } };
454
455 static void aes_exit(void)
456 {
457 crypto_unregister_skciphers(aes_algs, ARRAY_SIZE(aes_algs));
458 }
459
460 static int __init aes_init(void)
461 {
462 if (!cpu_have_named_feature(ASIMD))
463 return -ENODEV;
464
465 return crypto_register_skciphers(aes_algs, ARRAY_SIZE(aes_algs));
466 }
467
468 module_init(aes_init);
469 module_exit(aes_exit);
470
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.