Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Support for AES-NI and VAES instructions. This file contains glue code.
4 * The real AES implementations are in aesni-intel_asm.S and other .S files.
5 *
6 * Copyright (C) 2008, Intel Corp.
7 * Author: Huang Ying <ying.huang@intel.com>
8 *
9 * Added RFC4106 AES-GCM support for 128-bit keys under the AEAD
10 * interface for 64-bit kernels.
11 * Authors: Adrian Hoban <adrian.hoban@intel.com>
12 * Gabriele Paoloni <gabriele.paoloni@intel.com>
13 * Tadeusz Struk (tadeusz.struk@intel.com)
14 * Aidan O'Mahony (aidan.o.mahony@intel.com)
15 * Copyright (c) 2010, Intel Corporation.
16 *
17 * Copyright 2024 Google LLC
18 */
19
20#include <linux/hardirq.h>
21#include <linux/types.h>
22#include <linux/module.h>
23#include <linux/err.h>
24#include <crypto/algapi.h>
25#include <crypto/aes.h>
26#include <crypto/ctr.h>
27#include <crypto/b128ops.h>
28#include <crypto/gcm.h>
29#include <crypto/xts.h>
30#include <asm/cpu_device_id.h>
31#include <asm/simd.h>
32#include <crypto/scatterwalk.h>
33#include <crypto/internal/aead.h>
34#include <crypto/internal/simd.h>
35#include <crypto/internal/skcipher.h>
36#include <linux/jump_label.h>
37#include <linux/workqueue.h>
38#include <linux/spinlock.h>
39#include <linux/static_call.h>
40
41
42#define AESNI_ALIGN 16
43#define AESNI_ALIGN_ATTR __attribute__ ((__aligned__(AESNI_ALIGN)))
44#define AES_BLOCK_MASK (~(AES_BLOCK_SIZE - 1))
45#define AESNI_ALIGN_EXTRA ((AESNI_ALIGN - 1) & ~(CRYPTO_MINALIGN - 1))
46#define CRYPTO_AES_CTX_SIZE (sizeof(struct crypto_aes_ctx) + AESNI_ALIGN_EXTRA)
47#define XTS_AES_CTX_SIZE (sizeof(struct aesni_xts_ctx) + AESNI_ALIGN_EXTRA)
48
49struct aesni_xts_ctx {
50 struct crypto_aes_ctx tweak_ctx AESNI_ALIGN_ATTR;
51 struct crypto_aes_ctx crypt_ctx AESNI_ALIGN_ATTR;
52};
53
54static inline void *aes_align_addr(void *addr)
55{
56 if (crypto_tfm_ctx_alignment() >= AESNI_ALIGN)
57 return addr;
58 return PTR_ALIGN(addr, AESNI_ALIGN);
59}
60
61asmlinkage void aesni_set_key(struct crypto_aes_ctx *ctx, const u8 *in_key,
62 unsigned int key_len);
63asmlinkage void aesni_enc(const void *ctx, u8 *out, const u8 *in);
64asmlinkage void aesni_dec(const void *ctx, u8 *out, const u8 *in);
65asmlinkage void aesni_ecb_enc(struct crypto_aes_ctx *ctx, u8 *out,
66 const u8 *in, unsigned int len);
67asmlinkage void aesni_ecb_dec(struct crypto_aes_ctx *ctx, u8 *out,
68 const u8 *in, unsigned int len);
69asmlinkage void aesni_cbc_enc(struct crypto_aes_ctx *ctx, u8 *out,
70 const u8 *in, unsigned int len, u8 *iv);
71asmlinkage void aesni_cbc_dec(struct crypto_aes_ctx *ctx, u8 *out,
72 const u8 *in, unsigned int len, u8 *iv);
73asmlinkage void aesni_cts_cbc_enc(struct crypto_aes_ctx *ctx, u8 *out,
74 const u8 *in, unsigned int len, u8 *iv);
75asmlinkage void aesni_cts_cbc_dec(struct crypto_aes_ctx *ctx, u8 *out,
76 const u8 *in, unsigned int len, u8 *iv);
77
78asmlinkage void aesni_xts_enc(const struct crypto_aes_ctx *ctx, u8 *out,
79 const u8 *in, unsigned int len, u8 *iv);
80
81asmlinkage void aesni_xts_dec(const struct crypto_aes_ctx *ctx, u8 *out,
82 const u8 *in, unsigned int len, u8 *iv);
83
84#ifdef CONFIG_X86_64
85
86asmlinkage void aesni_ctr_enc(struct crypto_aes_ctx *ctx, u8 *out,
87 const u8 *in, unsigned int len, u8 *iv);
88DEFINE_STATIC_CALL(aesni_ctr_enc_tfm, aesni_ctr_enc);
89
90asmlinkage void aes_ctr_enc_128_avx_by8(const u8 *in, u8 *iv,
91 void *keys, u8 *out, unsigned int num_bytes);
92asmlinkage void aes_ctr_enc_192_avx_by8(const u8 *in, u8 *iv,
93 void *keys, u8 *out, unsigned int num_bytes);
94asmlinkage void aes_ctr_enc_256_avx_by8(const u8 *in, u8 *iv,
95 void *keys, u8 *out, unsigned int num_bytes);
96
97
98asmlinkage void aes_xctr_enc_128_avx_by8(const u8 *in, const u8 *iv,
99 const void *keys, u8 *out, unsigned int num_bytes,
100 unsigned int byte_ctr);
101
102asmlinkage void aes_xctr_enc_192_avx_by8(const u8 *in, const u8 *iv,
103 const void *keys, u8 *out, unsigned int num_bytes,
104 unsigned int byte_ctr);
105
106asmlinkage void aes_xctr_enc_256_avx_by8(const u8 *in, const u8 *iv,
107 const void *keys, u8 *out, unsigned int num_bytes,
108 unsigned int byte_ctr);
109#endif
110
111static inline struct crypto_aes_ctx *aes_ctx(void *raw_ctx)
112{
113 return aes_align_addr(raw_ctx);
114}
115
116static inline struct aesni_xts_ctx *aes_xts_ctx(struct crypto_skcipher *tfm)
117{
118 return aes_align_addr(crypto_skcipher_ctx(tfm));
119}
120
121static int aes_set_key_common(struct crypto_aes_ctx *ctx,
122 const u8 *in_key, unsigned int key_len)
123{
124 int err;
125
126 if (!crypto_simd_usable())
127 return aes_expandkey(ctx, in_key, key_len);
128
129 err = aes_check_keylen(key_len);
130 if (err)
131 return err;
132
133 kernel_fpu_begin();
134 aesni_set_key(ctx, in_key, key_len);
135 kernel_fpu_end();
136 return 0;
137}
138
139static int aes_set_key(struct crypto_tfm *tfm, const u8 *in_key,
140 unsigned int key_len)
141{
142 return aes_set_key_common(aes_ctx(crypto_tfm_ctx(tfm)), in_key,
143 key_len);
144}
145
146static void aesni_encrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
147{
148 struct crypto_aes_ctx *ctx = aes_ctx(crypto_tfm_ctx(tfm));
149
150 if (!crypto_simd_usable()) {
151 aes_encrypt(ctx, dst, src);
152 } else {
153 kernel_fpu_begin();
154 aesni_enc(ctx, dst, src);
155 kernel_fpu_end();
156 }
157}
158
159static void aesni_decrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
160{
161 struct crypto_aes_ctx *ctx = aes_ctx(crypto_tfm_ctx(tfm));
162
163 if (!crypto_simd_usable()) {
164 aes_decrypt(ctx, dst, src);
165 } else {
166 kernel_fpu_begin();
167 aesni_dec(ctx, dst, src);
168 kernel_fpu_end();
169 }
170}
171
172static int aesni_skcipher_setkey(struct crypto_skcipher *tfm, const u8 *key,
173 unsigned int len)
174{
175 return aes_set_key_common(aes_ctx(crypto_skcipher_ctx(tfm)), key, len);
176}
177
178static int ecb_encrypt(struct skcipher_request *req)
179{
180 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
181 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
182 struct skcipher_walk walk;
183 unsigned int nbytes;
184 int err;
185
186 err = skcipher_walk_virt(&walk, req, false);
187
188 while ((nbytes = walk.nbytes)) {
189 kernel_fpu_begin();
190 aesni_ecb_enc(ctx, walk.dst.virt.addr, walk.src.virt.addr,
191 nbytes & AES_BLOCK_MASK);
192 kernel_fpu_end();
193 nbytes &= AES_BLOCK_SIZE - 1;
194 err = skcipher_walk_done(&walk, nbytes);
195 }
196
197 return err;
198}
199
200static int ecb_decrypt(struct skcipher_request *req)
201{
202 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
203 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
204 struct skcipher_walk walk;
205 unsigned int nbytes;
206 int err;
207
208 err = skcipher_walk_virt(&walk, req, false);
209
210 while ((nbytes = walk.nbytes)) {
211 kernel_fpu_begin();
212 aesni_ecb_dec(ctx, walk.dst.virt.addr, walk.src.virt.addr,
213 nbytes & AES_BLOCK_MASK);
214 kernel_fpu_end();
215 nbytes &= AES_BLOCK_SIZE - 1;
216 err = skcipher_walk_done(&walk, nbytes);
217 }
218
219 return err;
220}
221
222static int cbc_encrypt(struct skcipher_request *req)
223{
224 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
225 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
226 struct skcipher_walk walk;
227 unsigned int nbytes;
228 int err;
229
230 err = skcipher_walk_virt(&walk, req, false);
231
232 while ((nbytes = walk.nbytes)) {
233 kernel_fpu_begin();
234 aesni_cbc_enc(ctx, walk.dst.virt.addr, walk.src.virt.addr,
235 nbytes & AES_BLOCK_MASK, walk.iv);
236 kernel_fpu_end();
237 nbytes &= AES_BLOCK_SIZE - 1;
238 err = skcipher_walk_done(&walk, nbytes);
239 }
240
241 return err;
242}
243
244static int cbc_decrypt(struct skcipher_request *req)
245{
246 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
247 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
248 struct skcipher_walk walk;
249 unsigned int nbytes;
250 int err;
251
252 err = skcipher_walk_virt(&walk, req, false);
253
254 while ((nbytes = walk.nbytes)) {
255 kernel_fpu_begin();
256 aesni_cbc_dec(ctx, walk.dst.virt.addr, walk.src.virt.addr,
257 nbytes & AES_BLOCK_MASK, walk.iv);
258 kernel_fpu_end();
259 nbytes &= AES_BLOCK_SIZE - 1;
260 err = skcipher_walk_done(&walk, nbytes);
261 }
262
263 return err;
264}
265
266static int cts_cbc_encrypt(struct skcipher_request *req)
267{
268 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
269 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
270 int cbc_blocks = DIV_ROUND_UP(req->cryptlen, AES_BLOCK_SIZE) - 2;
271 struct scatterlist *src = req->src, *dst = req->dst;
272 struct scatterlist sg_src[2], sg_dst[2];
273 struct skcipher_request subreq;
274 struct skcipher_walk walk;
275 int err;
276
277 skcipher_request_set_tfm(&subreq, tfm);
278 skcipher_request_set_callback(&subreq, skcipher_request_flags(req),
279 NULL, NULL);
280
281 if (req->cryptlen <= AES_BLOCK_SIZE) {
282 if (req->cryptlen < AES_BLOCK_SIZE)
283 return -EINVAL;
284 cbc_blocks = 1;
285 }
286
287 if (cbc_blocks > 0) {
288 skcipher_request_set_crypt(&subreq, req->src, req->dst,
289 cbc_blocks * AES_BLOCK_SIZE,
290 req->iv);
291
292 err = cbc_encrypt(&subreq);
293 if (err)
294 return err;
295
296 if (req->cryptlen == AES_BLOCK_SIZE)
297 return 0;
298
299 dst = src = scatterwalk_ffwd(sg_src, req->src, subreq.cryptlen);
300 if (req->dst != req->src)
301 dst = scatterwalk_ffwd(sg_dst, req->dst,
302 subreq.cryptlen);
303 }
304
305 /* handle ciphertext stealing */
306 skcipher_request_set_crypt(&subreq, src, dst,
307 req->cryptlen - cbc_blocks * AES_BLOCK_SIZE,
308 req->iv);
309
310 err = skcipher_walk_virt(&walk, &subreq, false);
311 if (err)
312 return err;
313
314 kernel_fpu_begin();
315 aesni_cts_cbc_enc(ctx, walk.dst.virt.addr, walk.src.virt.addr,
316 walk.nbytes, walk.iv);
317 kernel_fpu_end();
318
319 return skcipher_walk_done(&walk, 0);
320}
321
322static int cts_cbc_decrypt(struct skcipher_request *req)
323{
324 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
325 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
326 int cbc_blocks = DIV_ROUND_UP(req->cryptlen, AES_BLOCK_SIZE) - 2;
327 struct scatterlist *src = req->src, *dst = req->dst;
328 struct scatterlist sg_src[2], sg_dst[2];
329 struct skcipher_request subreq;
330 struct skcipher_walk walk;
331 int err;
332
333 skcipher_request_set_tfm(&subreq, tfm);
334 skcipher_request_set_callback(&subreq, skcipher_request_flags(req),
335 NULL, NULL);
336
337 if (req->cryptlen <= AES_BLOCK_SIZE) {
338 if (req->cryptlen < AES_BLOCK_SIZE)
339 return -EINVAL;
340 cbc_blocks = 1;
341 }
342
343 if (cbc_blocks > 0) {
344 skcipher_request_set_crypt(&subreq, req->src, req->dst,
345 cbc_blocks * AES_BLOCK_SIZE,
346 req->iv);
347
348 err = cbc_decrypt(&subreq);
349 if (err)
350 return err;
351
352 if (req->cryptlen == AES_BLOCK_SIZE)
353 return 0;
354
355 dst = src = scatterwalk_ffwd(sg_src, req->src, subreq.cryptlen);
356 if (req->dst != req->src)
357 dst = scatterwalk_ffwd(sg_dst, req->dst,
358 subreq.cryptlen);
359 }
360
361 /* handle ciphertext stealing */
362 skcipher_request_set_crypt(&subreq, src, dst,
363 req->cryptlen - cbc_blocks * AES_BLOCK_SIZE,
364 req->iv);
365
366 err = skcipher_walk_virt(&walk, &subreq, false);
367 if (err)
368 return err;
369
370 kernel_fpu_begin();
371 aesni_cts_cbc_dec(ctx, walk.dst.virt.addr, walk.src.virt.addr,
372 walk.nbytes, walk.iv);
373 kernel_fpu_end();
374
375 return skcipher_walk_done(&walk, 0);
376}
377
378#ifdef CONFIG_X86_64
379static void aesni_ctr_enc_avx_tfm(struct crypto_aes_ctx *ctx, u8 *out,
380 const u8 *in, unsigned int len, u8 *iv)
381{
382 /*
383 * based on key length, override with the by8 version
384 * of ctr mode encryption/decryption for improved performance
385 * aes_set_key_common() ensures that key length is one of
386 * {128,192,256}
387 */
388 if (ctx->key_length == AES_KEYSIZE_128)
389 aes_ctr_enc_128_avx_by8(in, iv, (void *)ctx, out, len);
390 else if (ctx->key_length == AES_KEYSIZE_192)
391 aes_ctr_enc_192_avx_by8(in, iv, (void *)ctx, out, len);
392 else
393 aes_ctr_enc_256_avx_by8(in, iv, (void *)ctx, out, len);
394}
395
396static int ctr_crypt(struct skcipher_request *req)
397{
398 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
399 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
400 u8 keystream[AES_BLOCK_SIZE];
401 struct skcipher_walk walk;
402 unsigned int nbytes;
403 int err;
404
405 err = skcipher_walk_virt(&walk, req, false);
406
407 while ((nbytes = walk.nbytes) > 0) {
408 kernel_fpu_begin();
409 if (nbytes & AES_BLOCK_MASK)
410 static_call(aesni_ctr_enc_tfm)(ctx, walk.dst.virt.addr,
411 walk.src.virt.addr,
412 nbytes & AES_BLOCK_MASK,
413 walk.iv);
414 nbytes &= ~AES_BLOCK_MASK;
415
416 if (walk.nbytes == walk.total && nbytes > 0) {
417 aesni_enc(ctx, keystream, walk.iv);
418 crypto_xor_cpy(walk.dst.virt.addr + walk.nbytes - nbytes,
419 walk.src.virt.addr + walk.nbytes - nbytes,
420 keystream, nbytes);
421 crypto_inc(walk.iv, AES_BLOCK_SIZE);
422 nbytes = 0;
423 }
424 kernel_fpu_end();
425 err = skcipher_walk_done(&walk, nbytes);
426 }
427 return err;
428}
429
430static void aesni_xctr_enc_avx_tfm(struct crypto_aes_ctx *ctx, u8 *out,
431 const u8 *in, unsigned int len, u8 *iv,
432 unsigned int byte_ctr)
433{
434 if (ctx->key_length == AES_KEYSIZE_128)
435 aes_xctr_enc_128_avx_by8(in, iv, (void *)ctx, out, len,
436 byte_ctr);
437 else if (ctx->key_length == AES_KEYSIZE_192)
438 aes_xctr_enc_192_avx_by8(in, iv, (void *)ctx, out, len,
439 byte_ctr);
440 else
441 aes_xctr_enc_256_avx_by8(in, iv, (void *)ctx, out, len,
442 byte_ctr);
443}
444
445static int xctr_crypt(struct skcipher_request *req)
446{
447 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
448 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
449 u8 keystream[AES_BLOCK_SIZE];
450 struct skcipher_walk walk;
451 unsigned int nbytes;
452 unsigned int byte_ctr = 0;
453 int err;
454 __le32 block[AES_BLOCK_SIZE / sizeof(__le32)];
455
456 err = skcipher_walk_virt(&walk, req, false);
457
458 while ((nbytes = walk.nbytes) > 0) {
459 kernel_fpu_begin();
460 if (nbytes & AES_BLOCK_MASK)
461 aesni_xctr_enc_avx_tfm(ctx, walk.dst.virt.addr,
462 walk.src.virt.addr, nbytes & AES_BLOCK_MASK,
463 walk.iv, byte_ctr);
464 nbytes &= ~AES_BLOCK_MASK;
465 byte_ctr += walk.nbytes - nbytes;
466
467 if (walk.nbytes == walk.total && nbytes > 0) {
468 memcpy(block, walk.iv, AES_BLOCK_SIZE);
469 block[0] ^= cpu_to_le32(1 + byte_ctr / AES_BLOCK_SIZE);
470 aesni_enc(ctx, keystream, (u8 *)block);
471 crypto_xor_cpy(walk.dst.virt.addr + walk.nbytes -
472 nbytes, walk.src.virt.addr + walk.nbytes
473 - nbytes, keystream, nbytes);
474 byte_ctr += nbytes;
475 nbytes = 0;
476 }
477 kernel_fpu_end();
478 err = skcipher_walk_done(&walk, nbytes);
479 }
480 return err;
481}
482#endif
483
484static int xts_setkey_aesni(struct crypto_skcipher *tfm, const u8 *key,
485 unsigned int keylen)
486{
487 struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
488 int err;
489
490 err = xts_verify_key(tfm, key, keylen);
491 if (err)
492 return err;
493
494 keylen /= 2;
495
496 /* first half of xts-key is for crypt */
497 err = aes_set_key_common(&ctx->crypt_ctx, key, keylen);
498 if (err)
499 return err;
500
501 /* second half of xts-key is for tweak */
502 return aes_set_key_common(&ctx->tweak_ctx, key + keylen, keylen);
503}
504
505typedef void (*xts_encrypt_iv_func)(const struct crypto_aes_ctx *tweak_key,
506 u8 iv[AES_BLOCK_SIZE]);
507typedef void (*xts_crypt_func)(const struct crypto_aes_ctx *key,
508 const u8 *src, u8 *dst, unsigned int len,
509 u8 tweak[AES_BLOCK_SIZE]);
510
511/* This handles cases where the source and/or destination span pages. */
512static noinline int
513xts_crypt_slowpath(struct skcipher_request *req, xts_crypt_func crypt_func)
514{
515 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
516 const struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
517 int tail = req->cryptlen % AES_BLOCK_SIZE;
518 struct scatterlist sg_src[2], sg_dst[2];
519 struct skcipher_request subreq;
520 struct skcipher_walk walk;
521 struct scatterlist *src, *dst;
522 int err;
523
524 /*
525 * If the message length isn't divisible by the AES block size, then
526 * separate off the last full block and the partial block. This ensures
527 * that they are processed in the same call to the assembly function,
528 * which is required for ciphertext stealing.
529 */
530 if (tail) {
531 skcipher_request_set_tfm(&subreq, tfm);
532 skcipher_request_set_callback(&subreq,
533 skcipher_request_flags(req),
534 NULL, NULL);
535 skcipher_request_set_crypt(&subreq, req->src, req->dst,
536 req->cryptlen - tail - AES_BLOCK_SIZE,
537 req->iv);
538 req = &subreq;
539 }
540
541 err = skcipher_walk_virt(&walk, req, false);
542
543 while (walk.nbytes) {
544 kernel_fpu_begin();
545 (*crypt_func)(&ctx->crypt_ctx,
546 walk.src.virt.addr, walk.dst.virt.addr,
547 walk.nbytes & ~(AES_BLOCK_SIZE - 1), req->iv);
548 kernel_fpu_end();
549 err = skcipher_walk_done(&walk,
550 walk.nbytes & (AES_BLOCK_SIZE - 1));
551 }
552
553 if (err || !tail)
554 return err;
555
556 /* Do ciphertext stealing with the last full block and partial block. */
557
558 dst = src = scatterwalk_ffwd(sg_src, req->src, req->cryptlen);
559 if (req->dst != req->src)
560 dst = scatterwalk_ffwd(sg_dst, req->dst, req->cryptlen);
561
562 skcipher_request_set_crypt(req, src, dst, AES_BLOCK_SIZE + tail,
563 req->iv);
564
565 err = skcipher_walk_virt(&walk, req, false);
566 if (err)
567 return err;
568
569 kernel_fpu_begin();
570 (*crypt_func)(&ctx->crypt_ctx, walk.src.virt.addr, walk.dst.virt.addr,
571 walk.nbytes, req->iv);
572 kernel_fpu_end();
573
574 return skcipher_walk_done(&walk, 0);
575}
576
577/* __always_inline to avoid indirect call in fastpath */
578static __always_inline int
579xts_crypt(struct skcipher_request *req, xts_encrypt_iv_func encrypt_iv,
580 xts_crypt_func crypt_func)
581{
582 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
583 const struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
584 const unsigned int cryptlen = req->cryptlen;
585 struct scatterlist *src = req->src;
586 struct scatterlist *dst = req->dst;
587
588 if (unlikely(cryptlen < AES_BLOCK_SIZE))
589 return -EINVAL;
590
591 kernel_fpu_begin();
592 (*encrypt_iv)(&ctx->tweak_ctx, req->iv);
593
594 /*
595 * In practice, virtually all XTS plaintexts and ciphertexts are either
596 * 512 or 4096 bytes, aligned such that they don't span page boundaries.
597 * To optimize the performance of these cases, and also any other case
598 * where no page boundary is spanned, the below fast-path handles
599 * single-page sources and destinations as efficiently as possible.
600 */
601 if (likely(src->length >= cryptlen && dst->length >= cryptlen &&
602 src->offset + cryptlen <= PAGE_SIZE &&
603 dst->offset + cryptlen <= PAGE_SIZE)) {
604 struct page *src_page = sg_page(src);
605 struct page *dst_page = sg_page(dst);
606 void *src_virt = kmap_local_page(src_page) + src->offset;
607 void *dst_virt = kmap_local_page(dst_page) + dst->offset;
608
609 (*crypt_func)(&ctx->crypt_ctx, src_virt, dst_virt, cryptlen,
610 req->iv);
611 kunmap_local(dst_virt);
612 kunmap_local(src_virt);
613 kernel_fpu_end();
614 return 0;
615 }
616 kernel_fpu_end();
617 return xts_crypt_slowpath(req, crypt_func);
618}
619
620static void aesni_xts_encrypt_iv(const struct crypto_aes_ctx *tweak_key,
621 u8 iv[AES_BLOCK_SIZE])
622{
623 aesni_enc(tweak_key, iv, iv);
624}
625
626static void aesni_xts_encrypt(const struct crypto_aes_ctx *key,
627 const u8 *src, u8 *dst, unsigned int len,
628 u8 tweak[AES_BLOCK_SIZE])
629{
630 aesni_xts_enc(key, dst, src, len, tweak);
631}
632
633static void aesni_xts_decrypt(const struct crypto_aes_ctx *key,
634 const u8 *src, u8 *dst, unsigned int len,
635 u8 tweak[AES_BLOCK_SIZE])
636{
637 aesni_xts_dec(key, dst, src, len, tweak);
638}
639
640static int xts_encrypt_aesni(struct skcipher_request *req)
641{
642 return xts_crypt(req, aesni_xts_encrypt_iv, aesni_xts_encrypt);
643}
644
645static int xts_decrypt_aesni(struct skcipher_request *req)
646{
647 return xts_crypt(req, aesni_xts_encrypt_iv, aesni_xts_decrypt);
648}
649
650static struct crypto_alg aesni_cipher_alg = {
651 .cra_name = "aes",
652 .cra_driver_name = "aes-aesni",
653 .cra_priority = 300,
654 .cra_flags = CRYPTO_ALG_TYPE_CIPHER,
655 .cra_blocksize = AES_BLOCK_SIZE,
656 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
657 .cra_module = THIS_MODULE,
658 .cra_u = {
659 .cipher = {
660 .cia_min_keysize = AES_MIN_KEY_SIZE,
661 .cia_max_keysize = AES_MAX_KEY_SIZE,
662 .cia_setkey = aes_set_key,
663 .cia_encrypt = aesni_encrypt,
664 .cia_decrypt = aesni_decrypt
665 }
666 }
667};
668
669static struct skcipher_alg aesni_skciphers[] = {
670 {
671 .base = {
672 .cra_name = "__ecb(aes)",
673 .cra_driver_name = "__ecb-aes-aesni",
674 .cra_priority = 400,
675 .cra_flags = CRYPTO_ALG_INTERNAL,
676 .cra_blocksize = AES_BLOCK_SIZE,
677 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
678 .cra_module = THIS_MODULE,
679 },
680 .min_keysize = AES_MIN_KEY_SIZE,
681 .max_keysize = AES_MAX_KEY_SIZE,
682 .setkey = aesni_skcipher_setkey,
683 .encrypt = ecb_encrypt,
684 .decrypt = ecb_decrypt,
685 }, {
686 .base = {
687 .cra_name = "__cbc(aes)",
688 .cra_driver_name = "__cbc-aes-aesni",
689 .cra_priority = 400,
690 .cra_flags = CRYPTO_ALG_INTERNAL,
691 .cra_blocksize = AES_BLOCK_SIZE,
692 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
693 .cra_module = THIS_MODULE,
694 },
695 .min_keysize = AES_MIN_KEY_SIZE,
696 .max_keysize = AES_MAX_KEY_SIZE,
697 .ivsize = AES_BLOCK_SIZE,
698 .setkey = aesni_skcipher_setkey,
699 .encrypt = cbc_encrypt,
700 .decrypt = cbc_decrypt,
701 }, {
702 .base = {
703 .cra_name = "__cts(cbc(aes))",
704 .cra_driver_name = "__cts-cbc-aes-aesni",
705 .cra_priority = 400,
706 .cra_flags = CRYPTO_ALG_INTERNAL,
707 .cra_blocksize = AES_BLOCK_SIZE,
708 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
709 .cra_module = THIS_MODULE,
710 },
711 .min_keysize = AES_MIN_KEY_SIZE,
712 .max_keysize = AES_MAX_KEY_SIZE,
713 .ivsize = AES_BLOCK_SIZE,
714 .walksize = 2 * AES_BLOCK_SIZE,
715 .setkey = aesni_skcipher_setkey,
716 .encrypt = cts_cbc_encrypt,
717 .decrypt = cts_cbc_decrypt,
718#ifdef CONFIG_X86_64
719 }, {
720 .base = {
721 .cra_name = "__ctr(aes)",
722 .cra_driver_name = "__ctr-aes-aesni",
723 .cra_priority = 400,
724 .cra_flags = CRYPTO_ALG_INTERNAL,
725 .cra_blocksize = 1,
726 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
727 .cra_module = THIS_MODULE,
728 },
729 .min_keysize = AES_MIN_KEY_SIZE,
730 .max_keysize = AES_MAX_KEY_SIZE,
731 .ivsize = AES_BLOCK_SIZE,
732 .chunksize = AES_BLOCK_SIZE,
733 .setkey = aesni_skcipher_setkey,
734 .encrypt = ctr_crypt,
735 .decrypt = ctr_crypt,
736#endif
737 }, {
738 .base = {
739 .cra_name = "__xts(aes)",
740 .cra_driver_name = "__xts-aes-aesni",
741 .cra_priority = 401,
742 .cra_flags = CRYPTO_ALG_INTERNAL,
743 .cra_blocksize = AES_BLOCK_SIZE,
744 .cra_ctxsize = XTS_AES_CTX_SIZE,
745 .cra_module = THIS_MODULE,
746 },
747 .min_keysize = 2 * AES_MIN_KEY_SIZE,
748 .max_keysize = 2 * AES_MAX_KEY_SIZE,
749 .ivsize = AES_BLOCK_SIZE,
750 .walksize = 2 * AES_BLOCK_SIZE,
751 .setkey = xts_setkey_aesni,
752 .encrypt = xts_encrypt_aesni,
753 .decrypt = xts_decrypt_aesni,
754 }
755};
756
757static
758struct simd_skcipher_alg *aesni_simd_skciphers[ARRAY_SIZE(aesni_skciphers)];
759
760#ifdef CONFIG_X86_64
761/*
762 * XCTR does not have a non-AVX implementation, so it must be enabled
763 * conditionally.
764 */
765static struct skcipher_alg aesni_xctr = {
766 .base = {
767 .cra_name = "__xctr(aes)",
768 .cra_driver_name = "__xctr-aes-aesni",
769 .cra_priority = 400,
770 .cra_flags = CRYPTO_ALG_INTERNAL,
771 .cra_blocksize = 1,
772 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
773 .cra_module = THIS_MODULE,
774 },
775 .min_keysize = AES_MIN_KEY_SIZE,
776 .max_keysize = AES_MAX_KEY_SIZE,
777 .ivsize = AES_BLOCK_SIZE,
778 .chunksize = AES_BLOCK_SIZE,
779 .setkey = aesni_skcipher_setkey,
780 .encrypt = xctr_crypt,
781 .decrypt = xctr_crypt,
782};
783
784static struct simd_skcipher_alg *aesni_simd_xctr;
785
786asmlinkage void aes_xts_encrypt_iv(const struct crypto_aes_ctx *tweak_key,
787 u8 iv[AES_BLOCK_SIZE]);
788
789#define DEFINE_XTS_ALG(suffix, driver_name, priority) \
790 \
791asmlinkage void \
792aes_xts_encrypt_##suffix(const struct crypto_aes_ctx *key, const u8 *src, \
793 u8 *dst, unsigned int len, u8 tweak[AES_BLOCK_SIZE]); \
794asmlinkage void \
795aes_xts_decrypt_##suffix(const struct crypto_aes_ctx *key, const u8 *src, \
796 u8 *dst, unsigned int len, u8 tweak[AES_BLOCK_SIZE]); \
797 \
798static int xts_encrypt_##suffix(struct skcipher_request *req) \
799{ \
800 return xts_crypt(req, aes_xts_encrypt_iv, aes_xts_encrypt_##suffix); \
801} \
802 \
803static int xts_decrypt_##suffix(struct skcipher_request *req) \
804{ \
805 return xts_crypt(req, aes_xts_encrypt_iv, aes_xts_decrypt_##suffix); \
806} \
807 \
808static struct skcipher_alg aes_xts_alg_##suffix = { \
809 .base = { \
810 .cra_name = "__xts(aes)", \
811 .cra_driver_name = "__" driver_name, \
812 .cra_priority = priority, \
813 .cra_flags = CRYPTO_ALG_INTERNAL, \
814 .cra_blocksize = AES_BLOCK_SIZE, \
815 .cra_ctxsize = XTS_AES_CTX_SIZE, \
816 .cra_module = THIS_MODULE, \
817 }, \
818 .min_keysize = 2 * AES_MIN_KEY_SIZE, \
819 .max_keysize = 2 * AES_MAX_KEY_SIZE, \
820 .ivsize = AES_BLOCK_SIZE, \
821 .walksize = 2 * AES_BLOCK_SIZE, \
822 .setkey = xts_setkey_aesni, \
823 .encrypt = xts_encrypt_##suffix, \
824 .decrypt = xts_decrypt_##suffix, \
825}; \
826 \
827static struct simd_skcipher_alg *aes_xts_simdalg_##suffix
828
829DEFINE_XTS_ALG(aesni_avx, "xts-aes-aesni-avx", 500);
830#if defined(CONFIG_AS_VAES) && defined(CONFIG_AS_VPCLMULQDQ)
831DEFINE_XTS_ALG(vaes_avx2, "xts-aes-vaes-avx2", 600);
832DEFINE_XTS_ALG(vaes_avx10_256, "xts-aes-vaes-avx10_256", 700);
833DEFINE_XTS_ALG(vaes_avx10_512, "xts-aes-vaes-avx10_512", 800);
834#endif
835
836/* The common part of the x86_64 AES-GCM key struct */
837struct aes_gcm_key {
838 /* Expanded AES key and the AES key length in bytes */
839 struct crypto_aes_ctx aes_key;
840
841 /* RFC4106 nonce (used only by the rfc4106 algorithms) */
842 u32 rfc4106_nonce;
843};
844
845/* Key struct used by the AES-NI implementations of AES-GCM */
846struct aes_gcm_key_aesni {
847 /*
848 * Common part of the key. The assembly code requires 16-byte alignment
849 * for the round keys; we get this by them being located at the start of
850 * the struct and the whole struct being 16-byte aligned.
851 */
852 struct aes_gcm_key base;
853
854 /*
855 * Powers of the hash key H^8 through H^1. These are 128-bit values.
856 * They all have an extra factor of x^-1 and are byte-reversed. 16-byte
857 * alignment is required by the assembly code.
858 */
859 u64 h_powers[8][2] __aligned(16);
860
861 /*
862 * h_powers_xored[i] contains the two 64-bit halves of h_powers[i] XOR'd
863 * together. It's used for Karatsuba multiplication. 16-byte alignment
864 * is required by the assembly code.
865 */
866 u64 h_powers_xored[8] __aligned(16);
867
868 /*
869 * H^1 times x^64 (and also the usual extra factor of x^-1). 16-byte
870 * alignment is required by the assembly code.
871 */
872 u64 h_times_x64[2] __aligned(16);
873};
874#define AES_GCM_KEY_AESNI(key) \
875 container_of((key), struct aes_gcm_key_aesni, base)
876#define AES_GCM_KEY_AESNI_SIZE \
877 (sizeof(struct aes_gcm_key_aesni) + (15 & ~(CRYPTO_MINALIGN - 1)))
878
879/* Key struct used by the VAES + AVX10 implementations of AES-GCM */
880struct aes_gcm_key_avx10 {
881 /*
882 * Common part of the key. The assembly code prefers 16-byte alignment
883 * for the round keys; we get this by them being located at the start of
884 * the struct and the whole struct being 64-byte aligned.
885 */
886 struct aes_gcm_key base;
887
888 /*
889 * Powers of the hash key H^16 through H^1. These are 128-bit values.
890 * They all have an extra factor of x^-1 and are byte-reversed. This
891 * array is aligned to a 64-byte boundary to make it naturally aligned
892 * for 512-bit loads, which can improve performance. (The assembly code
893 * doesn't *need* the alignment; this is just an optimization.)
894 */
895 u64 h_powers[16][2] __aligned(64);
896
897 /* Three padding blocks required by the assembly code */
898 u64 padding[3][2];
899};
900#define AES_GCM_KEY_AVX10(key) \
901 container_of((key), struct aes_gcm_key_avx10, base)
902#define AES_GCM_KEY_AVX10_SIZE \
903 (sizeof(struct aes_gcm_key_avx10) + (63 & ~(CRYPTO_MINALIGN - 1)))
904
905/*
906 * These flags are passed to the AES-GCM helper functions to specify the
907 * specific version of AES-GCM (RFC4106 or not), whether it's encryption or
908 * decryption, and which assembly functions should be called. Assembly
909 * functions are selected using flags instead of function pointers to avoid
910 * indirect calls (which are very expensive on x86) regardless of inlining.
911 */
912#define FLAG_RFC4106 BIT(0)
913#define FLAG_ENC BIT(1)
914#define FLAG_AVX BIT(2)
915#if defined(CONFIG_AS_VAES) && defined(CONFIG_AS_VPCLMULQDQ)
916# define FLAG_AVX10_256 BIT(3)
917# define FLAG_AVX10_512 BIT(4)
918#else
919 /*
920 * This should cause all calls to the AVX10 assembly functions to be
921 * optimized out, avoiding the need to ifdef each call individually.
922 */
923# define FLAG_AVX10_256 0
924# define FLAG_AVX10_512 0
925#endif
926
927static inline struct aes_gcm_key *
928aes_gcm_key_get(struct crypto_aead *tfm, int flags)
929{
930 if (flags & (FLAG_AVX10_256 | FLAG_AVX10_512))
931 return PTR_ALIGN(crypto_aead_ctx(tfm), 64);
932 else
933 return PTR_ALIGN(crypto_aead_ctx(tfm), 16);
934}
935
936asmlinkage void
937aes_gcm_precompute_aesni(struct aes_gcm_key_aesni *key);
938asmlinkage void
939aes_gcm_precompute_aesni_avx(struct aes_gcm_key_aesni *key);
940asmlinkage void
941aes_gcm_precompute_vaes_avx10_256(struct aes_gcm_key_avx10 *key);
942asmlinkage void
943aes_gcm_precompute_vaes_avx10_512(struct aes_gcm_key_avx10 *key);
944
945static void aes_gcm_precompute(struct aes_gcm_key *key, int flags)
946{
947 /*
948 * To make things a bit easier on the assembly side, the AVX10
949 * implementations use the same key format. Therefore, a single
950 * function using 256-bit vectors would suffice here. However, it's
951 * straightforward to provide a 512-bit one because of how the assembly
952 * code is structured, and it works nicely because the total size of the
953 * key powers is a multiple of 512 bits. So we take advantage of that.
954 *
955 * A similar situation applies to the AES-NI implementations.
956 */
957 if (flags & FLAG_AVX10_512)
958 aes_gcm_precompute_vaes_avx10_512(AES_GCM_KEY_AVX10(key));
959 else if (flags & FLAG_AVX10_256)
960 aes_gcm_precompute_vaes_avx10_256(AES_GCM_KEY_AVX10(key));
961 else if (flags & FLAG_AVX)
962 aes_gcm_precompute_aesni_avx(AES_GCM_KEY_AESNI(key));
963 else
964 aes_gcm_precompute_aesni(AES_GCM_KEY_AESNI(key));
965}
966
967asmlinkage void
968aes_gcm_aad_update_aesni(const struct aes_gcm_key_aesni *key,
969 u8 ghash_acc[16], const u8 *aad, int aadlen);
970asmlinkage void
971aes_gcm_aad_update_aesni_avx(const struct aes_gcm_key_aesni *key,
972 u8 ghash_acc[16], const u8 *aad, int aadlen);
973asmlinkage void
974aes_gcm_aad_update_vaes_avx10(const struct aes_gcm_key_avx10 *key,
975 u8 ghash_acc[16], const u8 *aad, int aadlen);
976
977static void aes_gcm_aad_update(const struct aes_gcm_key *key, u8 ghash_acc[16],
978 const u8 *aad, int aadlen, int flags)
979{
980 if (flags & (FLAG_AVX10_256 | FLAG_AVX10_512))
981 aes_gcm_aad_update_vaes_avx10(AES_GCM_KEY_AVX10(key), ghash_acc,
982 aad, aadlen);
983 else if (flags & FLAG_AVX)
984 aes_gcm_aad_update_aesni_avx(AES_GCM_KEY_AESNI(key), ghash_acc,
985 aad, aadlen);
986 else
987 aes_gcm_aad_update_aesni(AES_GCM_KEY_AESNI(key), ghash_acc,
988 aad, aadlen);
989}
990
991asmlinkage void
992aes_gcm_enc_update_aesni(const struct aes_gcm_key_aesni *key,
993 const u32 le_ctr[4], u8 ghash_acc[16],
994 const u8 *src, u8 *dst, int datalen);
995asmlinkage void
996aes_gcm_enc_update_aesni_avx(const struct aes_gcm_key_aesni *key,
997 const u32 le_ctr[4], u8 ghash_acc[16],
998 const u8 *src, u8 *dst, int datalen);
999asmlinkage void
1000aes_gcm_enc_update_vaes_avx10_256(const struct aes_gcm_key_avx10 *key,
1001 const u32 le_ctr[4], u8 ghash_acc[16],
1002 const u8 *src, u8 *dst, int datalen);
1003asmlinkage void
1004aes_gcm_enc_update_vaes_avx10_512(const struct aes_gcm_key_avx10 *key,
1005 const u32 le_ctr[4], u8 ghash_acc[16],
1006 const u8 *src, u8 *dst, int datalen);
1007
1008asmlinkage void
1009aes_gcm_dec_update_aesni(const struct aes_gcm_key_aesni *key,
1010 const u32 le_ctr[4], u8 ghash_acc[16],
1011 const u8 *src, u8 *dst, int datalen);
1012asmlinkage void
1013aes_gcm_dec_update_aesni_avx(const struct aes_gcm_key_aesni *key,
1014 const u32 le_ctr[4], u8 ghash_acc[16],
1015 const u8 *src, u8 *dst, int datalen);
1016asmlinkage void
1017aes_gcm_dec_update_vaes_avx10_256(const struct aes_gcm_key_avx10 *key,
1018 const u32 le_ctr[4], u8 ghash_acc[16],
1019 const u8 *src, u8 *dst, int datalen);
1020asmlinkage void
1021aes_gcm_dec_update_vaes_avx10_512(const struct aes_gcm_key_avx10 *key,
1022 const u32 le_ctr[4], u8 ghash_acc[16],
1023 const u8 *src, u8 *dst, int datalen);
1024
1025/* __always_inline to optimize out the branches based on @flags */
1026static __always_inline void
1027aes_gcm_update(const struct aes_gcm_key *key,
1028 const u32 le_ctr[4], u8 ghash_acc[16],
1029 const u8 *src, u8 *dst, int datalen, int flags)
1030{
1031 if (flags & FLAG_ENC) {
1032 if (flags & FLAG_AVX10_512)
1033 aes_gcm_enc_update_vaes_avx10_512(AES_GCM_KEY_AVX10(key),
1034 le_ctr, ghash_acc,
1035 src, dst, datalen);
1036 else if (flags & FLAG_AVX10_256)
1037 aes_gcm_enc_update_vaes_avx10_256(AES_GCM_KEY_AVX10(key),
1038 le_ctr, ghash_acc,
1039 src, dst, datalen);
1040 else if (flags & FLAG_AVX)
1041 aes_gcm_enc_update_aesni_avx(AES_GCM_KEY_AESNI(key),
1042 le_ctr, ghash_acc,
1043 src, dst, datalen);
1044 else
1045 aes_gcm_enc_update_aesni(AES_GCM_KEY_AESNI(key), le_ctr,
1046 ghash_acc, src, dst, datalen);
1047 } else {
1048 if (flags & FLAG_AVX10_512)
1049 aes_gcm_dec_update_vaes_avx10_512(AES_GCM_KEY_AVX10(key),
1050 le_ctr, ghash_acc,
1051 src, dst, datalen);
1052 else if (flags & FLAG_AVX10_256)
1053 aes_gcm_dec_update_vaes_avx10_256(AES_GCM_KEY_AVX10(key),
1054 le_ctr, ghash_acc,
1055 src, dst, datalen);
1056 else if (flags & FLAG_AVX)
1057 aes_gcm_dec_update_aesni_avx(AES_GCM_KEY_AESNI(key),
1058 le_ctr, ghash_acc,
1059 src, dst, datalen);
1060 else
1061 aes_gcm_dec_update_aesni(AES_GCM_KEY_AESNI(key),
1062 le_ctr, ghash_acc,
1063 src, dst, datalen);
1064 }
1065}
1066
1067asmlinkage void
1068aes_gcm_enc_final_aesni(const struct aes_gcm_key_aesni *key,
1069 const u32 le_ctr[4], u8 ghash_acc[16],
1070 u64 total_aadlen, u64 total_datalen);
1071asmlinkage void
1072aes_gcm_enc_final_aesni_avx(const struct aes_gcm_key_aesni *key,
1073 const u32 le_ctr[4], u8 ghash_acc[16],
1074 u64 total_aadlen, u64 total_datalen);
1075asmlinkage void
1076aes_gcm_enc_final_vaes_avx10(const struct aes_gcm_key_avx10 *key,
1077 const u32 le_ctr[4], u8 ghash_acc[16],
1078 u64 total_aadlen, u64 total_datalen);
1079
1080/* __always_inline to optimize out the branches based on @flags */
1081static __always_inline void
1082aes_gcm_enc_final(const struct aes_gcm_key *key,
1083 const u32 le_ctr[4], u8 ghash_acc[16],
1084 u64 total_aadlen, u64 total_datalen, int flags)
1085{
1086 if (flags & (FLAG_AVX10_256 | FLAG_AVX10_512))
1087 aes_gcm_enc_final_vaes_avx10(AES_GCM_KEY_AVX10(key),
1088 le_ctr, ghash_acc,
1089 total_aadlen, total_datalen);
1090 else if (flags & FLAG_AVX)
1091 aes_gcm_enc_final_aesni_avx(AES_GCM_KEY_AESNI(key),
1092 le_ctr, ghash_acc,
1093 total_aadlen, total_datalen);
1094 else
1095 aes_gcm_enc_final_aesni(AES_GCM_KEY_AESNI(key),
1096 le_ctr, ghash_acc,
1097 total_aadlen, total_datalen);
1098}
1099
1100asmlinkage bool __must_check
1101aes_gcm_dec_final_aesni(const struct aes_gcm_key_aesni *key,
1102 const u32 le_ctr[4], const u8 ghash_acc[16],
1103 u64 total_aadlen, u64 total_datalen,
1104 const u8 tag[16], int taglen);
1105asmlinkage bool __must_check
1106aes_gcm_dec_final_aesni_avx(const struct aes_gcm_key_aesni *key,
1107 const u32 le_ctr[4], const u8 ghash_acc[16],
1108 u64 total_aadlen, u64 total_datalen,
1109 const u8 tag[16], int taglen);
1110asmlinkage bool __must_check
1111aes_gcm_dec_final_vaes_avx10(const struct aes_gcm_key_avx10 *key,
1112 const u32 le_ctr[4], const u8 ghash_acc[16],
1113 u64 total_aadlen, u64 total_datalen,
1114 const u8 tag[16], int taglen);
1115
1116/* __always_inline to optimize out the branches based on @flags */
1117static __always_inline bool __must_check
1118aes_gcm_dec_final(const struct aes_gcm_key *key, const u32 le_ctr[4],
1119 u8 ghash_acc[16], u64 total_aadlen, u64 total_datalen,
1120 u8 tag[16], int taglen, int flags)
1121{
1122 if (flags & (FLAG_AVX10_256 | FLAG_AVX10_512))
1123 return aes_gcm_dec_final_vaes_avx10(AES_GCM_KEY_AVX10(key),
1124 le_ctr, ghash_acc,
1125 total_aadlen, total_datalen,
1126 tag, taglen);
1127 else if (flags & FLAG_AVX)
1128 return aes_gcm_dec_final_aesni_avx(AES_GCM_KEY_AESNI(key),
1129 le_ctr, ghash_acc,
1130 total_aadlen, total_datalen,
1131 tag, taglen);
1132 else
1133 return aes_gcm_dec_final_aesni(AES_GCM_KEY_AESNI(key),
1134 le_ctr, ghash_acc,
1135 total_aadlen, total_datalen,
1136 tag, taglen);
1137}
1138
1139/*
1140 * This is the Integrity Check Value (aka the authentication tag) length and can
1141 * be 8, 12 or 16 bytes long.
1142 */
1143static int common_rfc4106_set_authsize(struct crypto_aead *aead,
1144 unsigned int authsize)
1145{
1146 switch (authsize) {
1147 case 8:
1148 case 12:
1149 case 16:
1150 break;
1151 default:
1152 return -EINVAL;
1153 }
1154
1155 return 0;
1156}
1157
1158static int generic_gcmaes_set_authsize(struct crypto_aead *tfm,
1159 unsigned int authsize)
1160{
1161 switch (authsize) {
1162 case 4:
1163 case 8:
1164 case 12:
1165 case 13:
1166 case 14:
1167 case 15:
1168 case 16:
1169 break;
1170 default:
1171 return -EINVAL;
1172 }
1173
1174 return 0;
1175}
1176
1177/*
1178 * This is the setkey function for the x86_64 implementations of AES-GCM. It
1179 * saves the RFC4106 nonce if applicable, expands the AES key, and precomputes
1180 * powers of the hash key.
1181 *
1182 * To comply with the crypto_aead API, this has to be usable in no-SIMD context.
1183 * For that reason, this function includes a portable C implementation of the
1184 * needed logic. However, the portable C implementation is very slow, taking
1185 * about the same time as encrypting 37 KB of data. To be ready for users that
1186 * may set a key even somewhat frequently, we therefore also include a SIMD
1187 * assembly implementation, expanding the AES key using AES-NI and precomputing
1188 * the hash key powers using PCLMULQDQ or VPCLMULQDQ.
1189 */
1190static int gcm_setkey(struct crypto_aead *tfm, const u8 *raw_key,
1191 unsigned int keylen, int flags)
1192{
1193 struct aes_gcm_key *key = aes_gcm_key_get(tfm, flags);
1194 int err;
1195
1196 if (flags & FLAG_RFC4106) {
1197 if (keylen < 4)
1198 return -EINVAL;
1199 keylen -= 4;
1200 key->rfc4106_nonce = get_unaligned_be32(raw_key + keylen);
1201 }
1202
1203 /* The assembly code assumes the following offsets. */
1204 BUILD_BUG_ON(offsetof(struct aes_gcm_key_aesni, base.aes_key.key_enc) != 0);
1205 BUILD_BUG_ON(offsetof(struct aes_gcm_key_aesni, base.aes_key.key_length) != 480);
1206 BUILD_BUG_ON(offsetof(struct aes_gcm_key_aesni, h_powers) != 496);
1207 BUILD_BUG_ON(offsetof(struct aes_gcm_key_aesni, h_powers_xored) != 624);
1208 BUILD_BUG_ON(offsetof(struct aes_gcm_key_aesni, h_times_x64) != 688);
1209 BUILD_BUG_ON(offsetof(struct aes_gcm_key_avx10, base.aes_key.key_enc) != 0);
1210 BUILD_BUG_ON(offsetof(struct aes_gcm_key_avx10, base.aes_key.key_length) != 480);
1211 BUILD_BUG_ON(offsetof(struct aes_gcm_key_avx10, h_powers) != 512);
1212 BUILD_BUG_ON(offsetof(struct aes_gcm_key_avx10, padding) != 768);
1213
1214 if (likely(crypto_simd_usable())) {
1215 err = aes_check_keylen(keylen);
1216 if (err)
1217 return err;
1218 kernel_fpu_begin();
1219 aesni_set_key(&key->aes_key, raw_key, keylen);
1220 aes_gcm_precompute(key, flags);
1221 kernel_fpu_end();
1222 } else {
1223 static const u8 x_to_the_minus1[16] __aligned(__alignof__(be128)) = {
1224 [0] = 0xc2, [15] = 1
1225 };
1226 static const u8 x_to_the_63[16] __aligned(__alignof__(be128)) = {
1227 [7] = 1,
1228 };
1229 be128 h1 = {};
1230 be128 h;
1231 int i;
1232
1233 err = aes_expandkey(&key->aes_key, raw_key, keylen);
1234 if (err)
1235 return err;
1236
1237 /* Encrypt the all-zeroes block to get the hash key H^1 */
1238 aes_encrypt(&key->aes_key, (u8 *)&h1, (u8 *)&h1);
1239
1240 /* Compute H^1 * x^-1 */
1241 h = h1;
1242 gf128mul_lle(&h, (const be128 *)x_to_the_minus1);
1243
1244 /* Compute the needed key powers */
1245 if (flags & (FLAG_AVX10_256 | FLAG_AVX10_512)) {
1246 struct aes_gcm_key_avx10 *k = AES_GCM_KEY_AVX10(key);
1247
1248 for (i = ARRAY_SIZE(k->h_powers) - 1; i >= 0; i--) {
1249 k->h_powers[i][0] = be64_to_cpu(h.b);
1250 k->h_powers[i][1] = be64_to_cpu(h.a);
1251 gf128mul_lle(&h, &h1);
1252 }
1253 memset(k->padding, 0, sizeof(k->padding));
1254 } else {
1255 struct aes_gcm_key_aesni *k = AES_GCM_KEY_AESNI(key);
1256
1257 for (i = ARRAY_SIZE(k->h_powers) - 1; i >= 0; i--) {
1258 k->h_powers[i][0] = be64_to_cpu(h.b);
1259 k->h_powers[i][1] = be64_to_cpu(h.a);
1260 k->h_powers_xored[i] = k->h_powers[i][0] ^
1261 k->h_powers[i][1];
1262 gf128mul_lle(&h, &h1);
1263 }
1264 gf128mul_lle(&h1, (const be128 *)x_to_the_63);
1265 k->h_times_x64[0] = be64_to_cpu(h1.b);
1266 k->h_times_x64[1] = be64_to_cpu(h1.a);
1267 }
1268 }
1269 return 0;
1270}
1271
1272/*
1273 * Initialize @ghash_acc, then pass all @assoclen bytes of associated data
1274 * (a.k.a. additional authenticated data) from @sg_src through the GHASH update
1275 * assembly function. kernel_fpu_begin() must have already been called.
1276 */
1277static void gcm_process_assoc(const struct aes_gcm_key *key, u8 ghash_acc[16],
1278 struct scatterlist *sg_src, unsigned int assoclen,
1279 int flags)
1280{
1281 struct scatter_walk walk;
1282 /*
1283 * The assembly function requires that the length of any non-last
1284 * segment of associated data be a multiple of 16 bytes, so this
1285 * function does the buffering needed to achieve that.
1286 */
1287 unsigned int pos = 0;
1288 u8 buf[16];
1289
1290 memset(ghash_acc, 0, 16);
1291 scatterwalk_start(&walk, sg_src);
1292
1293 while (assoclen) {
1294 unsigned int len_this_page = scatterwalk_clamp(&walk, assoclen);
1295 void *mapped = scatterwalk_map(&walk);
1296 const void *src = mapped;
1297 unsigned int len;
1298
1299 assoclen -= len_this_page;
1300 scatterwalk_advance(&walk, len_this_page);
1301 if (unlikely(pos)) {
1302 len = min(len_this_page, 16 - pos);
1303 memcpy(&buf[pos], src, len);
1304 pos += len;
1305 src += len;
1306 len_this_page -= len;
1307 if (pos < 16)
1308 goto next;
1309 aes_gcm_aad_update(key, ghash_acc, buf, 16, flags);
1310 pos = 0;
1311 }
1312 len = len_this_page;
1313 if (unlikely(assoclen)) /* Not the last segment yet? */
1314 len = round_down(len, 16);
1315 aes_gcm_aad_update(key, ghash_acc, src, len, flags);
1316 src += len;
1317 len_this_page -= len;
1318 if (unlikely(len_this_page)) {
1319 memcpy(buf, src, len_this_page);
1320 pos = len_this_page;
1321 }
1322next:
1323 scatterwalk_unmap(mapped);
1324 scatterwalk_pagedone(&walk, 0, assoclen);
1325 if (need_resched()) {
1326 kernel_fpu_end();
1327 kernel_fpu_begin();
1328 }
1329 }
1330 if (unlikely(pos))
1331 aes_gcm_aad_update(key, ghash_acc, buf, pos, flags);
1332}
1333
1334
1335/* __always_inline to optimize out the branches based on @flags */
1336static __always_inline int
1337gcm_crypt(struct aead_request *req, int flags)
1338{
1339 struct crypto_aead *tfm = crypto_aead_reqtfm(req);
1340 const struct aes_gcm_key *key = aes_gcm_key_get(tfm, flags);
1341 unsigned int assoclen = req->assoclen;
1342 struct skcipher_walk walk;
1343 unsigned int nbytes;
1344 u8 ghash_acc[16]; /* GHASH accumulator */
1345 u32 le_ctr[4]; /* Counter in little-endian format */
1346 int taglen;
1347 int err;
1348
1349 /* Initialize the counter and determine the associated data length. */
1350 le_ctr[0] = 2;
1351 if (flags & FLAG_RFC4106) {
1352 if (unlikely(assoclen != 16 && assoclen != 20))
1353 return -EINVAL;
1354 assoclen -= 8;
1355 le_ctr[1] = get_unaligned_be32(req->iv + 4);
1356 le_ctr[2] = get_unaligned_be32(req->iv + 0);
1357 le_ctr[3] = key->rfc4106_nonce; /* already byte-swapped */
1358 } else {
1359 le_ctr[1] = get_unaligned_be32(req->iv + 8);
1360 le_ctr[2] = get_unaligned_be32(req->iv + 4);
1361 le_ctr[3] = get_unaligned_be32(req->iv + 0);
1362 }
1363
1364 /* Begin walking through the plaintext or ciphertext. */
1365 if (flags & FLAG_ENC)
1366 err = skcipher_walk_aead_encrypt(&walk, req, false);
1367 else
1368 err = skcipher_walk_aead_decrypt(&walk, req, false);
1369 if (err)
1370 return err;
1371
1372 /*
1373 * Since the AES-GCM assembly code requires that at least three assembly
1374 * functions be called to process any message (this is needed to support
1375 * incremental updates cleanly), to reduce overhead we try to do all
1376 * three calls in the same kernel FPU section if possible. We close the
1377 * section and start a new one if there are multiple data segments or if
1378 * rescheduling is needed while processing the associated data.
1379 */
1380 kernel_fpu_begin();
1381
1382 /* Pass the associated data through GHASH. */
1383 gcm_process_assoc(key, ghash_acc, req->src, assoclen, flags);
1384
1385 /* En/decrypt the data and pass the ciphertext through GHASH. */
1386 while (unlikely((nbytes = walk.nbytes) < walk.total)) {
1387 /*
1388 * Non-last segment. In this case, the assembly function
1389 * requires that the length be a multiple of 16 (AES_BLOCK_SIZE)
1390 * bytes. The needed buffering of up to 16 bytes is handled by
1391 * the skcipher_walk. Here we just need to round down to a
1392 * multiple of 16.
1393 */
1394 nbytes = round_down(nbytes, AES_BLOCK_SIZE);
1395 aes_gcm_update(key, le_ctr, ghash_acc, walk.src.virt.addr,
1396 walk.dst.virt.addr, nbytes, flags);
1397 le_ctr[0] += nbytes / AES_BLOCK_SIZE;
1398 kernel_fpu_end();
1399 err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
1400 if (err)
1401 return err;
1402 kernel_fpu_begin();
1403 }
1404 /* Last segment: process all remaining data. */
1405 aes_gcm_update(key, le_ctr, ghash_acc, walk.src.virt.addr,
1406 walk.dst.virt.addr, nbytes, flags);
1407 /*
1408 * The low word of the counter isn't used by the finalize, so there's no
1409 * need to increment it here.
1410 */
1411
1412 /* Finalize */
1413 taglen = crypto_aead_authsize(tfm);
1414 if (flags & FLAG_ENC) {
1415 /* Finish computing the auth tag. */
1416 aes_gcm_enc_final(key, le_ctr, ghash_acc, assoclen,
1417 req->cryptlen, flags);
1418
1419 /* Store the computed auth tag in the dst scatterlist. */
1420 scatterwalk_map_and_copy(ghash_acc, req->dst, req->assoclen +
1421 req->cryptlen, taglen, 1);
1422 } else {
1423 unsigned int datalen = req->cryptlen - taglen;
1424 u8 tag[16];
1425
1426 /* Get the transmitted auth tag from the src scatterlist. */
1427 scatterwalk_map_and_copy(tag, req->src, req->assoclen + datalen,
1428 taglen, 0);
1429 /*
1430 * Finish computing the auth tag and compare it to the
1431 * transmitted one. The assembly function does the actual tag
1432 * comparison. Here, just check the boolean result.
1433 */
1434 if (!aes_gcm_dec_final(key, le_ctr, ghash_acc, assoclen,
1435 datalen, tag, taglen, flags))
1436 err = -EBADMSG;
1437 }
1438 kernel_fpu_end();
1439 if (nbytes)
1440 skcipher_walk_done(&walk, 0);
1441 return err;
1442}
1443
1444#define DEFINE_GCM_ALGS(suffix, flags, generic_driver_name, rfc_driver_name, \
1445 ctxsize, priority) \
1446 \
1447static int gcm_setkey_##suffix(struct crypto_aead *tfm, const u8 *raw_key, \
1448 unsigned int keylen) \
1449{ \
1450 return gcm_setkey(tfm, raw_key, keylen, (flags)); \
1451} \
1452 \
1453static int gcm_encrypt_##suffix(struct aead_request *req) \
1454{ \
1455 return gcm_crypt(req, (flags) | FLAG_ENC); \
1456} \
1457 \
1458static int gcm_decrypt_##suffix(struct aead_request *req) \
1459{ \
1460 return gcm_crypt(req, (flags)); \
1461} \
1462 \
1463static int rfc4106_setkey_##suffix(struct crypto_aead *tfm, const u8 *raw_key, \
1464 unsigned int keylen) \
1465{ \
1466 return gcm_setkey(tfm, raw_key, keylen, (flags) | FLAG_RFC4106); \
1467} \
1468 \
1469static int rfc4106_encrypt_##suffix(struct aead_request *req) \
1470{ \
1471 return gcm_crypt(req, (flags) | FLAG_RFC4106 | FLAG_ENC); \
1472} \
1473 \
1474static int rfc4106_decrypt_##suffix(struct aead_request *req) \
1475{ \
1476 return gcm_crypt(req, (flags) | FLAG_RFC4106); \
1477} \
1478 \
1479static struct aead_alg aes_gcm_algs_##suffix[] = { { \
1480 .setkey = gcm_setkey_##suffix, \
1481 .setauthsize = generic_gcmaes_set_authsize, \
1482 .encrypt = gcm_encrypt_##suffix, \
1483 .decrypt = gcm_decrypt_##suffix, \
1484 .ivsize = GCM_AES_IV_SIZE, \
1485 .chunksize = AES_BLOCK_SIZE, \
1486 .maxauthsize = 16, \
1487 .base = { \
1488 .cra_name = "__gcm(aes)", \
1489 .cra_driver_name = "__" generic_driver_name, \
1490 .cra_priority = (priority), \
1491 .cra_flags = CRYPTO_ALG_INTERNAL, \
1492 .cra_blocksize = 1, \
1493 .cra_ctxsize = (ctxsize), \
1494 .cra_module = THIS_MODULE, \
1495 }, \
1496}, { \
1497 .setkey = rfc4106_setkey_##suffix, \
1498 .setauthsize = common_rfc4106_set_authsize, \
1499 .encrypt = rfc4106_encrypt_##suffix, \
1500 .decrypt = rfc4106_decrypt_##suffix, \
1501 .ivsize = GCM_RFC4106_IV_SIZE, \
1502 .chunksize = AES_BLOCK_SIZE, \
1503 .maxauthsize = 16, \
1504 .base = { \
1505 .cra_name = "__rfc4106(gcm(aes))", \
1506 .cra_driver_name = "__" rfc_driver_name, \
1507 .cra_priority = (priority), \
1508 .cra_flags = CRYPTO_ALG_INTERNAL, \
1509 .cra_blocksize = 1, \
1510 .cra_ctxsize = (ctxsize), \
1511 .cra_module = THIS_MODULE, \
1512 }, \
1513} }; \
1514 \
1515static struct simd_aead_alg *aes_gcm_simdalgs_##suffix[2] \
1516
1517/* aes_gcm_algs_aesni */
1518DEFINE_GCM_ALGS(aesni, /* no flags */ 0,
1519 "generic-gcm-aesni", "rfc4106-gcm-aesni",
1520 AES_GCM_KEY_AESNI_SIZE, 400);
1521
1522/* aes_gcm_algs_aesni_avx */
1523DEFINE_GCM_ALGS(aesni_avx, FLAG_AVX,
1524 "generic-gcm-aesni-avx", "rfc4106-gcm-aesni-avx",
1525 AES_GCM_KEY_AESNI_SIZE, 500);
1526
1527#if defined(CONFIG_AS_VAES) && defined(CONFIG_AS_VPCLMULQDQ)
1528/* aes_gcm_algs_vaes_avx10_256 */
1529DEFINE_GCM_ALGS(vaes_avx10_256, FLAG_AVX10_256,
1530 "generic-gcm-vaes-avx10_256", "rfc4106-gcm-vaes-avx10_256",
1531 AES_GCM_KEY_AVX10_SIZE, 700);
1532
1533/* aes_gcm_algs_vaes_avx10_512 */
1534DEFINE_GCM_ALGS(vaes_avx10_512, FLAG_AVX10_512,
1535 "generic-gcm-vaes-avx10_512", "rfc4106-gcm-vaes-avx10_512",
1536 AES_GCM_KEY_AVX10_SIZE, 800);
1537#endif /* CONFIG_AS_VAES && CONFIG_AS_VPCLMULQDQ */
1538
1539/*
1540 * This is a list of CPU models that are known to suffer from downclocking when
1541 * zmm registers (512-bit vectors) are used. On these CPUs, the AES mode
1542 * implementations with zmm registers won't be used by default. Implementations
1543 * with ymm registers (256-bit vectors) will be used by default instead.
1544 */
1545static const struct x86_cpu_id zmm_exclusion_list[] = {
1546 X86_MATCH_VFM(INTEL_SKYLAKE_X, 0),
1547 X86_MATCH_VFM(INTEL_ICELAKE_X, 0),
1548 X86_MATCH_VFM(INTEL_ICELAKE_D, 0),
1549 X86_MATCH_VFM(INTEL_ICELAKE, 0),
1550 X86_MATCH_VFM(INTEL_ICELAKE_L, 0),
1551 X86_MATCH_VFM(INTEL_ICELAKE_NNPI, 0),
1552 X86_MATCH_VFM(INTEL_TIGERLAKE_L, 0),
1553 X86_MATCH_VFM(INTEL_TIGERLAKE, 0),
1554 /* Allow Rocket Lake and later, and Sapphire Rapids and later. */
1555 /* Also allow AMD CPUs (starting with Zen 4, the first with AVX-512). */
1556 {},
1557};
1558
1559static int __init register_avx_algs(void)
1560{
1561 int err;
1562
1563 if (!boot_cpu_has(X86_FEATURE_AVX))
1564 return 0;
1565 err = simd_register_skciphers_compat(&aes_xts_alg_aesni_avx, 1,
1566 &aes_xts_simdalg_aesni_avx);
1567 if (err)
1568 return err;
1569 err = simd_register_aeads_compat(aes_gcm_algs_aesni_avx,
1570 ARRAY_SIZE(aes_gcm_algs_aesni_avx),
1571 aes_gcm_simdalgs_aesni_avx);
1572 if (err)
1573 return err;
1574#if defined(CONFIG_AS_VAES) && defined(CONFIG_AS_VPCLMULQDQ)
1575 if (!boot_cpu_has(X86_FEATURE_AVX2) ||
1576 !boot_cpu_has(X86_FEATURE_VAES) ||
1577 !boot_cpu_has(X86_FEATURE_VPCLMULQDQ) ||
1578 !boot_cpu_has(X86_FEATURE_PCLMULQDQ) ||
1579 !cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM, NULL))
1580 return 0;
1581 err = simd_register_skciphers_compat(&aes_xts_alg_vaes_avx2, 1,
1582 &aes_xts_simdalg_vaes_avx2);
1583 if (err)
1584 return err;
1585
1586 if (!boot_cpu_has(X86_FEATURE_AVX512BW) ||
1587 !boot_cpu_has(X86_FEATURE_AVX512VL) ||
1588 !boot_cpu_has(X86_FEATURE_BMI2) ||
1589 !cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM |
1590 XFEATURE_MASK_AVX512, NULL))
1591 return 0;
1592
1593 err = simd_register_skciphers_compat(&aes_xts_alg_vaes_avx10_256, 1,
1594 &aes_xts_simdalg_vaes_avx10_256);
1595 if (err)
1596 return err;
1597 err = simd_register_aeads_compat(aes_gcm_algs_vaes_avx10_256,
1598 ARRAY_SIZE(aes_gcm_algs_vaes_avx10_256),
1599 aes_gcm_simdalgs_vaes_avx10_256);
1600 if (err)
1601 return err;
1602
1603 if (x86_match_cpu(zmm_exclusion_list)) {
1604 int i;
1605
1606 aes_xts_alg_vaes_avx10_512.base.cra_priority = 1;
1607 for (i = 0; i < ARRAY_SIZE(aes_gcm_algs_vaes_avx10_512); i++)
1608 aes_gcm_algs_vaes_avx10_512[i].base.cra_priority = 1;
1609 }
1610
1611 err = simd_register_skciphers_compat(&aes_xts_alg_vaes_avx10_512, 1,
1612 &aes_xts_simdalg_vaes_avx10_512);
1613 if (err)
1614 return err;
1615 err = simd_register_aeads_compat(aes_gcm_algs_vaes_avx10_512,
1616 ARRAY_SIZE(aes_gcm_algs_vaes_avx10_512),
1617 aes_gcm_simdalgs_vaes_avx10_512);
1618 if (err)
1619 return err;
1620#endif /* CONFIG_AS_VAES && CONFIG_AS_VPCLMULQDQ */
1621 return 0;
1622}
1623
1624static void unregister_avx_algs(void)
1625{
1626 if (aes_xts_simdalg_aesni_avx)
1627 simd_unregister_skciphers(&aes_xts_alg_aesni_avx, 1,
1628 &aes_xts_simdalg_aesni_avx);
1629 if (aes_gcm_simdalgs_aesni_avx[0])
1630 simd_unregister_aeads(aes_gcm_algs_aesni_avx,
1631 ARRAY_SIZE(aes_gcm_algs_aesni_avx),
1632 aes_gcm_simdalgs_aesni_avx);
1633#if defined(CONFIG_AS_VAES) && defined(CONFIG_AS_VPCLMULQDQ)
1634 if (aes_xts_simdalg_vaes_avx2)
1635 simd_unregister_skciphers(&aes_xts_alg_vaes_avx2, 1,
1636 &aes_xts_simdalg_vaes_avx2);
1637 if (aes_xts_simdalg_vaes_avx10_256)
1638 simd_unregister_skciphers(&aes_xts_alg_vaes_avx10_256, 1,
1639 &aes_xts_simdalg_vaes_avx10_256);
1640 if (aes_gcm_simdalgs_vaes_avx10_256[0])
1641 simd_unregister_aeads(aes_gcm_algs_vaes_avx10_256,
1642 ARRAY_SIZE(aes_gcm_algs_vaes_avx10_256),
1643 aes_gcm_simdalgs_vaes_avx10_256);
1644 if (aes_xts_simdalg_vaes_avx10_512)
1645 simd_unregister_skciphers(&aes_xts_alg_vaes_avx10_512, 1,
1646 &aes_xts_simdalg_vaes_avx10_512);
1647 if (aes_gcm_simdalgs_vaes_avx10_512[0])
1648 simd_unregister_aeads(aes_gcm_algs_vaes_avx10_512,
1649 ARRAY_SIZE(aes_gcm_algs_vaes_avx10_512),
1650 aes_gcm_simdalgs_vaes_avx10_512);
1651#endif
1652}
1653#else /* CONFIG_X86_64 */
1654static struct aead_alg aes_gcm_algs_aesni[0];
1655static struct simd_aead_alg *aes_gcm_simdalgs_aesni[0];
1656
1657static int __init register_avx_algs(void)
1658{
1659 return 0;
1660}
1661
1662static void unregister_avx_algs(void)
1663{
1664}
1665#endif /* !CONFIG_X86_64 */
1666
1667static const struct x86_cpu_id aesni_cpu_id[] = {
1668 X86_MATCH_FEATURE(X86_FEATURE_AES, NULL),
1669 {}
1670};
1671MODULE_DEVICE_TABLE(x86cpu, aesni_cpu_id);
1672
1673static int __init aesni_init(void)
1674{
1675 int err;
1676
1677 if (!x86_match_cpu(aesni_cpu_id))
1678 return -ENODEV;
1679#ifdef CONFIG_X86_64
1680 if (boot_cpu_has(X86_FEATURE_AVX)) {
1681 /* optimize performance of ctr mode encryption transform */
1682 static_call_update(aesni_ctr_enc_tfm, aesni_ctr_enc_avx_tfm);
1683 pr_info("AES CTR mode by8 optimization enabled\n");
1684 }
1685#endif /* CONFIG_X86_64 */
1686
1687 err = crypto_register_alg(&aesni_cipher_alg);
1688 if (err)
1689 return err;
1690
1691 err = simd_register_skciphers_compat(aesni_skciphers,
1692 ARRAY_SIZE(aesni_skciphers),
1693 aesni_simd_skciphers);
1694 if (err)
1695 goto unregister_cipher;
1696
1697 err = simd_register_aeads_compat(aes_gcm_algs_aesni,
1698 ARRAY_SIZE(aes_gcm_algs_aesni),
1699 aes_gcm_simdalgs_aesni);
1700 if (err)
1701 goto unregister_skciphers;
1702
1703#ifdef CONFIG_X86_64
1704 if (boot_cpu_has(X86_FEATURE_AVX))
1705 err = simd_register_skciphers_compat(&aesni_xctr, 1,
1706 &aesni_simd_xctr);
1707 if (err)
1708 goto unregister_aeads;
1709#endif /* CONFIG_X86_64 */
1710
1711 err = register_avx_algs();
1712 if (err)
1713 goto unregister_avx;
1714
1715 return 0;
1716
1717unregister_avx:
1718 unregister_avx_algs();
1719#ifdef CONFIG_X86_64
1720 if (aesni_simd_xctr)
1721 simd_unregister_skciphers(&aesni_xctr, 1, &aesni_simd_xctr);
1722unregister_aeads:
1723#endif /* CONFIG_X86_64 */
1724 simd_unregister_aeads(aes_gcm_algs_aesni,
1725 ARRAY_SIZE(aes_gcm_algs_aesni),
1726 aes_gcm_simdalgs_aesni);
1727unregister_skciphers:
1728 simd_unregister_skciphers(aesni_skciphers, ARRAY_SIZE(aesni_skciphers),
1729 aesni_simd_skciphers);
1730unregister_cipher:
1731 crypto_unregister_alg(&aesni_cipher_alg);
1732 return err;
1733}
1734
1735static void __exit aesni_exit(void)
1736{
1737 simd_unregister_aeads(aes_gcm_algs_aesni,
1738 ARRAY_SIZE(aes_gcm_algs_aesni),
1739 aes_gcm_simdalgs_aesni);
1740 simd_unregister_skciphers(aesni_skciphers, ARRAY_SIZE(aesni_skciphers),
1741 aesni_simd_skciphers);
1742 crypto_unregister_alg(&aesni_cipher_alg);
1743#ifdef CONFIG_X86_64
1744 if (boot_cpu_has(X86_FEATURE_AVX))
1745 simd_unregister_skciphers(&aesni_xctr, 1, &aesni_simd_xctr);
1746#endif /* CONFIG_X86_64 */
1747 unregister_avx_algs();
1748}
1749
1750module_init(aesni_init);
1751module_exit(aesni_exit);
1752
1753MODULE_DESCRIPTION("AES cipher and modes, optimized with AES-NI or VAES instructions");
1754MODULE_LICENSE("GPL");
1755MODULE_ALIAS_CRYPTO("aes");
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Support for Intel AES-NI instructions. This file contains glue
4 * code, the real AES implementation is in intel-aes_asm.S.
5 *
6 * Copyright (C) 2008, Intel Corp.
7 * Author: Huang Ying <ying.huang@intel.com>
8 *
9 * Added RFC4106 AES-GCM support for 128-bit keys under the AEAD
10 * interface for 64-bit kernels.
11 * Authors: Adrian Hoban <adrian.hoban@intel.com>
12 * Gabriele Paoloni <gabriele.paoloni@intel.com>
13 * Tadeusz Struk (tadeusz.struk@intel.com)
14 * Aidan O'Mahony (aidan.o.mahony@intel.com)
15 * Copyright (c) 2010, Intel Corporation.
16 */
17
18#include <linux/hardirq.h>
19#include <linux/types.h>
20#include <linux/module.h>
21#include <linux/err.h>
22#include <crypto/algapi.h>
23#include <crypto/aes.h>
24#include <crypto/ctr.h>
25#include <crypto/b128ops.h>
26#include <crypto/gcm.h>
27#include <crypto/xts.h>
28#include <asm/cpu_device_id.h>
29#include <asm/simd.h>
30#include <crypto/scatterwalk.h>
31#include <crypto/internal/aead.h>
32#include <crypto/internal/simd.h>
33#include <crypto/internal/skcipher.h>
34#include <linux/jump_label.h>
35#include <linux/workqueue.h>
36#include <linux/spinlock.h>
37#include <linux/static_call.h>
38
39
40#define AESNI_ALIGN 16
41#define AESNI_ALIGN_ATTR __attribute__ ((__aligned__(AESNI_ALIGN)))
42#define AES_BLOCK_MASK (~(AES_BLOCK_SIZE - 1))
43#define RFC4106_HASH_SUBKEY_SIZE 16
44#define AESNI_ALIGN_EXTRA ((AESNI_ALIGN - 1) & ~(CRYPTO_MINALIGN - 1))
45#define CRYPTO_AES_CTX_SIZE (sizeof(struct crypto_aes_ctx) + AESNI_ALIGN_EXTRA)
46#define XTS_AES_CTX_SIZE (sizeof(struct aesni_xts_ctx) + AESNI_ALIGN_EXTRA)
47
48/* This data is stored at the end of the crypto_tfm struct.
49 * It's a type of per "session" data storage location.
50 * This needs to be 16 byte aligned.
51 */
52struct aesni_rfc4106_gcm_ctx {
53 u8 hash_subkey[16] AESNI_ALIGN_ATTR;
54 struct crypto_aes_ctx aes_key_expanded AESNI_ALIGN_ATTR;
55 u8 nonce[4];
56};
57
58struct generic_gcmaes_ctx {
59 u8 hash_subkey[16] AESNI_ALIGN_ATTR;
60 struct crypto_aes_ctx aes_key_expanded AESNI_ALIGN_ATTR;
61};
62
63struct aesni_xts_ctx {
64 struct crypto_aes_ctx tweak_ctx AESNI_ALIGN_ATTR;
65 struct crypto_aes_ctx crypt_ctx AESNI_ALIGN_ATTR;
66};
67
68#define GCM_BLOCK_LEN 16
69
70struct gcm_context_data {
71 /* init, update and finalize context data */
72 u8 aad_hash[GCM_BLOCK_LEN];
73 u64 aad_length;
74 u64 in_length;
75 u8 partial_block_enc_key[GCM_BLOCK_LEN];
76 u8 orig_IV[GCM_BLOCK_LEN];
77 u8 current_counter[GCM_BLOCK_LEN];
78 u64 partial_block_len;
79 u64 unused;
80 u8 hash_keys[GCM_BLOCK_LEN * 16];
81};
82
83static inline void *aes_align_addr(void *addr)
84{
85 if (crypto_tfm_ctx_alignment() >= AESNI_ALIGN)
86 return addr;
87 return PTR_ALIGN(addr, AESNI_ALIGN);
88}
89
90asmlinkage int aesni_set_key(struct crypto_aes_ctx *ctx, const u8 *in_key,
91 unsigned int key_len);
92asmlinkage void aesni_enc(const void *ctx, u8 *out, const u8 *in);
93asmlinkage void aesni_dec(const void *ctx, u8 *out, const u8 *in);
94asmlinkage void aesni_ecb_enc(struct crypto_aes_ctx *ctx, u8 *out,
95 const u8 *in, unsigned int len);
96asmlinkage void aesni_ecb_dec(struct crypto_aes_ctx *ctx, u8 *out,
97 const u8 *in, unsigned int len);
98asmlinkage void aesni_cbc_enc(struct crypto_aes_ctx *ctx, u8 *out,
99 const u8 *in, unsigned int len, u8 *iv);
100asmlinkage void aesni_cbc_dec(struct crypto_aes_ctx *ctx, u8 *out,
101 const u8 *in, unsigned int len, u8 *iv);
102asmlinkage void aesni_cts_cbc_enc(struct crypto_aes_ctx *ctx, u8 *out,
103 const u8 *in, unsigned int len, u8 *iv);
104asmlinkage void aesni_cts_cbc_dec(struct crypto_aes_ctx *ctx, u8 *out,
105 const u8 *in, unsigned int len, u8 *iv);
106
107#define AVX_GEN2_OPTSIZE 640
108#define AVX_GEN4_OPTSIZE 4096
109
110asmlinkage void aesni_xts_encrypt(const struct crypto_aes_ctx *ctx, u8 *out,
111 const u8 *in, unsigned int len, u8 *iv);
112
113asmlinkage void aesni_xts_decrypt(const struct crypto_aes_ctx *ctx, u8 *out,
114 const u8 *in, unsigned int len, u8 *iv);
115
116#ifdef CONFIG_X86_64
117
118asmlinkage void aesni_ctr_enc(struct crypto_aes_ctx *ctx, u8 *out,
119 const u8 *in, unsigned int len, u8 *iv);
120DEFINE_STATIC_CALL(aesni_ctr_enc_tfm, aesni_ctr_enc);
121
122/* Scatter / Gather routines, with args similar to above */
123asmlinkage void aesni_gcm_init(void *ctx,
124 struct gcm_context_data *gdata,
125 u8 *iv,
126 u8 *hash_subkey, const u8 *aad,
127 unsigned long aad_len);
128asmlinkage void aesni_gcm_enc_update(void *ctx,
129 struct gcm_context_data *gdata, u8 *out,
130 const u8 *in, unsigned long plaintext_len);
131asmlinkage void aesni_gcm_dec_update(void *ctx,
132 struct gcm_context_data *gdata, u8 *out,
133 const u8 *in,
134 unsigned long ciphertext_len);
135asmlinkage void aesni_gcm_finalize(void *ctx,
136 struct gcm_context_data *gdata,
137 u8 *auth_tag, unsigned long auth_tag_len);
138
139asmlinkage void aes_ctr_enc_128_avx_by8(const u8 *in, u8 *iv,
140 void *keys, u8 *out, unsigned int num_bytes);
141asmlinkage void aes_ctr_enc_192_avx_by8(const u8 *in, u8 *iv,
142 void *keys, u8 *out, unsigned int num_bytes);
143asmlinkage void aes_ctr_enc_256_avx_by8(const u8 *in, u8 *iv,
144 void *keys, u8 *out, unsigned int num_bytes);
145
146
147asmlinkage void aes_xctr_enc_128_avx_by8(const u8 *in, const u8 *iv,
148 const void *keys, u8 *out, unsigned int num_bytes,
149 unsigned int byte_ctr);
150
151asmlinkage void aes_xctr_enc_192_avx_by8(const u8 *in, const u8 *iv,
152 const void *keys, u8 *out, unsigned int num_bytes,
153 unsigned int byte_ctr);
154
155asmlinkage void aes_xctr_enc_256_avx_by8(const u8 *in, const u8 *iv,
156 const void *keys, u8 *out, unsigned int num_bytes,
157 unsigned int byte_ctr);
158
159/*
160 * asmlinkage void aesni_gcm_init_avx_gen2()
161 * gcm_data *my_ctx_data, context data
162 * u8 *hash_subkey, the Hash sub key input. Data starts on a 16-byte boundary.
163 */
164asmlinkage void aesni_gcm_init_avx_gen2(void *my_ctx_data,
165 struct gcm_context_data *gdata,
166 u8 *iv,
167 u8 *hash_subkey,
168 const u8 *aad,
169 unsigned long aad_len);
170
171asmlinkage void aesni_gcm_enc_update_avx_gen2(void *ctx,
172 struct gcm_context_data *gdata, u8 *out,
173 const u8 *in, unsigned long plaintext_len);
174asmlinkage void aesni_gcm_dec_update_avx_gen2(void *ctx,
175 struct gcm_context_data *gdata, u8 *out,
176 const u8 *in,
177 unsigned long ciphertext_len);
178asmlinkage void aesni_gcm_finalize_avx_gen2(void *ctx,
179 struct gcm_context_data *gdata,
180 u8 *auth_tag, unsigned long auth_tag_len);
181
182/*
183 * asmlinkage void aesni_gcm_init_avx_gen4()
184 * gcm_data *my_ctx_data, context data
185 * u8 *hash_subkey, the Hash sub key input. Data starts on a 16-byte boundary.
186 */
187asmlinkage void aesni_gcm_init_avx_gen4(void *my_ctx_data,
188 struct gcm_context_data *gdata,
189 u8 *iv,
190 u8 *hash_subkey,
191 const u8 *aad,
192 unsigned long aad_len);
193
194asmlinkage void aesni_gcm_enc_update_avx_gen4(void *ctx,
195 struct gcm_context_data *gdata, u8 *out,
196 const u8 *in, unsigned long plaintext_len);
197asmlinkage void aesni_gcm_dec_update_avx_gen4(void *ctx,
198 struct gcm_context_data *gdata, u8 *out,
199 const u8 *in,
200 unsigned long ciphertext_len);
201asmlinkage void aesni_gcm_finalize_avx_gen4(void *ctx,
202 struct gcm_context_data *gdata,
203 u8 *auth_tag, unsigned long auth_tag_len);
204
205static __ro_after_init DEFINE_STATIC_KEY_FALSE(gcm_use_avx);
206static __ro_after_init DEFINE_STATIC_KEY_FALSE(gcm_use_avx2);
207
208static inline struct
209aesni_rfc4106_gcm_ctx *aesni_rfc4106_gcm_ctx_get(struct crypto_aead *tfm)
210{
211 return aes_align_addr(crypto_aead_ctx(tfm));
212}
213
214static inline struct
215generic_gcmaes_ctx *generic_gcmaes_ctx_get(struct crypto_aead *tfm)
216{
217 return aes_align_addr(crypto_aead_ctx(tfm));
218}
219#endif
220
221static inline struct crypto_aes_ctx *aes_ctx(void *raw_ctx)
222{
223 return aes_align_addr(raw_ctx);
224}
225
226static inline struct aesni_xts_ctx *aes_xts_ctx(struct crypto_skcipher *tfm)
227{
228 return aes_align_addr(crypto_skcipher_ctx(tfm));
229}
230
231static int aes_set_key_common(struct crypto_aes_ctx *ctx,
232 const u8 *in_key, unsigned int key_len)
233{
234 int err;
235
236 if (key_len != AES_KEYSIZE_128 && key_len != AES_KEYSIZE_192 &&
237 key_len != AES_KEYSIZE_256)
238 return -EINVAL;
239
240 if (!crypto_simd_usable())
241 err = aes_expandkey(ctx, in_key, key_len);
242 else {
243 kernel_fpu_begin();
244 err = aesni_set_key(ctx, in_key, key_len);
245 kernel_fpu_end();
246 }
247
248 return err;
249}
250
251static int aes_set_key(struct crypto_tfm *tfm, const u8 *in_key,
252 unsigned int key_len)
253{
254 return aes_set_key_common(aes_ctx(crypto_tfm_ctx(tfm)), in_key,
255 key_len);
256}
257
258static void aesni_encrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
259{
260 struct crypto_aes_ctx *ctx = aes_ctx(crypto_tfm_ctx(tfm));
261
262 if (!crypto_simd_usable()) {
263 aes_encrypt(ctx, dst, src);
264 } else {
265 kernel_fpu_begin();
266 aesni_enc(ctx, dst, src);
267 kernel_fpu_end();
268 }
269}
270
271static void aesni_decrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
272{
273 struct crypto_aes_ctx *ctx = aes_ctx(crypto_tfm_ctx(tfm));
274
275 if (!crypto_simd_usable()) {
276 aes_decrypt(ctx, dst, src);
277 } else {
278 kernel_fpu_begin();
279 aesni_dec(ctx, dst, src);
280 kernel_fpu_end();
281 }
282}
283
284static int aesni_skcipher_setkey(struct crypto_skcipher *tfm, const u8 *key,
285 unsigned int len)
286{
287 return aes_set_key_common(aes_ctx(crypto_skcipher_ctx(tfm)), key, len);
288}
289
290static int ecb_encrypt(struct skcipher_request *req)
291{
292 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
293 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
294 struct skcipher_walk walk;
295 unsigned int nbytes;
296 int err;
297
298 err = skcipher_walk_virt(&walk, req, false);
299
300 while ((nbytes = walk.nbytes)) {
301 kernel_fpu_begin();
302 aesni_ecb_enc(ctx, walk.dst.virt.addr, walk.src.virt.addr,
303 nbytes & AES_BLOCK_MASK);
304 kernel_fpu_end();
305 nbytes &= AES_BLOCK_SIZE - 1;
306 err = skcipher_walk_done(&walk, nbytes);
307 }
308
309 return err;
310}
311
312static int ecb_decrypt(struct skcipher_request *req)
313{
314 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
315 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
316 struct skcipher_walk walk;
317 unsigned int nbytes;
318 int err;
319
320 err = skcipher_walk_virt(&walk, req, false);
321
322 while ((nbytes = walk.nbytes)) {
323 kernel_fpu_begin();
324 aesni_ecb_dec(ctx, walk.dst.virt.addr, walk.src.virt.addr,
325 nbytes & AES_BLOCK_MASK);
326 kernel_fpu_end();
327 nbytes &= AES_BLOCK_SIZE - 1;
328 err = skcipher_walk_done(&walk, nbytes);
329 }
330
331 return err;
332}
333
334static int cbc_encrypt(struct skcipher_request *req)
335{
336 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
337 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
338 struct skcipher_walk walk;
339 unsigned int nbytes;
340 int err;
341
342 err = skcipher_walk_virt(&walk, req, false);
343
344 while ((nbytes = walk.nbytes)) {
345 kernel_fpu_begin();
346 aesni_cbc_enc(ctx, walk.dst.virt.addr, walk.src.virt.addr,
347 nbytes & AES_BLOCK_MASK, walk.iv);
348 kernel_fpu_end();
349 nbytes &= AES_BLOCK_SIZE - 1;
350 err = skcipher_walk_done(&walk, nbytes);
351 }
352
353 return err;
354}
355
356static int cbc_decrypt(struct skcipher_request *req)
357{
358 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
359 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
360 struct skcipher_walk walk;
361 unsigned int nbytes;
362 int err;
363
364 err = skcipher_walk_virt(&walk, req, false);
365
366 while ((nbytes = walk.nbytes)) {
367 kernel_fpu_begin();
368 aesni_cbc_dec(ctx, walk.dst.virt.addr, walk.src.virt.addr,
369 nbytes & AES_BLOCK_MASK, walk.iv);
370 kernel_fpu_end();
371 nbytes &= AES_BLOCK_SIZE - 1;
372 err = skcipher_walk_done(&walk, nbytes);
373 }
374
375 return err;
376}
377
378static int cts_cbc_encrypt(struct skcipher_request *req)
379{
380 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
381 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
382 int cbc_blocks = DIV_ROUND_UP(req->cryptlen, AES_BLOCK_SIZE) - 2;
383 struct scatterlist *src = req->src, *dst = req->dst;
384 struct scatterlist sg_src[2], sg_dst[2];
385 struct skcipher_request subreq;
386 struct skcipher_walk walk;
387 int err;
388
389 skcipher_request_set_tfm(&subreq, tfm);
390 skcipher_request_set_callback(&subreq, skcipher_request_flags(req),
391 NULL, NULL);
392
393 if (req->cryptlen <= AES_BLOCK_SIZE) {
394 if (req->cryptlen < AES_BLOCK_SIZE)
395 return -EINVAL;
396 cbc_blocks = 1;
397 }
398
399 if (cbc_blocks > 0) {
400 skcipher_request_set_crypt(&subreq, req->src, req->dst,
401 cbc_blocks * AES_BLOCK_SIZE,
402 req->iv);
403
404 err = cbc_encrypt(&subreq);
405 if (err)
406 return err;
407
408 if (req->cryptlen == AES_BLOCK_SIZE)
409 return 0;
410
411 dst = src = scatterwalk_ffwd(sg_src, req->src, subreq.cryptlen);
412 if (req->dst != req->src)
413 dst = scatterwalk_ffwd(sg_dst, req->dst,
414 subreq.cryptlen);
415 }
416
417 /* handle ciphertext stealing */
418 skcipher_request_set_crypt(&subreq, src, dst,
419 req->cryptlen - cbc_blocks * AES_BLOCK_SIZE,
420 req->iv);
421
422 err = skcipher_walk_virt(&walk, &subreq, false);
423 if (err)
424 return err;
425
426 kernel_fpu_begin();
427 aesni_cts_cbc_enc(ctx, walk.dst.virt.addr, walk.src.virt.addr,
428 walk.nbytes, walk.iv);
429 kernel_fpu_end();
430
431 return skcipher_walk_done(&walk, 0);
432}
433
434static int cts_cbc_decrypt(struct skcipher_request *req)
435{
436 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
437 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
438 int cbc_blocks = DIV_ROUND_UP(req->cryptlen, AES_BLOCK_SIZE) - 2;
439 struct scatterlist *src = req->src, *dst = req->dst;
440 struct scatterlist sg_src[2], sg_dst[2];
441 struct skcipher_request subreq;
442 struct skcipher_walk walk;
443 int err;
444
445 skcipher_request_set_tfm(&subreq, tfm);
446 skcipher_request_set_callback(&subreq, skcipher_request_flags(req),
447 NULL, NULL);
448
449 if (req->cryptlen <= AES_BLOCK_SIZE) {
450 if (req->cryptlen < AES_BLOCK_SIZE)
451 return -EINVAL;
452 cbc_blocks = 1;
453 }
454
455 if (cbc_blocks > 0) {
456 skcipher_request_set_crypt(&subreq, req->src, req->dst,
457 cbc_blocks * AES_BLOCK_SIZE,
458 req->iv);
459
460 err = cbc_decrypt(&subreq);
461 if (err)
462 return err;
463
464 if (req->cryptlen == AES_BLOCK_SIZE)
465 return 0;
466
467 dst = src = scatterwalk_ffwd(sg_src, req->src, subreq.cryptlen);
468 if (req->dst != req->src)
469 dst = scatterwalk_ffwd(sg_dst, req->dst,
470 subreq.cryptlen);
471 }
472
473 /* handle ciphertext stealing */
474 skcipher_request_set_crypt(&subreq, src, dst,
475 req->cryptlen - cbc_blocks * AES_BLOCK_SIZE,
476 req->iv);
477
478 err = skcipher_walk_virt(&walk, &subreq, false);
479 if (err)
480 return err;
481
482 kernel_fpu_begin();
483 aesni_cts_cbc_dec(ctx, walk.dst.virt.addr, walk.src.virt.addr,
484 walk.nbytes, walk.iv);
485 kernel_fpu_end();
486
487 return skcipher_walk_done(&walk, 0);
488}
489
490#ifdef CONFIG_X86_64
491static void aesni_ctr_enc_avx_tfm(struct crypto_aes_ctx *ctx, u8 *out,
492 const u8 *in, unsigned int len, u8 *iv)
493{
494 /*
495 * based on key length, override with the by8 version
496 * of ctr mode encryption/decryption for improved performance
497 * aes_set_key_common() ensures that key length is one of
498 * {128,192,256}
499 */
500 if (ctx->key_length == AES_KEYSIZE_128)
501 aes_ctr_enc_128_avx_by8(in, iv, (void *)ctx, out, len);
502 else if (ctx->key_length == AES_KEYSIZE_192)
503 aes_ctr_enc_192_avx_by8(in, iv, (void *)ctx, out, len);
504 else
505 aes_ctr_enc_256_avx_by8(in, iv, (void *)ctx, out, len);
506}
507
508static int ctr_crypt(struct skcipher_request *req)
509{
510 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
511 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
512 u8 keystream[AES_BLOCK_SIZE];
513 struct skcipher_walk walk;
514 unsigned int nbytes;
515 int err;
516
517 err = skcipher_walk_virt(&walk, req, false);
518
519 while ((nbytes = walk.nbytes) > 0) {
520 kernel_fpu_begin();
521 if (nbytes & AES_BLOCK_MASK)
522 static_call(aesni_ctr_enc_tfm)(ctx, walk.dst.virt.addr,
523 walk.src.virt.addr,
524 nbytes & AES_BLOCK_MASK,
525 walk.iv);
526 nbytes &= ~AES_BLOCK_MASK;
527
528 if (walk.nbytes == walk.total && nbytes > 0) {
529 aesni_enc(ctx, keystream, walk.iv);
530 crypto_xor_cpy(walk.dst.virt.addr + walk.nbytes - nbytes,
531 walk.src.virt.addr + walk.nbytes - nbytes,
532 keystream, nbytes);
533 crypto_inc(walk.iv, AES_BLOCK_SIZE);
534 nbytes = 0;
535 }
536 kernel_fpu_end();
537 err = skcipher_walk_done(&walk, nbytes);
538 }
539 return err;
540}
541
542static void aesni_xctr_enc_avx_tfm(struct crypto_aes_ctx *ctx, u8 *out,
543 const u8 *in, unsigned int len, u8 *iv,
544 unsigned int byte_ctr)
545{
546 if (ctx->key_length == AES_KEYSIZE_128)
547 aes_xctr_enc_128_avx_by8(in, iv, (void *)ctx, out, len,
548 byte_ctr);
549 else if (ctx->key_length == AES_KEYSIZE_192)
550 aes_xctr_enc_192_avx_by8(in, iv, (void *)ctx, out, len,
551 byte_ctr);
552 else
553 aes_xctr_enc_256_avx_by8(in, iv, (void *)ctx, out, len,
554 byte_ctr);
555}
556
557static int xctr_crypt(struct skcipher_request *req)
558{
559 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
560 struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
561 u8 keystream[AES_BLOCK_SIZE];
562 struct skcipher_walk walk;
563 unsigned int nbytes;
564 unsigned int byte_ctr = 0;
565 int err;
566 __le32 block[AES_BLOCK_SIZE / sizeof(__le32)];
567
568 err = skcipher_walk_virt(&walk, req, false);
569
570 while ((nbytes = walk.nbytes) > 0) {
571 kernel_fpu_begin();
572 if (nbytes & AES_BLOCK_MASK)
573 aesni_xctr_enc_avx_tfm(ctx, walk.dst.virt.addr,
574 walk.src.virt.addr, nbytes & AES_BLOCK_MASK,
575 walk.iv, byte_ctr);
576 nbytes &= ~AES_BLOCK_MASK;
577 byte_ctr += walk.nbytes - nbytes;
578
579 if (walk.nbytes == walk.total && nbytes > 0) {
580 memcpy(block, walk.iv, AES_BLOCK_SIZE);
581 block[0] ^= cpu_to_le32(1 + byte_ctr / AES_BLOCK_SIZE);
582 aesni_enc(ctx, keystream, (u8 *)block);
583 crypto_xor_cpy(walk.dst.virt.addr + walk.nbytes -
584 nbytes, walk.src.virt.addr + walk.nbytes
585 - nbytes, keystream, nbytes);
586 byte_ctr += nbytes;
587 nbytes = 0;
588 }
589 kernel_fpu_end();
590 err = skcipher_walk_done(&walk, nbytes);
591 }
592 return err;
593}
594
595static int
596rfc4106_set_hash_subkey(u8 *hash_subkey, const u8 *key, unsigned int key_len)
597{
598 struct crypto_aes_ctx ctx;
599 int ret;
600
601 ret = aes_expandkey(&ctx, key, key_len);
602 if (ret)
603 return ret;
604
605 /* Clear the data in the hash sub key container to zero.*/
606 /* We want to cipher all zeros to create the hash sub key. */
607 memset(hash_subkey, 0, RFC4106_HASH_SUBKEY_SIZE);
608
609 aes_encrypt(&ctx, hash_subkey, hash_subkey);
610
611 memzero_explicit(&ctx, sizeof(ctx));
612 return 0;
613}
614
615static int common_rfc4106_set_key(struct crypto_aead *aead, const u8 *key,
616 unsigned int key_len)
617{
618 struct aesni_rfc4106_gcm_ctx *ctx = aesni_rfc4106_gcm_ctx_get(aead);
619
620 if (key_len < 4)
621 return -EINVAL;
622
623 /*Account for 4 byte nonce at the end.*/
624 key_len -= 4;
625
626 memcpy(ctx->nonce, key + key_len, sizeof(ctx->nonce));
627
628 return aes_set_key_common(&ctx->aes_key_expanded, key, key_len) ?:
629 rfc4106_set_hash_subkey(ctx->hash_subkey, key, key_len);
630}
631
632/* This is the Integrity Check Value (aka the authentication tag) length and can
633 * be 8, 12 or 16 bytes long. */
634static int common_rfc4106_set_authsize(struct crypto_aead *aead,
635 unsigned int authsize)
636{
637 switch (authsize) {
638 case 8:
639 case 12:
640 case 16:
641 break;
642 default:
643 return -EINVAL;
644 }
645
646 return 0;
647}
648
649static int generic_gcmaes_set_authsize(struct crypto_aead *tfm,
650 unsigned int authsize)
651{
652 switch (authsize) {
653 case 4:
654 case 8:
655 case 12:
656 case 13:
657 case 14:
658 case 15:
659 case 16:
660 break;
661 default:
662 return -EINVAL;
663 }
664
665 return 0;
666}
667
668static int gcmaes_crypt_by_sg(bool enc, struct aead_request *req,
669 unsigned int assoclen, u8 *hash_subkey,
670 u8 *iv, void *aes_ctx, u8 *auth_tag,
671 unsigned long auth_tag_len)
672{
673 u8 databuf[sizeof(struct gcm_context_data) + (AESNI_ALIGN - 8)] __aligned(8);
674 struct gcm_context_data *data = PTR_ALIGN((void *)databuf, AESNI_ALIGN);
675 unsigned long left = req->cryptlen;
676 struct scatter_walk assoc_sg_walk;
677 struct skcipher_walk walk;
678 bool do_avx, do_avx2;
679 u8 *assocmem = NULL;
680 u8 *assoc;
681 int err;
682
683 if (!enc)
684 left -= auth_tag_len;
685
686 do_avx = (left >= AVX_GEN2_OPTSIZE);
687 do_avx2 = (left >= AVX_GEN4_OPTSIZE);
688
689 /* Linearize assoc, if not already linear */
690 if (req->src->length >= assoclen && req->src->length) {
691 scatterwalk_start(&assoc_sg_walk, req->src);
692 assoc = scatterwalk_map(&assoc_sg_walk);
693 } else {
694 gfp_t flags = (req->base.flags & CRYPTO_TFM_REQ_MAY_SLEEP) ?
695 GFP_KERNEL : GFP_ATOMIC;
696
697 /* assoc can be any length, so must be on heap */
698 assocmem = kmalloc(assoclen, flags);
699 if (unlikely(!assocmem))
700 return -ENOMEM;
701 assoc = assocmem;
702
703 scatterwalk_map_and_copy(assoc, req->src, 0, assoclen, 0);
704 }
705
706 kernel_fpu_begin();
707 if (static_branch_likely(&gcm_use_avx2) && do_avx2)
708 aesni_gcm_init_avx_gen4(aes_ctx, data, iv, hash_subkey, assoc,
709 assoclen);
710 else if (static_branch_likely(&gcm_use_avx) && do_avx)
711 aesni_gcm_init_avx_gen2(aes_ctx, data, iv, hash_subkey, assoc,
712 assoclen);
713 else
714 aesni_gcm_init(aes_ctx, data, iv, hash_subkey, assoc, assoclen);
715 kernel_fpu_end();
716
717 if (!assocmem)
718 scatterwalk_unmap(assoc);
719 else
720 kfree(assocmem);
721
722 err = enc ? skcipher_walk_aead_encrypt(&walk, req, false)
723 : skcipher_walk_aead_decrypt(&walk, req, false);
724
725 while (walk.nbytes > 0) {
726 kernel_fpu_begin();
727 if (static_branch_likely(&gcm_use_avx2) && do_avx2) {
728 if (enc)
729 aesni_gcm_enc_update_avx_gen4(aes_ctx, data,
730 walk.dst.virt.addr,
731 walk.src.virt.addr,
732 walk.nbytes);
733 else
734 aesni_gcm_dec_update_avx_gen4(aes_ctx, data,
735 walk.dst.virt.addr,
736 walk.src.virt.addr,
737 walk.nbytes);
738 } else if (static_branch_likely(&gcm_use_avx) && do_avx) {
739 if (enc)
740 aesni_gcm_enc_update_avx_gen2(aes_ctx, data,
741 walk.dst.virt.addr,
742 walk.src.virt.addr,
743 walk.nbytes);
744 else
745 aesni_gcm_dec_update_avx_gen2(aes_ctx, data,
746 walk.dst.virt.addr,
747 walk.src.virt.addr,
748 walk.nbytes);
749 } else if (enc) {
750 aesni_gcm_enc_update(aes_ctx, data, walk.dst.virt.addr,
751 walk.src.virt.addr, walk.nbytes);
752 } else {
753 aesni_gcm_dec_update(aes_ctx, data, walk.dst.virt.addr,
754 walk.src.virt.addr, walk.nbytes);
755 }
756 kernel_fpu_end();
757
758 err = skcipher_walk_done(&walk, 0);
759 }
760
761 if (err)
762 return err;
763
764 kernel_fpu_begin();
765 if (static_branch_likely(&gcm_use_avx2) && do_avx2)
766 aesni_gcm_finalize_avx_gen4(aes_ctx, data, auth_tag,
767 auth_tag_len);
768 else if (static_branch_likely(&gcm_use_avx) && do_avx)
769 aesni_gcm_finalize_avx_gen2(aes_ctx, data, auth_tag,
770 auth_tag_len);
771 else
772 aesni_gcm_finalize(aes_ctx, data, auth_tag, auth_tag_len);
773 kernel_fpu_end();
774
775 return 0;
776}
777
778static int gcmaes_encrypt(struct aead_request *req, unsigned int assoclen,
779 u8 *hash_subkey, u8 *iv, void *aes_ctx)
780{
781 struct crypto_aead *tfm = crypto_aead_reqtfm(req);
782 unsigned long auth_tag_len = crypto_aead_authsize(tfm);
783 u8 auth_tag[16];
784 int err;
785
786 err = gcmaes_crypt_by_sg(true, req, assoclen, hash_subkey, iv, aes_ctx,
787 auth_tag, auth_tag_len);
788 if (err)
789 return err;
790
791 scatterwalk_map_and_copy(auth_tag, req->dst,
792 req->assoclen + req->cryptlen,
793 auth_tag_len, 1);
794 return 0;
795}
796
797static int gcmaes_decrypt(struct aead_request *req, unsigned int assoclen,
798 u8 *hash_subkey, u8 *iv, void *aes_ctx)
799{
800 struct crypto_aead *tfm = crypto_aead_reqtfm(req);
801 unsigned long auth_tag_len = crypto_aead_authsize(tfm);
802 u8 auth_tag_msg[16];
803 u8 auth_tag[16];
804 int err;
805
806 err = gcmaes_crypt_by_sg(false, req, assoclen, hash_subkey, iv, aes_ctx,
807 auth_tag, auth_tag_len);
808 if (err)
809 return err;
810
811 /* Copy out original auth_tag */
812 scatterwalk_map_and_copy(auth_tag_msg, req->src,
813 req->assoclen + req->cryptlen - auth_tag_len,
814 auth_tag_len, 0);
815
816 /* Compare generated tag with passed in tag. */
817 if (crypto_memneq(auth_tag_msg, auth_tag, auth_tag_len)) {
818 memzero_explicit(auth_tag, sizeof(auth_tag));
819 return -EBADMSG;
820 }
821 return 0;
822}
823
824static int helper_rfc4106_encrypt(struct aead_request *req)
825{
826 struct crypto_aead *tfm = crypto_aead_reqtfm(req);
827 struct aesni_rfc4106_gcm_ctx *ctx = aesni_rfc4106_gcm_ctx_get(tfm);
828 void *aes_ctx = &(ctx->aes_key_expanded);
829 u8 ivbuf[16 + (AESNI_ALIGN - 8)] __aligned(8);
830 u8 *iv = PTR_ALIGN(&ivbuf[0], AESNI_ALIGN);
831 unsigned int i;
832 __be32 counter = cpu_to_be32(1);
833
834 /* Assuming we are supporting rfc4106 64-bit extended */
835 /* sequence numbers We need to have the AAD length equal */
836 /* to 16 or 20 bytes */
837 if (unlikely(req->assoclen != 16 && req->assoclen != 20))
838 return -EINVAL;
839
840 /* IV below built */
841 for (i = 0; i < 4; i++)
842 *(iv+i) = ctx->nonce[i];
843 for (i = 0; i < 8; i++)
844 *(iv+4+i) = req->iv[i];
845 *((__be32 *)(iv+12)) = counter;
846
847 return gcmaes_encrypt(req, req->assoclen - 8, ctx->hash_subkey, iv,
848 aes_ctx);
849}
850
851static int helper_rfc4106_decrypt(struct aead_request *req)
852{
853 __be32 counter = cpu_to_be32(1);
854 struct crypto_aead *tfm = crypto_aead_reqtfm(req);
855 struct aesni_rfc4106_gcm_ctx *ctx = aesni_rfc4106_gcm_ctx_get(tfm);
856 void *aes_ctx = &(ctx->aes_key_expanded);
857 u8 ivbuf[16 + (AESNI_ALIGN - 8)] __aligned(8);
858 u8 *iv = PTR_ALIGN(&ivbuf[0], AESNI_ALIGN);
859 unsigned int i;
860
861 if (unlikely(req->assoclen != 16 && req->assoclen != 20))
862 return -EINVAL;
863
864 /* Assuming we are supporting rfc4106 64-bit extended */
865 /* sequence numbers We need to have the AAD length */
866 /* equal to 16 or 20 bytes */
867
868 /* IV below built */
869 for (i = 0; i < 4; i++)
870 *(iv+i) = ctx->nonce[i];
871 for (i = 0; i < 8; i++)
872 *(iv+4+i) = req->iv[i];
873 *((__be32 *)(iv+12)) = counter;
874
875 return gcmaes_decrypt(req, req->assoclen - 8, ctx->hash_subkey, iv,
876 aes_ctx);
877}
878#endif
879
880static int xts_aesni_setkey(struct crypto_skcipher *tfm, const u8 *key,
881 unsigned int keylen)
882{
883 struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
884 int err;
885
886 err = xts_verify_key(tfm, key, keylen);
887 if (err)
888 return err;
889
890 keylen /= 2;
891
892 /* first half of xts-key is for crypt */
893 err = aes_set_key_common(&ctx->crypt_ctx, key, keylen);
894 if (err)
895 return err;
896
897 /* second half of xts-key is for tweak */
898 return aes_set_key_common(&ctx->tweak_ctx, key + keylen, keylen);
899}
900
901static int xts_crypt(struct skcipher_request *req, bool encrypt)
902{
903 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
904 struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
905 int tail = req->cryptlen % AES_BLOCK_SIZE;
906 struct skcipher_request subreq;
907 struct skcipher_walk walk;
908 int err;
909
910 if (req->cryptlen < AES_BLOCK_SIZE)
911 return -EINVAL;
912
913 err = skcipher_walk_virt(&walk, req, false);
914 if (!walk.nbytes)
915 return err;
916
917 if (unlikely(tail > 0 && walk.nbytes < walk.total)) {
918 int blocks = DIV_ROUND_UP(req->cryptlen, AES_BLOCK_SIZE) - 2;
919
920 skcipher_walk_abort(&walk);
921
922 skcipher_request_set_tfm(&subreq, tfm);
923 skcipher_request_set_callback(&subreq,
924 skcipher_request_flags(req),
925 NULL, NULL);
926 skcipher_request_set_crypt(&subreq, req->src, req->dst,
927 blocks * AES_BLOCK_SIZE, req->iv);
928 req = &subreq;
929
930 err = skcipher_walk_virt(&walk, req, false);
931 if (!walk.nbytes)
932 return err;
933 } else {
934 tail = 0;
935 }
936
937 kernel_fpu_begin();
938
939 /* calculate first value of T */
940 aesni_enc(&ctx->tweak_ctx, walk.iv, walk.iv);
941
942 while (walk.nbytes > 0) {
943 int nbytes = walk.nbytes;
944
945 if (nbytes < walk.total)
946 nbytes &= ~(AES_BLOCK_SIZE - 1);
947
948 if (encrypt)
949 aesni_xts_encrypt(&ctx->crypt_ctx,
950 walk.dst.virt.addr, walk.src.virt.addr,
951 nbytes, walk.iv);
952 else
953 aesni_xts_decrypt(&ctx->crypt_ctx,
954 walk.dst.virt.addr, walk.src.virt.addr,
955 nbytes, walk.iv);
956 kernel_fpu_end();
957
958 err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
959
960 if (walk.nbytes > 0)
961 kernel_fpu_begin();
962 }
963
964 if (unlikely(tail > 0 && !err)) {
965 struct scatterlist sg_src[2], sg_dst[2];
966 struct scatterlist *src, *dst;
967
968 dst = src = scatterwalk_ffwd(sg_src, req->src, req->cryptlen);
969 if (req->dst != req->src)
970 dst = scatterwalk_ffwd(sg_dst, req->dst, req->cryptlen);
971
972 skcipher_request_set_crypt(req, src, dst, AES_BLOCK_SIZE + tail,
973 req->iv);
974
975 err = skcipher_walk_virt(&walk, &subreq, false);
976 if (err)
977 return err;
978
979 kernel_fpu_begin();
980 if (encrypt)
981 aesni_xts_encrypt(&ctx->crypt_ctx,
982 walk.dst.virt.addr, walk.src.virt.addr,
983 walk.nbytes, walk.iv);
984 else
985 aesni_xts_decrypt(&ctx->crypt_ctx,
986 walk.dst.virt.addr, walk.src.virt.addr,
987 walk.nbytes, walk.iv);
988 kernel_fpu_end();
989
990 err = skcipher_walk_done(&walk, 0);
991 }
992 return err;
993}
994
995static int xts_encrypt(struct skcipher_request *req)
996{
997 return xts_crypt(req, true);
998}
999
1000static int xts_decrypt(struct skcipher_request *req)
1001{
1002 return xts_crypt(req, false);
1003}
1004
1005static struct crypto_alg aesni_cipher_alg = {
1006 .cra_name = "aes",
1007 .cra_driver_name = "aes-aesni",
1008 .cra_priority = 300,
1009 .cra_flags = CRYPTO_ALG_TYPE_CIPHER,
1010 .cra_blocksize = AES_BLOCK_SIZE,
1011 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
1012 .cra_module = THIS_MODULE,
1013 .cra_u = {
1014 .cipher = {
1015 .cia_min_keysize = AES_MIN_KEY_SIZE,
1016 .cia_max_keysize = AES_MAX_KEY_SIZE,
1017 .cia_setkey = aes_set_key,
1018 .cia_encrypt = aesni_encrypt,
1019 .cia_decrypt = aesni_decrypt
1020 }
1021 }
1022};
1023
1024static struct skcipher_alg aesni_skciphers[] = {
1025 {
1026 .base = {
1027 .cra_name = "__ecb(aes)",
1028 .cra_driver_name = "__ecb-aes-aesni",
1029 .cra_priority = 400,
1030 .cra_flags = CRYPTO_ALG_INTERNAL,
1031 .cra_blocksize = AES_BLOCK_SIZE,
1032 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
1033 .cra_module = THIS_MODULE,
1034 },
1035 .min_keysize = AES_MIN_KEY_SIZE,
1036 .max_keysize = AES_MAX_KEY_SIZE,
1037 .setkey = aesni_skcipher_setkey,
1038 .encrypt = ecb_encrypt,
1039 .decrypt = ecb_decrypt,
1040 }, {
1041 .base = {
1042 .cra_name = "__cbc(aes)",
1043 .cra_driver_name = "__cbc-aes-aesni",
1044 .cra_priority = 400,
1045 .cra_flags = CRYPTO_ALG_INTERNAL,
1046 .cra_blocksize = AES_BLOCK_SIZE,
1047 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
1048 .cra_module = THIS_MODULE,
1049 },
1050 .min_keysize = AES_MIN_KEY_SIZE,
1051 .max_keysize = AES_MAX_KEY_SIZE,
1052 .ivsize = AES_BLOCK_SIZE,
1053 .setkey = aesni_skcipher_setkey,
1054 .encrypt = cbc_encrypt,
1055 .decrypt = cbc_decrypt,
1056 }, {
1057 .base = {
1058 .cra_name = "__cts(cbc(aes))",
1059 .cra_driver_name = "__cts-cbc-aes-aesni",
1060 .cra_priority = 400,
1061 .cra_flags = CRYPTO_ALG_INTERNAL,
1062 .cra_blocksize = AES_BLOCK_SIZE,
1063 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
1064 .cra_module = THIS_MODULE,
1065 },
1066 .min_keysize = AES_MIN_KEY_SIZE,
1067 .max_keysize = AES_MAX_KEY_SIZE,
1068 .ivsize = AES_BLOCK_SIZE,
1069 .walksize = 2 * AES_BLOCK_SIZE,
1070 .setkey = aesni_skcipher_setkey,
1071 .encrypt = cts_cbc_encrypt,
1072 .decrypt = cts_cbc_decrypt,
1073#ifdef CONFIG_X86_64
1074 }, {
1075 .base = {
1076 .cra_name = "__ctr(aes)",
1077 .cra_driver_name = "__ctr-aes-aesni",
1078 .cra_priority = 400,
1079 .cra_flags = CRYPTO_ALG_INTERNAL,
1080 .cra_blocksize = 1,
1081 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
1082 .cra_module = THIS_MODULE,
1083 },
1084 .min_keysize = AES_MIN_KEY_SIZE,
1085 .max_keysize = AES_MAX_KEY_SIZE,
1086 .ivsize = AES_BLOCK_SIZE,
1087 .chunksize = AES_BLOCK_SIZE,
1088 .setkey = aesni_skcipher_setkey,
1089 .encrypt = ctr_crypt,
1090 .decrypt = ctr_crypt,
1091#endif
1092 }, {
1093 .base = {
1094 .cra_name = "__xts(aes)",
1095 .cra_driver_name = "__xts-aes-aesni",
1096 .cra_priority = 401,
1097 .cra_flags = CRYPTO_ALG_INTERNAL,
1098 .cra_blocksize = AES_BLOCK_SIZE,
1099 .cra_ctxsize = XTS_AES_CTX_SIZE,
1100 .cra_module = THIS_MODULE,
1101 },
1102 .min_keysize = 2 * AES_MIN_KEY_SIZE,
1103 .max_keysize = 2 * AES_MAX_KEY_SIZE,
1104 .ivsize = AES_BLOCK_SIZE,
1105 .walksize = 2 * AES_BLOCK_SIZE,
1106 .setkey = xts_aesni_setkey,
1107 .encrypt = xts_encrypt,
1108 .decrypt = xts_decrypt,
1109 }
1110};
1111
1112static
1113struct simd_skcipher_alg *aesni_simd_skciphers[ARRAY_SIZE(aesni_skciphers)];
1114
1115#ifdef CONFIG_X86_64
1116/*
1117 * XCTR does not have a non-AVX implementation, so it must be enabled
1118 * conditionally.
1119 */
1120static struct skcipher_alg aesni_xctr = {
1121 .base = {
1122 .cra_name = "__xctr(aes)",
1123 .cra_driver_name = "__xctr-aes-aesni",
1124 .cra_priority = 400,
1125 .cra_flags = CRYPTO_ALG_INTERNAL,
1126 .cra_blocksize = 1,
1127 .cra_ctxsize = CRYPTO_AES_CTX_SIZE,
1128 .cra_module = THIS_MODULE,
1129 },
1130 .min_keysize = AES_MIN_KEY_SIZE,
1131 .max_keysize = AES_MAX_KEY_SIZE,
1132 .ivsize = AES_BLOCK_SIZE,
1133 .chunksize = AES_BLOCK_SIZE,
1134 .setkey = aesni_skcipher_setkey,
1135 .encrypt = xctr_crypt,
1136 .decrypt = xctr_crypt,
1137};
1138
1139static struct simd_skcipher_alg *aesni_simd_xctr;
1140#endif /* CONFIG_X86_64 */
1141
1142#ifdef CONFIG_X86_64
1143static int generic_gcmaes_set_key(struct crypto_aead *aead, const u8 *key,
1144 unsigned int key_len)
1145{
1146 struct generic_gcmaes_ctx *ctx = generic_gcmaes_ctx_get(aead);
1147
1148 return aes_set_key_common(&ctx->aes_key_expanded, key, key_len) ?:
1149 rfc4106_set_hash_subkey(ctx->hash_subkey, key, key_len);
1150}
1151
1152static int generic_gcmaes_encrypt(struct aead_request *req)
1153{
1154 struct crypto_aead *tfm = crypto_aead_reqtfm(req);
1155 struct generic_gcmaes_ctx *ctx = generic_gcmaes_ctx_get(tfm);
1156 void *aes_ctx = &(ctx->aes_key_expanded);
1157 u8 ivbuf[16 + (AESNI_ALIGN - 8)] __aligned(8);
1158 u8 *iv = PTR_ALIGN(&ivbuf[0], AESNI_ALIGN);
1159 __be32 counter = cpu_to_be32(1);
1160
1161 memcpy(iv, req->iv, 12);
1162 *((__be32 *)(iv+12)) = counter;
1163
1164 return gcmaes_encrypt(req, req->assoclen, ctx->hash_subkey, iv,
1165 aes_ctx);
1166}
1167
1168static int generic_gcmaes_decrypt(struct aead_request *req)
1169{
1170 __be32 counter = cpu_to_be32(1);
1171 struct crypto_aead *tfm = crypto_aead_reqtfm(req);
1172 struct generic_gcmaes_ctx *ctx = generic_gcmaes_ctx_get(tfm);
1173 void *aes_ctx = &(ctx->aes_key_expanded);
1174 u8 ivbuf[16 + (AESNI_ALIGN - 8)] __aligned(8);
1175 u8 *iv = PTR_ALIGN(&ivbuf[0], AESNI_ALIGN);
1176
1177 memcpy(iv, req->iv, 12);
1178 *((__be32 *)(iv+12)) = counter;
1179
1180 return gcmaes_decrypt(req, req->assoclen, ctx->hash_subkey, iv,
1181 aes_ctx);
1182}
1183
1184static struct aead_alg aesni_aeads[] = { {
1185 .setkey = common_rfc4106_set_key,
1186 .setauthsize = common_rfc4106_set_authsize,
1187 .encrypt = helper_rfc4106_encrypt,
1188 .decrypt = helper_rfc4106_decrypt,
1189 .ivsize = GCM_RFC4106_IV_SIZE,
1190 .maxauthsize = 16,
1191 .base = {
1192 .cra_name = "__rfc4106(gcm(aes))",
1193 .cra_driver_name = "__rfc4106-gcm-aesni",
1194 .cra_priority = 400,
1195 .cra_flags = CRYPTO_ALG_INTERNAL,
1196 .cra_blocksize = 1,
1197 .cra_ctxsize = sizeof(struct aesni_rfc4106_gcm_ctx),
1198 .cra_alignmask = 0,
1199 .cra_module = THIS_MODULE,
1200 },
1201}, {
1202 .setkey = generic_gcmaes_set_key,
1203 .setauthsize = generic_gcmaes_set_authsize,
1204 .encrypt = generic_gcmaes_encrypt,
1205 .decrypt = generic_gcmaes_decrypt,
1206 .ivsize = GCM_AES_IV_SIZE,
1207 .maxauthsize = 16,
1208 .base = {
1209 .cra_name = "__gcm(aes)",
1210 .cra_driver_name = "__generic-gcm-aesni",
1211 .cra_priority = 400,
1212 .cra_flags = CRYPTO_ALG_INTERNAL,
1213 .cra_blocksize = 1,
1214 .cra_ctxsize = sizeof(struct generic_gcmaes_ctx),
1215 .cra_alignmask = 0,
1216 .cra_module = THIS_MODULE,
1217 },
1218} };
1219#else
1220static struct aead_alg aesni_aeads[0];
1221#endif
1222
1223static struct simd_aead_alg *aesni_simd_aeads[ARRAY_SIZE(aesni_aeads)];
1224
1225static const struct x86_cpu_id aesni_cpu_id[] = {
1226 X86_MATCH_FEATURE(X86_FEATURE_AES, NULL),
1227 {}
1228};
1229MODULE_DEVICE_TABLE(x86cpu, aesni_cpu_id);
1230
1231static int __init aesni_init(void)
1232{
1233 int err;
1234
1235 if (!x86_match_cpu(aesni_cpu_id))
1236 return -ENODEV;
1237#ifdef CONFIG_X86_64
1238 if (boot_cpu_has(X86_FEATURE_AVX2)) {
1239 pr_info("AVX2 version of gcm_enc/dec engaged.\n");
1240 static_branch_enable(&gcm_use_avx);
1241 static_branch_enable(&gcm_use_avx2);
1242 } else
1243 if (boot_cpu_has(X86_FEATURE_AVX)) {
1244 pr_info("AVX version of gcm_enc/dec engaged.\n");
1245 static_branch_enable(&gcm_use_avx);
1246 } else {
1247 pr_info("SSE version of gcm_enc/dec engaged.\n");
1248 }
1249 if (boot_cpu_has(X86_FEATURE_AVX)) {
1250 /* optimize performance of ctr mode encryption transform */
1251 static_call_update(aesni_ctr_enc_tfm, aesni_ctr_enc_avx_tfm);
1252 pr_info("AES CTR mode by8 optimization enabled\n");
1253 }
1254#endif /* CONFIG_X86_64 */
1255
1256 err = crypto_register_alg(&aesni_cipher_alg);
1257 if (err)
1258 return err;
1259
1260 err = simd_register_skciphers_compat(aesni_skciphers,
1261 ARRAY_SIZE(aesni_skciphers),
1262 aesni_simd_skciphers);
1263 if (err)
1264 goto unregister_cipher;
1265
1266 err = simd_register_aeads_compat(aesni_aeads, ARRAY_SIZE(aesni_aeads),
1267 aesni_simd_aeads);
1268 if (err)
1269 goto unregister_skciphers;
1270
1271#ifdef CONFIG_X86_64
1272 if (boot_cpu_has(X86_FEATURE_AVX))
1273 err = simd_register_skciphers_compat(&aesni_xctr, 1,
1274 &aesni_simd_xctr);
1275 if (err)
1276 goto unregister_aeads;
1277#endif /* CONFIG_X86_64 */
1278
1279 return 0;
1280
1281#ifdef CONFIG_X86_64
1282unregister_aeads:
1283 simd_unregister_aeads(aesni_aeads, ARRAY_SIZE(aesni_aeads),
1284 aesni_simd_aeads);
1285#endif /* CONFIG_X86_64 */
1286
1287unregister_skciphers:
1288 simd_unregister_skciphers(aesni_skciphers, ARRAY_SIZE(aesni_skciphers),
1289 aesni_simd_skciphers);
1290unregister_cipher:
1291 crypto_unregister_alg(&aesni_cipher_alg);
1292 return err;
1293}
1294
1295static void __exit aesni_exit(void)
1296{
1297 simd_unregister_aeads(aesni_aeads, ARRAY_SIZE(aesni_aeads),
1298 aesni_simd_aeads);
1299 simd_unregister_skciphers(aesni_skciphers, ARRAY_SIZE(aesni_skciphers),
1300 aesni_simd_skciphers);
1301 crypto_unregister_alg(&aesni_cipher_alg);
1302#ifdef CONFIG_X86_64
1303 if (boot_cpu_has(X86_FEATURE_AVX))
1304 simd_unregister_skciphers(&aesni_xctr, 1, &aesni_simd_xctr);
1305#endif /* CONFIG_X86_64 */
1306}
1307
1308late_initcall(aesni_init);
1309module_exit(aesni_exit);
1310
1311MODULE_DESCRIPTION("Rijndael (AES) Cipher Algorithm, Intel AES-NI instructions optimized");
1312MODULE_LICENSE("GPL");
1313MODULE_ALIAS_CRYPTO("aes");