xref: /openbmc/linux/drivers/md/dm-crypt.c (revision 95e9fd10)
1 /*
2  * Copyright (C) 2003 Christophe Saout <christophe@saout.de>
3  * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
4  * Copyright (C) 2006-2009 Red Hat, Inc. All rights reserved.
5  *
6  * This file is released under the GPL.
7  */
8 
9 #include <linux/completion.h>
10 #include <linux/err.h>
11 #include <linux/module.h>
12 #include <linux/init.h>
13 #include <linux/kernel.h>
14 #include <linux/bio.h>
15 #include <linux/blkdev.h>
16 #include <linux/mempool.h>
17 #include <linux/slab.h>
18 #include <linux/crypto.h>
19 #include <linux/workqueue.h>
20 #include <linux/backing-dev.h>
21 #include <linux/percpu.h>
22 #include <linux/atomic.h>
23 #include <linux/scatterlist.h>
24 #include <asm/page.h>
25 #include <asm/unaligned.h>
26 #include <crypto/hash.h>
27 #include <crypto/md5.h>
28 #include <crypto/algapi.h>
29 
30 #include <linux/device-mapper.h>
31 
32 #define DM_MSG_PREFIX "crypt"
33 
34 /*
35  * context holding the current state of a multi-part conversion
36  */
37 struct convert_context {
38 	struct completion restart;
39 	struct bio *bio_in;
40 	struct bio *bio_out;
41 	unsigned int offset_in;
42 	unsigned int offset_out;
43 	unsigned int idx_in;
44 	unsigned int idx_out;
45 	sector_t cc_sector;
46 	atomic_t cc_pending;
47 };
48 
49 /*
50  * per bio private data
51  */
52 struct dm_crypt_io {
53 	struct crypt_config *cc;
54 	struct bio *base_bio;
55 	struct work_struct work;
56 
57 	struct convert_context ctx;
58 
59 	atomic_t io_pending;
60 	int error;
61 	sector_t sector;
62 	struct dm_crypt_io *base_io;
63 };
64 
65 struct dm_crypt_request {
66 	struct convert_context *ctx;
67 	struct scatterlist sg_in;
68 	struct scatterlist sg_out;
69 	sector_t iv_sector;
70 };
71 
72 struct crypt_config;
73 
74 struct crypt_iv_operations {
75 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
76 		   const char *opts);
77 	void (*dtr)(struct crypt_config *cc);
78 	int (*init)(struct crypt_config *cc);
79 	int (*wipe)(struct crypt_config *cc);
80 	int (*generator)(struct crypt_config *cc, u8 *iv,
81 			 struct dm_crypt_request *dmreq);
82 	int (*post)(struct crypt_config *cc, u8 *iv,
83 		    struct dm_crypt_request *dmreq);
84 };
85 
86 struct iv_essiv_private {
87 	struct crypto_hash *hash_tfm;
88 	u8 *salt;
89 };
90 
91 struct iv_benbi_private {
92 	int shift;
93 };
94 
95 #define LMK_SEED_SIZE 64 /* hash + 0 */
96 struct iv_lmk_private {
97 	struct crypto_shash *hash_tfm;
98 	u8 *seed;
99 };
100 
101 /*
102  * Crypt: maps a linear range of a block device
103  * and encrypts / decrypts at the same time.
104  */
105 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID };
106 
107 /*
108  * Duplicated per-CPU state for cipher.
109  */
110 struct crypt_cpu {
111 	struct ablkcipher_request *req;
112 };
113 
114 /*
115  * The fields in here must be read only after initialization,
116  * changing state should be in crypt_cpu.
117  */
118 struct crypt_config {
119 	struct dm_dev *dev;
120 	sector_t start;
121 
122 	/*
123 	 * pool for per bio private data, crypto requests and
124 	 * encryption requeusts/buffer pages
125 	 */
126 	mempool_t *io_pool;
127 	mempool_t *req_pool;
128 	mempool_t *page_pool;
129 	struct bio_set *bs;
130 
131 	struct workqueue_struct *io_queue;
132 	struct workqueue_struct *crypt_queue;
133 
134 	char *cipher;
135 	char *cipher_string;
136 
137 	struct crypt_iv_operations *iv_gen_ops;
138 	union {
139 		struct iv_essiv_private essiv;
140 		struct iv_benbi_private benbi;
141 		struct iv_lmk_private lmk;
142 	} iv_gen_private;
143 	sector_t iv_offset;
144 	unsigned int iv_size;
145 
146 	/*
147 	 * Duplicated per cpu state. Access through
148 	 * per_cpu_ptr() only.
149 	 */
150 	struct crypt_cpu __percpu *cpu;
151 
152 	/* ESSIV: struct crypto_cipher *essiv_tfm */
153 	void *iv_private;
154 	struct crypto_ablkcipher **tfms;
155 	unsigned tfms_count;
156 
157 	/*
158 	 * Layout of each crypto request:
159 	 *
160 	 *   struct ablkcipher_request
161 	 *      context
162 	 *      padding
163 	 *   struct dm_crypt_request
164 	 *      padding
165 	 *   IV
166 	 *
167 	 * The padding is added so that dm_crypt_request and the IV are
168 	 * correctly aligned.
169 	 */
170 	unsigned int dmreq_start;
171 
172 	unsigned long flags;
173 	unsigned int key_size;
174 	unsigned int key_parts;
175 	u8 key[0];
176 };
177 
178 #define MIN_IOS        16
179 #define MIN_POOL_PAGES 32
180 
181 static struct kmem_cache *_crypt_io_pool;
182 
183 static void clone_init(struct dm_crypt_io *, struct bio *);
184 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
185 static u8 *iv_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq);
186 
187 static struct crypt_cpu *this_crypt_config(struct crypt_config *cc)
188 {
189 	return this_cpu_ptr(cc->cpu);
190 }
191 
192 /*
193  * Use this to access cipher attributes that are the same for each CPU.
194  */
195 static struct crypto_ablkcipher *any_tfm(struct crypt_config *cc)
196 {
197 	return cc->tfms[0];
198 }
199 
200 /*
201  * Different IV generation algorithms:
202  *
203  * plain: the initial vector is the 32-bit little-endian version of the sector
204  *        number, padded with zeros if necessary.
205  *
206  * plain64: the initial vector is the 64-bit little-endian version of the sector
207  *        number, padded with zeros if necessary.
208  *
209  * essiv: "encrypted sector|salt initial vector", the sector number is
210  *        encrypted with the bulk cipher using a salt as key. The salt
211  *        should be derived from the bulk cipher's key via hashing.
212  *
213  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
214  *        (needed for LRW-32-AES and possible other narrow block modes)
215  *
216  * null: the initial vector is always zero.  Provides compatibility with
217  *       obsolete loop_fish2 devices.  Do not use for new devices.
218  *
219  * lmk:  Compatible implementation of the block chaining mode used
220  *       by the Loop-AES block device encryption system
221  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
222  *       It operates on full 512 byte sectors and uses CBC
223  *       with an IV derived from the sector number, the data and
224  *       optionally extra IV seed.
225  *       This means that after decryption the first block
226  *       of sector must be tweaked according to decrypted data.
227  *       Loop-AES can use three encryption schemes:
228  *         version 1: is plain aes-cbc mode
229  *         version 2: uses 64 multikey scheme with lmk IV generator
230  *         version 3: the same as version 2 with additional IV seed
231  *                   (it uses 65 keys, last key is used as IV seed)
232  *
233  * plumb: unimplemented, see:
234  * http://article.gmane.org/gmane.linux.kernel.device-mapper.dm-crypt/454
235  */
236 
237 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
238 			      struct dm_crypt_request *dmreq)
239 {
240 	memset(iv, 0, cc->iv_size);
241 	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
242 
243 	return 0;
244 }
245 
246 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
247 				struct dm_crypt_request *dmreq)
248 {
249 	memset(iv, 0, cc->iv_size);
250 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
251 
252 	return 0;
253 }
254 
255 /* Initialise ESSIV - compute salt but no local memory allocations */
256 static int crypt_iv_essiv_init(struct crypt_config *cc)
257 {
258 	struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv;
259 	struct hash_desc desc;
260 	struct scatterlist sg;
261 	struct crypto_cipher *essiv_tfm;
262 	int err;
263 
264 	sg_init_one(&sg, cc->key, cc->key_size);
265 	desc.tfm = essiv->hash_tfm;
266 	desc.flags = CRYPTO_TFM_REQ_MAY_SLEEP;
267 
268 	err = crypto_hash_digest(&desc, &sg, cc->key_size, essiv->salt);
269 	if (err)
270 		return err;
271 
272 	essiv_tfm = cc->iv_private;
273 
274 	err = crypto_cipher_setkey(essiv_tfm, essiv->salt,
275 			    crypto_hash_digestsize(essiv->hash_tfm));
276 	if (err)
277 		return err;
278 
279 	return 0;
280 }
281 
282 /* Wipe salt and reset key derived from volume key */
283 static int crypt_iv_essiv_wipe(struct crypt_config *cc)
284 {
285 	struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv;
286 	unsigned salt_size = crypto_hash_digestsize(essiv->hash_tfm);
287 	struct crypto_cipher *essiv_tfm;
288 	int r, err = 0;
289 
290 	memset(essiv->salt, 0, salt_size);
291 
292 	essiv_tfm = cc->iv_private;
293 	r = crypto_cipher_setkey(essiv_tfm, essiv->salt, salt_size);
294 	if (r)
295 		err = r;
296 
297 	return err;
298 }
299 
300 /* Set up per cpu cipher state */
301 static struct crypto_cipher *setup_essiv_cpu(struct crypt_config *cc,
302 					     struct dm_target *ti,
303 					     u8 *salt, unsigned saltsize)
304 {
305 	struct crypto_cipher *essiv_tfm;
306 	int err;
307 
308 	/* Setup the essiv_tfm with the given salt */
309 	essiv_tfm = crypto_alloc_cipher(cc->cipher, 0, CRYPTO_ALG_ASYNC);
310 	if (IS_ERR(essiv_tfm)) {
311 		ti->error = "Error allocating crypto tfm for ESSIV";
312 		return essiv_tfm;
313 	}
314 
315 	if (crypto_cipher_blocksize(essiv_tfm) !=
316 	    crypto_ablkcipher_ivsize(any_tfm(cc))) {
317 		ti->error = "Block size of ESSIV cipher does "
318 			    "not match IV size of block cipher";
319 		crypto_free_cipher(essiv_tfm);
320 		return ERR_PTR(-EINVAL);
321 	}
322 
323 	err = crypto_cipher_setkey(essiv_tfm, salt, saltsize);
324 	if (err) {
325 		ti->error = "Failed to set key for ESSIV cipher";
326 		crypto_free_cipher(essiv_tfm);
327 		return ERR_PTR(err);
328 	}
329 
330 	return essiv_tfm;
331 }
332 
333 static void crypt_iv_essiv_dtr(struct crypt_config *cc)
334 {
335 	struct crypto_cipher *essiv_tfm;
336 	struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv;
337 
338 	crypto_free_hash(essiv->hash_tfm);
339 	essiv->hash_tfm = NULL;
340 
341 	kzfree(essiv->salt);
342 	essiv->salt = NULL;
343 
344 	essiv_tfm = cc->iv_private;
345 
346 	if (essiv_tfm)
347 		crypto_free_cipher(essiv_tfm);
348 
349 	cc->iv_private = NULL;
350 }
351 
352 static int crypt_iv_essiv_ctr(struct crypt_config *cc, struct dm_target *ti,
353 			      const char *opts)
354 {
355 	struct crypto_cipher *essiv_tfm = NULL;
356 	struct crypto_hash *hash_tfm = NULL;
357 	u8 *salt = NULL;
358 	int err;
359 
360 	if (!opts) {
361 		ti->error = "Digest algorithm missing for ESSIV mode";
362 		return -EINVAL;
363 	}
364 
365 	/* Allocate hash algorithm */
366 	hash_tfm = crypto_alloc_hash(opts, 0, CRYPTO_ALG_ASYNC);
367 	if (IS_ERR(hash_tfm)) {
368 		ti->error = "Error initializing ESSIV hash";
369 		err = PTR_ERR(hash_tfm);
370 		goto bad;
371 	}
372 
373 	salt = kzalloc(crypto_hash_digestsize(hash_tfm), GFP_KERNEL);
374 	if (!salt) {
375 		ti->error = "Error kmallocing salt storage in ESSIV";
376 		err = -ENOMEM;
377 		goto bad;
378 	}
379 
380 	cc->iv_gen_private.essiv.salt = salt;
381 	cc->iv_gen_private.essiv.hash_tfm = hash_tfm;
382 
383 	essiv_tfm = setup_essiv_cpu(cc, ti, salt,
384 				crypto_hash_digestsize(hash_tfm));
385 	if (IS_ERR(essiv_tfm)) {
386 		crypt_iv_essiv_dtr(cc);
387 		return PTR_ERR(essiv_tfm);
388 	}
389 	cc->iv_private = essiv_tfm;
390 
391 	return 0;
392 
393 bad:
394 	if (hash_tfm && !IS_ERR(hash_tfm))
395 		crypto_free_hash(hash_tfm);
396 	kfree(salt);
397 	return err;
398 }
399 
400 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
401 			      struct dm_crypt_request *dmreq)
402 {
403 	struct crypto_cipher *essiv_tfm = cc->iv_private;
404 
405 	memset(iv, 0, cc->iv_size);
406 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
407 	crypto_cipher_encrypt_one(essiv_tfm, iv, iv);
408 
409 	return 0;
410 }
411 
412 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
413 			      const char *opts)
414 {
415 	unsigned bs = crypto_ablkcipher_blocksize(any_tfm(cc));
416 	int log = ilog2(bs);
417 
418 	/* we need to calculate how far we must shift the sector count
419 	 * to get the cipher block count, we use this shift in _gen */
420 
421 	if (1 << log != bs) {
422 		ti->error = "cypher blocksize is not a power of 2";
423 		return -EINVAL;
424 	}
425 
426 	if (log > 9) {
427 		ti->error = "cypher blocksize is > 512";
428 		return -EINVAL;
429 	}
430 
431 	cc->iv_gen_private.benbi.shift = 9 - log;
432 
433 	return 0;
434 }
435 
436 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
437 {
438 }
439 
440 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
441 			      struct dm_crypt_request *dmreq)
442 {
443 	__be64 val;
444 
445 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
446 
447 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
448 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
449 
450 	return 0;
451 }
452 
453 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
454 			     struct dm_crypt_request *dmreq)
455 {
456 	memset(iv, 0, cc->iv_size);
457 
458 	return 0;
459 }
460 
461 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
462 {
463 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
464 
465 	if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
466 		crypto_free_shash(lmk->hash_tfm);
467 	lmk->hash_tfm = NULL;
468 
469 	kzfree(lmk->seed);
470 	lmk->seed = NULL;
471 }
472 
473 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
474 			    const char *opts)
475 {
476 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
477 
478 	lmk->hash_tfm = crypto_alloc_shash("md5", 0, 0);
479 	if (IS_ERR(lmk->hash_tfm)) {
480 		ti->error = "Error initializing LMK hash";
481 		return PTR_ERR(lmk->hash_tfm);
482 	}
483 
484 	/* No seed in LMK version 2 */
485 	if (cc->key_parts == cc->tfms_count) {
486 		lmk->seed = NULL;
487 		return 0;
488 	}
489 
490 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
491 	if (!lmk->seed) {
492 		crypt_iv_lmk_dtr(cc);
493 		ti->error = "Error kmallocing seed storage in LMK";
494 		return -ENOMEM;
495 	}
496 
497 	return 0;
498 }
499 
500 static int crypt_iv_lmk_init(struct crypt_config *cc)
501 {
502 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
503 	int subkey_size = cc->key_size / cc->key_parts;
504 
505 	/* LMK seed is on the position of LMK_KEYS + 1 key */
506 	if (lmk->seed)
507 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
508 		       crypto_shash_digestsize(lmk->hash_tfm));
509 
510 	return 0;
511 }
512 
513 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
514 {
515 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
516 
517 	if (lmk->seed)
518 		memset(lmk->seed, 0, LMK_SEED_SIZE);
519 
520 	return 0;
521 }
522 
523 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
524 			    struct dm_crypt_request *dmreq,
525 			    u8 *data)
526 {
527 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
528 	struct {
529 		struct shash_desc desc;
530 		char ctx[crypto_shash_descsize(lmk->hash_tfm)];
531 	} sdesc;
532 	struct md5_state md5state;
533 	u32 buf[4];
534 	int i, r;
535 
536 	sdesc.desc.tfm = lmk->hash_tfm;
537 	sdesc.desc.flags = CRYPTO_TFM_REQ_MAY_SLEEP;
538 
539 	r = crypto_shash_init(&sdesc.desc);
540 	if (r)
541 		return r;
542 
543 	if (lmk->seed) {
544 		r = crypto_shash_update(&sdesc.desc, lmk->seed, LMK_SEED_SIZE);
545 		if (r)
546 			return r;
547 	}
548 
549 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
550 	r = crypto_shash_update(&sdesc.desc, data + 16, 16 * 31);
551 	if (r)
552 		return r;
553 
554 	/* Sector is cropped to 56 bits here */
555 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
556 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
557 	buf[2] = cpu_to_le32(4024);
558 	buf[3] = 0;
559 	r = crypto_shash_update(&sdesc.desc, (u8 *)buf, sizeof(buf));
560 	if (r)
561 		return r;
562 
563 	/* No MD5 padding here */
564 	r = crypto_shash_export(&sdesc.desc, &md5state);
565 	if (r)
566 		return r;
567 
568 	for (i = 0; i < MD5_HASH_WORDS; i++)
569 		__cpu_to_le32s(&md5state.hash[i]);
570 	memcpy(iv, &md5state.hash, cc->iv_size);
571 
572 	return 0;
573 }
574 
575 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
576 			    struct dm_crypt_request *dmreq)
577 {
578 	u8 *src;
579 	int r = 0;
580 
581 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
582 		src = kmap_atomic(sg_page(&dmreq->sg_in));
583 		r = crypt_iv_lmk_one(cc, iv, dmreq, src + dmreq->sg_in.offset);
584 		kunmap_atomic(src);
585 	} else
586 		memset(iv, 0, cc->iv_size);
587 
588 	return r;
589 }
590 
591 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
592 			     struct dm_crypt_request *dmreq)
593 {
594 	u8 *dst;
595 	int r;
596 
597 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
598 		return 0;
599 
600 	dst = kmap_atomic(sg_page(&dmreq->sg_out));
601 	r = crypt_iv_lmk_one(cc, iv, dmreq, dst + dmreq->sg_out.offset);
602 
603 	/* Tweak the first block of plaintext sector */
604 	if (!r)
605 		crypto_xor(dst + dmreq->sg_out.offset, iv, cc->iv_size);
606 
607 	kunmap_atomic(dst);
608 	return r;
609 }
610 
611 static struct crypt_iv_operations crypt_iv_plain_ops = {
612 	.generator = crypt_iv_plain_gen
613 };
614 
615 static struct crypt_iv_operations crypt_iv_plain64_ops = {
616 	.generator = crypt_iv_plain64_gen
617 };
618 
619 static struct crypt_iv_operations crypt_iv_essiv_ops = {
620 	.ctr       = crypt_iv_essiv_ctr,
621 	.dtr       = crypt_iv_essiv_dtr,
622 	.init      = crypt_iv_essiv_init,
623 	.wipe      = crypt_iv_essiv_wipe,
624 	.generator = crypt_iv_essiv_gen
625 };
626 
627 static struct crypt_iv_operations crypt_iv_benbi_ops = {
628 	.ctr	   = crypt_iv_benbi_ctr,
629 	.dtr	   = crypt_iv_benbi_dtr,
630 	.generator = crypt_iv_benbi_gen
631 };
632 
633 static struct crypt_iv_operations crypt_iv_null_ops = {
634 	.generator = crypt_iv_null_gen
635 };
636 
637 static struct crypt_iv_operations crypt_iv_lmk_ops = {
638 	.ctr	   = crypt_iv_lmk_ctr,
639 	.dtr	   = crypt_iv_lmk_dtr,
640 	.init	   = crypt_iv_lmk_init,
641 	.wipe	   = crypt_iv_lmk_wipe,
642 	.generator = crypt_iv_lmk_gen,
643 	.post	   = crypt_iv_lmk_post
644 };
645 
646 static void crypt_convert_init(struct crypt_config *cc,
647 			       struct convert_context *ctx,
648 			       struct bio *bio_out, struct bio *bio_in,
649 			       sector_t sector)
650 {
651 	ctx->bio_in = bio_in;
652 	ctx->bio_out = bio_out;
653 	ctx->offset_in = 0;
654 	ctx->offset_out = 0;
655 	ctx->idx_in = bio_in ? bio_in->bi_idx : 0;
656 	ctx->idx_out = bio_out ? bio_out->bi_idx : 0;
657 	ctx->cc_sector = sector + cc->iv_offset;
658 	init_completion(&ctx->restart);
659 }
660 
661 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
662 					     struct ablkcipher_request *req)
663 {
664 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
665 }
666 
667 static struct ablkcipher_request *req_of_dmreq(struct crypt_config *cc,
668 					       struct dm_crypt_request *dmreq)
669 {
670 	return (struct ablkcipher_request *)((char *)dmreq - cc->dmreq_start);
671 }
672 
673 static u8 *iv_of_dmreq(struct crypt_config *cc,
674 		       struct dm_crypt_request *dmreq)
675 {
676 	return (u8 *)ALIGN((unsigned long)(dmreq + 1),
677 		crypto_ablkcipher_alignmask(any_tfm(cc)) + 1);
678 }
679 
680 static int crypt_convert_block(struct crypt_config *cc,
681 			       struct convert_context *ctx,
682 			       struct ablkcipher_request *req)
683 {
684 	struct bio_vec *bv_in = bio_iovec_idx(ctx->bio_in, ctx->idx_in);
685 	struct bio_vec *bv_out = bio_iovec_idx(ctx->bio_out, ctx->idx_out);
686 	struct dm_crypt_request *dmreq;
687 	u8 *iv;
688 	int r;
689 
690 	dmreq = dmreq_of_req(cc, req);
691 	iv = iv_of_dmreq(cc, dmreq);
692 
693 	dmreq->iv_sector = ctx->cc_sector;
694 	dmreq->ctx = ctx;
695 	sg_init_table(&dmreq->sg_in, 1);
696 	sg_set_page(&dmreq->sg_in, bv_in->bv_page, 1 << SECTOR_SHIFT,
697 		    bv_in->bv_offset + ctx->offset_in);
698 
699 	sg_init_table(&dmreq->sg_out, 1);
700 	sg_set_page(&dmreq->sg_out, bv_out->bv_page, 1 << SECTOR_SHIFT,
701 		    bv_out->bv_offset + ctx->offset_out);
702 
703 	ctx->offset_in += 1 << SECTOR_SHIFT;
704 	if (ctx->offset_in >= bv_in->bv_len) {
705 		ctx->offset_in = 0;
706 		ctx->idx_in++;
707 	}
708 
709 	ctx->offset_out += 1 << SECTOR_SHIFT;
710 	if (ctx->offset_out >= bv_out->bv_len) {
711 		ctx->offset_out = 0;
712 		ctx->idx_out++;
713 	}
714 
715 	if (cc->iv_gen_ops) {
716 		r = cc->iv_gen_ops->generator(cc, iv, dmreq);
717 		if (r < 0)
718 			return r;
719 	}
720 
721 	ablkcipher_request_set_crypt(req, &dmreq->sg_in, &dmreq->sg_out,
722 				     1 << SECTOR_SHIFT, iv);
723 
724 	if (bio_data_dir(ctx->bio_in) == WRITE)
725 		r = crypto_ablkcipher_encrypt(req);
726 	else
727 		r = crypto_ablkcipher_decrypt(req);
728 
729 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
730 		r = cc->iv_gen_ops->post(cc, iv, dmreq);
731 
732 	return r;
733 }
734 
735 static void kcryptd_async_done(struct crypto_async_request *async_req,
736 			       int error);
737 
738 static void crypt_alloc_req(struct crypt_config *cc,
739 			    struct convert_context *ctx)
740 {
741 	struct crypt_cpu *this_cc = this_crypt_config(cc);
742 	unsigned key_index = ctx->cc_sector & (cc->tfms_count - 1);
743 
744 	if (!this_cc->req)
745 		this_cc->req = mempool_alloc(cc->req_pool, GFP_NOIO);
746 
747 	ablkcipher_request_set_tfm(this_cc->req, cc->tfms[key_index]);
748 	ablkcipher_request_set_callback(this_cc->req,
749 	    CRYPTO_TFM_REQ_MAY_BACKLOG | CRYPTO_TFM_REQ_MAY_SLEEP,
750 	    kcryptd_async_done, dmreq_of_req(cc, this_cc->req));
751 }
752 
753 /*
754  * Encrypt / decrypt data from one bio to another one (can be the same one)
755  */
756 static int crypt_convert(struct crypt_config *cc,
757 			 struct convert_context *ctx)
758 {
759 	struct crypt_cpu *this_cc = this_crypt_config(cc);
760 	int r;
761 
762 	atomic_set(&ctx->cc_pending, 1);
763 
764 	while(ctx->idx_in < ctx->bio_in->bi_vcnt &&
765 	      ctx->idx_out < ctx->bio_out->bi_vcnt) {
766 
767 		crypt_alloc_req(cc, ctx);
768 
769 		atomic_inc(&ctx->cc_pending);
770 
771 		r = crypt_convert_block(cc, ctx, this_cc->req);
772 
773 		switch (r) {
774 		/* async */
775 		case -EBUSY:
776 			wait_for_completion(&ctx->restart);
777 			INIT_COMPLETION(ctx->restart);
778 			/* fall through*/
779 		case -EINPROGRESS:
780 			this_cc->req = NULL;
781 			ctx->cc_sector++;
782 			continue;
783 
784 		/* sync */
785 		case 0:
786 			atomic_dec(&ctx->cc_pending);
787 			ctx->cc_sector++;
788 			cond_resched();
789 			continue;
790 
791 		/* error */
792 		default:
793 			atomic_dec(&ctx->cc_pending);
794 			return r;
795 		}
796 	}
797 
798 	return 0;
799 }
800 
801 static void dm_crypt_bio_destructor(struct bio *bio)
802 {
803 	struct dm_crypt_io *io = bio->bi_private;
804 	struct crypt_config *cc = io->cc;
805 
806 	bio_free(bio, cc->bs);
807 }
808 
809 /*
810  * Generate a new unfragmented bio with the given size
811  * This should never violate the device limitations
812  * May return a smaller bio when running out of pages, indicated by
813  * *out_of_pages set to 1.
814  */
815 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size,
816 				      unsigned *out_of_pages)
817 {
818 	struct crypt_config *cc = io->cc;
819 	struct bio *clone;
820 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
821 	gfp_t gfp_mask = GFP_NOIO | __GFP_HIGHMEM;
822 	unsigned i, len;
823 	struct page *page;
824 
825 	clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, cc->bs);
826 	if (!clone)
827 		return NULL;
828 
829 	clone_init(io, clone);
830 	*out_of_pages = 0;
831 
832 	for (i = 0; i < nr_iovecs; i++) {
833 		page = mempool_alloc(cc->page_pool, gfp_mask);
834 		if (!page) {
835 			*out_of_pages = 1;
836 			break;
837 		}
838 
839 		/*
840 		 * If additional pages cannot be allocated without waiting,
841 		 * return a partially-allocated bio.  The caller will then try
842 		 * to allocate more bios while submitting this partial bio.
843 		 */
844 		gfp_mask = (gfp_mask | __GFP_NOWARN) & ~__GFP_WAIT;
845 
846 		len = (size > PAGE_SIZE) ? PAGE_SIZE : size;
847 
848 		if (!bio_add_page(clone, page, len, 0)) {
849 			mempool_free(page, cc->page_pool);
850 			break;
851 		}
852 
853 		size -= len;
854 	}
855 
856 	if (!clone->bi_size) {
857 		bio_put(clone);
858 		return NULL;
859 	}
860 
861 	return clone;
862 }
863 
864 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
865 {
866 	unsigned int i;
867 	struct bio_vec *bv;
868 
869 	for (i = 0; i < clone->bi_vcnt; i++) {
870 		bv = bio_iovec_idx(clone, i);
871 		BUG_ON(!bv->bv_page);
872 		mempool_free(bv->bv_page, cc->page_pool);
873 		bv->bv_page = NULL;
874 	}
875 }
876 
877 static struct dm_crypt_io *crypt_io_alloc(struct crypt_config *cc,
878 					  struct bio *bio, sector_t sector)
879 {
880 	struct dm_crypt_io *io;
881 
882 	io = mempool_alloc(cc->io_pool, GFP_NOIO);
883 	io->cc = cc;
884 	io->base_bio = bio;
885 	io->sector = sector;
886 	io->error = 0;
887 	io->base_io = NULL;
888 	atomic_set(&io->io_pending, 0);
889 
890 	return io;
891 }
892 
893 static void crypt_inc_pending(struct dm_crypt_io *io)
894 {
895 	atomic_inc(&io->io_pending);
896 }
897 
898 /*
899  * One of the bios was finished. Check for completion of
900  * the whole request and correctly clean up the buffer.
901  * If base_io is set, wait for the last fragment to complete.
902  */
903 static void crypt_dec_pending(struct dm_crypt_io *io)
904 {
905 	struct crypt_config *cc = io->cc;
906 	struct bio *base_bio = io->base_bio;
907 	struct dm_crypt_io *base_io = io->base_io;
908 	int error = io->error;
909 
910 	if (!atomic_dec_and_test(&io->io_pending))
911 		return;
912 
913 	mempool_free(io, cc->io_pool);
914 
915 	if (likely(!base_io))
916 		bio_endio(base_bio, error);
917 	else {
918 		if (error && !base_io->error)
919 			base_io->error = error;
920 		crypt_dec_pending(base_io);
921 	}
922 }
923 
924 /*
925  * kcryptd/kcryptd_io:
926  *
927  * Needed because it would be very unwise to do decryption in an
928  * interrupt context.
929  *
930  * kcryptd performs the actual encryption or decryption.
931  *
932  * kcryptd_io performs the IO submission.
933  *
934  * They must be separated as otherwise the final stages could be
935  * starved by new requests which can block in the first stages due
936  * to memory allocation.
937  *
938  * The work is done per CPU global for all dm-crypt instances.
939  * They should not depend on each other and do not block.
940  */
941 static void crypt_endio(struct bio *clone, int error)
942 {
943 	struct dm_crypt_io *io = clone->bi_private;
944 	struct crypt_config *cc = io->cc;
945 	unsigned rw = bio_data_dir(clone);
946 
947 	if (unlikely(!bio_flagged(clone, BIO_UPTODATE) && !error))
948 		error = -EIO;
949 
950 	/*
951 	 * free the processed pages
952 	 */
953 	if (rw == WRITE)
954 		crypt_free_buffer_pages(cc, clone);
955 
956 	bio_put(clone);
957 
958 	if (rw == READ && !error) {
959 		kcryptd_queue_crypt(io);
960 		return;
961 	}
962 
963 	if (unlikely(error))
964 		io->error = error;
965 
966 	crypt_dec_pending(io);
967 }
968 
969 static void clone_init(struct dm_crypt_io *io, struct bio *clone)
970 {
971 	struct crypt_config *cc = io->cc;
972 
973 	clone->bi_private = io;
974 	clone->bi_end_io  = crypt_endio;
975 	clone->bi_bdev    = cc->dev->bdev;
976 	clone->bi_rw      = io->base_bio->bi_rw;
977 	clone->bi_destructor = dm_crypt_bio_destructor;
978 }
979 
980 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
981 {
982 	struct crypt_config *cc = io->cc;
983 	struct bio *base_bio = io->base_bio;
984 	struct bio *clone;
985 
986 	/*
987 	 * The block layer might modify the bvec array, so always
988 	 * copy the required bvecs because we need the original
989 	 * one in order to decrypt the whole bio data *afterwards*.
990 	 */
991 	clone = bio_alloc_bioset(gfp, bio_segments(base_bio), cc->bs);
992 	if (!clone)
993 		return 1;
994 
995 	crypt_inc_pending(io);
996 
997 	clone_init(io, clone);
998 	clone->bi_idx = 0;
999 	clone->bi_vcnt = bio_segments(base_bio);
1000 	clone->bi_size = base_bio->bi_size;
1001 	clone->bi_sector = cc->start + io->sector;
1002 	memcpy(clone->bi_io_vec, bio_iovec(base_bio),
1003 	       sizeof(struct bio_vec) * clone->bi_vcnt);
1004 
1005 	generic_make_request(clone);
1006 	return 0;
1007 }
1008 
1009 static void kcryptd_io_write(struct dm_crypt_io *io)
1010 {
1011 	struct bio *clone = io->ctx.bio_out;
1012 	generic_make_request(clone);
1013 }
1014 
1015 static void kcryptd_io(struct work_struct *work)
1016 {
1017 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1018 
1019 	if (bio_data_dir(io->base_bio) == READ) {
1020 		crypt_inc_pending(io);
1021 		if (kcryptd_io_read(io, GFP_NOIO))
1022 			io->error = -ENOMEM;
1023 		crypt_dec_pending(io);
1024 	} else
1025 		kcryptd_io_write(io);
1026 }
1027 
1028 static void kcryptd_queue_io(struct dm_crypt_io *io)
1029 {
1030 	struct crypt_config *cc = io->cc;
1031 
1032 	INIT_WORK(&io->work, kcryptd_io);
1033 	queue_work(cc->io_queue, &io->work);
1034 }
1035 
1036 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1037 {
1038 	struct bio *clone = io->ctx.bio_out;
1039 	struct crypt_config *cc = io->cc;
1040 
1041 	if (unlikely(io->error < 0)) {
1042 		crypt_free_buffer_pages(cc, clone);
1043 		bio_put(clone);
1044 		crypt_dec_pending(io);
1045 		return;
1046 	}
1047 
1048 	/* crypt_convert should have filled the clone bio */
1049 	BUG_ON(io->ctx.idx_out < clone->bi_vcnt);
1050 
1051 	clone->bi_sector = cc->start + io->sector;
1052 
1053 	if (async)
1054 		kcryptd_queue_io(io);
1055 	else
1056 		generic_make_request(clone);
1057 }
1058 
1059 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
1060 {
1061 	struct crypt_config *cc = io->cc;
1062 	struct bio *clone;
1063 	struct dm_crypt_io *new_io;
1064 	int crypt_finished;
1065 	unsigned out_of_pages = 0;
1066 	unsigned remaining = io->base_bio->bi_size;
1067 	sector_t sector = io->sector;
1068 	int r;
1069 
1070 	/*
1071 	 * Prevent io from disappearing until this function completes.
1072 	 */
1073 	crypt_inc_pending(io);
1074 	crypt_convert_init(cc, &io->ctx, NULL, io->base_bio, sector);
1075 
1076 	/*
1077 	 * The allocated buffers can be smaller than the whole bio,
1078 	 * so repeat the whole process until all the data can be handled.
1079 	 */
1080 	while (remaining) {
1081 		clone = crypt_alloc_buffer(io, remaining, &out_of_pages);
1082 		if (unlikely(!clone)) {
1083 			io->error = -ENOMEM;
1084 			break;
1085 		}
1086 
1087 		io->ctx.bio_out = clone;
1088 		io->ctx.idx_out = 0;
1089 
1090 		remaining -= clone->bi_size;
1091 		sector += bio_sectors(clone);
1092 
1093 		crypt_inc_pending(io);
1094 
1095 		r = crypt_convert(cc, &io->ctx);
1096 		if (r < 0)
1097 			io->error = -EIO;
1098 
1099 		crypt_finished = atomic_dec_and_test(&io->ctx.cc_pending);
1100 
1101 		/* Encryption was already finished, submit io now */
1102 		if (crypt_finished) {
1103 			kcryptd_crypt_write_io_submit(io, 0);
1104 
1105 			/*
1106 			 * If there was an error, do not try next fragments.
1107 			 * For async, error is processed in async handler.
1108 			 */
1109 			if (unlikely(r < 0))
1110 				break;
1111 
1112 			io->sector = sector;
1113 		}
1114 
1115 		/*
1116 		 * Out of memory -> run queues
1117 		 * But don't wait if split was due to the io size restriction
1118 		 */
1119 		if (unlikely(out_of_pages))
1120 			congestion_wait(BLK_RW_ASYNC, HZ/100);
1121 
1122 		/*
1123 		 * With async crypto it is unsafe to share the crypto context
1124 		 * between fragments, so switch to a new dm_crypt_io structure.
1125 		 */
1126 		if (unlikely(!crypt_finished && remaining)) {
1127 			new_io = crypt_io_alloc(io->cc, io->base_bio,
1128 						sector);
1129 			crypt_inc_pending(new_io);
1130 			crypt_convert_init(cc, &new_io->ctx, NULL,
1131 					   io->base_bio, sector);
1132 			new_io->ctx.idx_in = io->ctx.idx_in;
1133 			new_io->ctx.offset_in = io->ctx.offset_in;
1134 
1135 			/*
1136 			 * Fragments after the first use the base_io
1137 			 * pending count.
1138 			 */
1139 			if (!io->base_io)
1140 				new_io->base_io = io;
1141 			else {
1142 				new_io->base_io = io->base_io;
1143 				crypt_inc_pending(io->base_io);
1144 				crypt_dec_pending(io);
1145 			}
1146 
1147 			io = new_io;
1148 		}
1149 	}
1150 
1151 	crypt_dec_pending(io);
1152 }
1153 
1154 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
1155 {
1156 	crypt_dec_pending(io);
1157 }
1158 
1159 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
1160 {
1161 	struct crypt_config *cc = io->cc;
1162 	int r = 0;
1163 
1164 	crypt_inc_pending(io);
1165 
1166 	crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
1167 			   io->sector);
1168 
1169 	r = crypt_convert(cc, &io->ctx);
1170 	if (r < 0)
1171 		io->error = -EIO;
1172 
1173 	if (atomic_dec_and_test(&io->ctx.cc_pending))
1174 		kcryptd_crypt_read_done(io);
1175 
1176 	crypt_dec_pending(io);
1177 }
1178 
1179 static void kcryptd_async_done(struct crypto_async_request *async_req,
1180 			       int error)
1181 {
1182 	struct dm_crypt_request *dmreq = async_req->data;
1183 	struct convert_context *ctx = dmreq->ctx;
1184 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1185 	struct crypt_config *cc = io->cc;
1186 
1187 	if (error == -EINPROGRESS) {
1188 		complete(&ctx->restart);
1189 		return;
1190 	}
1191 
1192 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
1193 		error = cc->iv_gen_ops->post(cc, iv_of_dmreq(cc, dmreq), dmreq);
1194 
1195 	if (error < 0)
1196 		io->error = -EIO;
1197 
1198 	mempool_free(req_of_dmreq(cc, dmreq), cc->req_pool);
1199 
1200 	if (!atomic_dec_and_test(&ctx->cc_pending))
1201 		return;
1202 
1203 	if (bio_data_dir(io->base_bio) == READ)
1204 		kcryptd_crypt_read_done(io);
1205 	else
1206 		kcryptd_crypt_write_io_submit(io, 1);
1207 }
1208 
1209 static void kcryptd_crypt(struct work_struct *work)
1210 {
1211 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1212 
1213 	if (bio_data_dir(io->base_bio) == READ)
1214 		kcryptd_crypt_read_convert(io);
1215 	else
1216 		kcryptd_crypt_write_convert(io);
1217 }
1218 
1219 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
1220 {
1221 	struct crypt_config *cc = io->cc;
1222 
1223 	INIT_WORK(&io->work, kcryptd_crypt);
1224 	queue_work(cc->crypt_queue, &io->work);
1225 }
1226 
1227 /*
1228  * Decode key from its hex representation
1229  */
1230 static int crypt_decode_key(u8 *key, char *hex, unsigned int size)
1231 {
1232 	char buffer[3];
1233 	unsigned int i;
1234 
1235 	buffer[2] = '\0';
1236 
1237 	for (i = 0; i < size; i++) {
1238 		buffer[0] = *hex++;
1239 		buffer[1] = *hex++;
1240 
1241 		if (kstrtou8(buffer, 16, &key[i]))
1242 			return -EINVAL;
1243 	}
1244 
1245 	if (*hex != '\0')
1246 		return -EINVAL;
1247 
1248 	return 0;
1249 }
1250 
1251 /*
1252  * Encode key into its hex representation
1253  */
1254 static void crypt_encode_key(char *hex, u8 *key, unsigned int size)
1255 {
1256 	unsigned int i;
1257 
1258 	for (i = 0; i < size; i++) {
1259 		sprintf(hex, "%02x", *key);
1260 		hex += 2;
1261 		key++;
1262 	}
1263 }
1264 
1265 static void crypt_free_tfms(struct crypt_config *cc)
1266 {
1267 	unsigned i;
1268 
1269 	if (!cc->tfms)
1270 		return;
1271 
1272 	for (i = 0; i < cc->tfms_count; i++)
1273 		if (cc->tfms[i] && !IS_ERR(cc->tfms[i])) {
1274 			crypto_free_ablkcipher(cc->tfms[i]);
1275 			cc->tfms[i] = NULL;
1276 		}
1277 
1278 	kfree(cc->tfms);
1279 	cc->tfms = NULL;
1280 }
1281 
1282 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
1283 {
1284 	unsigned i;
1285 	int err;
1286 
1287 	cc->tfms = kmalloc(cc->tfms_count * sizeof(struct crypto_ablkcipher *),
1288 			   GFP_KERNEL);
1289 	if (!cc->tfms)
1290 		return -ENOMEM;
1291 
1292 	for (i = 0; i < cc->tfms_count; i++) {
1293 		cc->tfms[i] = crypto_alloc_ablkcipher(ciphermode, 0, 0);
1294 		if (IS_ERR(cc->tfms[i])) {
1295 			err = PTR_ERR(cc->tfms[i]);
1296 			crypt_free_tfms(cc);
1297 			return err;
1298 		}
1299 	}
1300 
1301 	return 0;
1302 }
1303 
1304 static int crypt_setkey_allcpus(struct crypt_config *cc)
1305 {
1306 	unsigned subkey_size = cc->key_size >> ilog2(cc->tfms_count);
1307 	int err = 0, i, r;
1308 
1309 	for (i = 0; i < cc->tfms_count; i++) {
1310 		r = crypto_ablkcipher_setkey(cc->tfms[i],
1311 					     cc->key + (i * subkey_size),
1312 					     subkey_size);
1313 		if (r)
1314 			err = r;
1315 	}
1316 
1317 	return err;
1318 }
1319 
1320 static int crypt_set_key(struct crypt_config *cc, char *key)
1321 {
1322 	int r = -EINVAL;
1323 	int key_string_len = strlen(key);
1324 
1325 	/* The key size may not be changed. */
1326 	if (cc->key_size != (key_string_len >> 1))
1327 		goto out;
1328 
1329 	/* Hyphen (which gives a key_size of zero) means there is no key. */
1330 	if (!cc->key_size && strcmp(key, "-"))
1331 		goto out;
1332 
1333 	if (cc->key_size && crypt_decode_key(cc->key, key, cc->key_size) < 0)
1334 		goto out;
1335 
1336 	set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
1337 
1338 	r = crypt_setkey_allcpus(cc);
1339 
1340 out:
1341 	/* Hex key string not needed after here, so wipe it. */
1342 	memset(key, '0', key_string_len);
1343 
1344 	return r;
1345 }
1346 
1347 static int crypt_wipe_key(struct crypt_config *cc)
1348 {
1349 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
1350 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
1351 
1352 	return crypt_setkey_allcpus(cc);
1353 }
1354 
1355 static void crypt_dtr(struct dm_target *ti)
1356 {
1357 	struct crypt_config *cc = ti->private;
1358 	struct crypt_cpu *cpu_cc;
1359 	int cpu;
1360 
1361 	ti->private = NULL;
1362 
1363 	if (!cc)
1364 		return;
1365 
1366 	if (cc->io_queue)
1367 		destroy_workqueue(cc->io_queue);
1368 	if (cc->crypt_queue)
1369 		destroy_workqueue(cc->crypt_queue);
1370 
1371 	if (cc->cpu)
1372 		for_each_possible_cpu(cpu) {
1373 			cpu_cc = per_cpu_ptr(cc->cpu, cpu);
1374 			if (cpu_cc->req)
1375 				mempool_free(cpu_cc->req, cc->req_pool);
1376 		}
1377 
1378 	crypt_free_tfms(cc);
1379 
1380 	if (cc->bs)
1381 		bioset_free(cc->bs);
1382 
1383 	if (cc->page_pool)
1384 		mempool_destroy(cc->page_pool);
1385 	if (cc->req_pool)
1386 		mempool_destroy(cc->req_pool);
1387 	if (cc->io_pool)
1388 		mempool_destroy(cc->io_pool);
1389 
1390 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
1391 		cc->iv_gen_ops->dtr(cc);
1392 
1393 	if (cc->dev)
1394 		dm_put_device(ti, cc->dev);
1395 
1396 	if (cc->cpu)
1397 		free_percpu(cc->cpu);
1398 
1399 	kzfree(cc->cipher);
1400 	kzfree(cc->cipher_string);
1401 
1402 	/* Must zero key material before freeing */
1403 	kzfree(cc);
1404 }
1405 
1406 static int crypt_ctr_cipher(struct dm_target *ti,
1407 			    char *cipher_in, char *key)
1408 {
1409 	struct crypt_config *cc = ti->private;
1410 	char *tmp, *cipher, *chainmode, *ivmode, *ivopts, *keycount;
1411 	char *cipher_api = NULL;
1412 	int ret = -EINVAL;
1413 	char dummy;
1414 
1415 	/* Convert to crypto api definition? */
1416 	if (strchr(cipher_in, '(')) {
1417 		ti->error = "Bad cipher specification";
1418 		return -EINVAL;
1419 	}
1420 
1421 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
1422 	if (!cc->cipher_string)
1423 		goto bad_mem;
1424 
1425 	/*
1426 	 * Legacy dm-crypt cipher specification
1427 	 * cipher[:keycount]-mode-iv:ivopts
1428 	 */
1429 	tmp = cipher_in;
1430 	keycount = strsep(&tmp, "-");
1431 	cipher = strsep(&keycount, ":");
1432 
1433 	if (!keycount)
1434 		cc->tfms_count = 1;
1435 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
1436 		 !is_power_of_2(cc->tfms_count)) {
1437 		ti->error = "Bad cipher key count specification";
1438 		return -EINVAL;
1439 	}
1440 	cc->key_parts = cc->tfms_count;
1441 
1442 	cc->cipher = kstrdup(cipher, GFP_KERNEL);
1443 	if (!cc->cipher)
1444 		goto bad_mem;
1445 
1446 	chainmode = strsep(&tmp, "-");
1447 	ivopts = strsep(&tmp, "-");
1448 	ivmode = strsep(&ivopts, ":");
1449 
1450 	if (tmp)
1451 		DMWARN("Ignoring unexpected additional cipher options");
1452 
1453 	cc->cpu = __alloc_percpu(sizeof(*(cc->cpu)),
1454 				 __alignof__(struct crypt_cpu));
1455 	if (!cc->cpu) {
1456 		ti->error = "Cannot allocate per cpu state";
1457 		goto bad_mem;
1458 	}
1459 
1460 	/*
1461 	 * For compatibility with the original dm-crypt mapping format, if
1462 	 * only the cipher name is supplied, use cbc-plain.
1463 	 */
1464 	if (!chainmode || (!strcmp(chainmode, "plain") && !ivmode)) {
1465 		chainmode = "cbc";
1466 		ivmode = "plain";
1467 	}
1468 
1469 	if (strcmp(chainmode, "ecb") && !ivmode) {
1470 		ti->error = "IV mechanism required";
1471 		return -EINVAL;
1472 	}
1473 
1474 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
1475 	if (!cipher_api)
1476 		goto bad_mem;
1477 
1478 	ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
1479 		       "%s(%s)", chainmode, cipher);
1480 	if (ret < 0) {
1481 		kfree(cipher_api);
1482 		goto bad_mem;
1483 	}
1484 
1485 	/* Allocate cipher */
1486 	ret = crypt_alloc_tfms(cc, cipher_api);
1487 	if (ret < 0) {
1488 		ti->error = "Error allocating crypto tfm";
1489 		goto bad;
1490 	}
1491 
1492 	/* Initialize and set key */
1493 	ret = crypt_set_key(cc, key);
1494 	if (ret < 0) {
1495 		ti->error = "Error decoding and setting key";
1496 		goto bad;
1497 	}
1498 
1499 	/* Initialize IV */
1500 	cc->iv_size = crypto_ablkcipher_ivsize(any_tfm(cc));
1501 	if (cc->iv_size)
1502 		/* at least a 64 bit sector number should fit in our buffer */
1503 		cc->iv_size = max(cc->iv_size,
1504 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
1505 	else if (ivmode) {
1506 		DMWARN("Selected cipher does not support IVs");
1507 		ivmode = NULL;
1508 	}
1509 
1510 	/* Choose ivmode, see comments at iv code. */
1511 	if (ivmode == NULL)
1512 		cc->iv_gen_ops = NULL;
1513 	else if (strcmp(ivmode, "plain") == 0)
1514 		cc->iv_gen_ops = &crypt_iv_plain_ops;
1515 	else if (strcmp(ivmode, "plain64") == 0)
1516 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
1517 	else if (strcmp(ivmode, "essiv") == 0)
1518 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
1519 	else if (strcmp(ivmode, "benbi") == 0)
1520 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
1521 	else if (strcmp(ivmode, "null") == 0)
1522 		cc->iv_gen_ops = &crypt_iv_null_ops;
1523 	else if (strcmp(ivmode, "lmk") == 0) {
1524 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
1525 		/* Version 2 and 3 is recognised according
1526 		 * to length of provided multi-key string.
1527 		 * If present (version 3), last key is used as IV seed.
1528 		 */
1529 		if (cc->key_size % cc->key_parts)
1530 			cc->key_parts++;
1531 	} else {
1532 		ret = -EINVAL;
1533 		ti->error = "Invalid IV mode";
1534 		goto bad;
1535 	}
1536 
1537 	/* Allocate IV */
1538 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
1539 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
1540 		if (ret < 0) {
1541 			ti->error = "Error creating IV";
1542 			goto bad;
1543 		}
1544 	}
1545 
1546 	/* Initialize IV (set keys for ESSIV etc) */
1547 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
1548 		ret = cc->iv_gen_ops->init(cc);
1549 		if (ret < 0) {
1550 			ti->error = "Error initialising IV";
1551 			goto bad;
1552 		}
1553 	}
1554 
1555 	ret = 0;
1556 bad:
1557 	kfree(cipher_api);
1558 	return ret;
1559 
1560 bad_mem:
1561 	ti->error = "Cannot allocate cipher strings";
1562 	return -ENOMEM;
1563 }
1564 
1565 /*
1566  * Construct an encryption mapping:
1567  * <cipher> <key> <iv_offset> <dev_path> <start>
1568  */
1569 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1570 {
1571 	struct crypt_config *cc;
1572 	unsigned int key_size, opt_params;
1573 	unsigned long long tmpll;
1574 	int ret;
1575 	struct dm_arg_set as;
1576 	const char *opt_string;
1577 	char dummy;
1578 
1579 	static struct dm_arg _args[] = {
1580 		{0, 1, "Invalid number of feature args"},
1581 	};
1582 
1583 	if (argc < 5) {
1584 		ti->error = "Not enough arguments";
1585 		return -EINVAL;
1586 	}
1587 
1588 	key_size = strlen(argv[1]) >> 1;
1589 
1590 	cc = kzalloc(sizeof(*cc) + key_size * sizeof(u8), GFP_KERNEL);
1591 	if (!cc) {
1592 		ti->error = "Cannot allocate encryption context";
1593 		return -ENOMEM;
1594 	}
1595 	cc->key_size = key_size;
1596 
1597 	ti->private = cc;
1598 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
1599 	if (ret < 0)
1600 		goto bad;
1601 
1602 	ret = -ENOMEM;
1603 	cc->io_pool = mempool_create_slab_pool(MIN_IOS, _crypt_io_pool);
1604 	if (!cc->io_pool) {
1605 		ti->error = "Cannot allocate crypt io mempool";
1606 		goto bad;
1607 	}
1608 
1609 	cc->dmreq_start = sizeof(struct ablkcipher_request);
1610 	cc->dmreq_start += crypto_ablkcipher_reqsize(any_tfm(cc));
1611 	cc->dmreq_start = ALIGN(cc->dmreq_start, crypto_tfm_ctx_alignment());
1612 	cc->dmreq_start += crypto_ablkcipher_alignmask(any_tfm(cc)) &
1613 			   ~(crypto_tfm_ctx_alignment() - 1);
1614 
1615 	cc->req_pool = mempool_create_kmalloc_pool(MIN_IOS, cc->dmreq_start +
1616 			sizeof(struct dm_crypt_request) + cc->iv_size);
1617 	if (!cc->req_pool) {
1618 		ti->error = "Cannot allocate crypt request mempool";
1619 		goto bad;
1620 	}
1621 
1622 	cc->page_pool = mempool_create_page_pool(MIN_POOL_PAGES, 0);
1623 	if (!cc->page_pool) {
1624 		ti->error = "Cannot allocate page mempool";
1625 		goto bad;
1626 	}
1627 
1628 	cc->bs = bioset_create(MIN_IOS, 0);
1629 	if (!cc->bs) {
1630 		ti->error = "Cannot allocate crypt bioset";
1631 		goto bad;
1632 	}
1633 
1634 	ret = -EINVAL;
1635 	if (sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) {
1636 		ti->error = "Invalid iv_offset sector";
1637 		goto bad;
1638 	}
1639 	cc->iv_offset = tmpll;
1640 
1641 	if (dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev)) {
1642 		ti->error = "Device lookup failed";
1643 		goto bad;
1644 	}
1645 
1646 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1) {
1647 		ti->error = "Invalid device sector";
1648 		goto bad;
1649 	}
1650 	cc->start = tmpll;
1651 
1652 	argv += 5;
1653 	argc -= 5;
1654 
1655 	/* Optional parameters */
1656 	if (argc) {
1657 		as.argc = argc;
1658 		as.argv = argv;
1659 
1660 		ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
1661 		if (ret)
1662 			goto bad;
1663 
1664 		opt_string = dm_shift_arg(&as);
1665 
1666 		if (opt_params == 1 && opt_string &&
1667 		    !strcasecmp(opt_string, "allow_discards"))
1668 			ti->num_discard_requests = 1;
1669 		else if (opt_params) {
1670 			ret = -EINVAL;
1671 			ti->error = "Invalid feature arguments";
1672 			goto bad;
1673 		}
1674 	}
1675 
1676 	ret = -ENOMEM;
1677 	cc->io_queue = alloc_workqueue("kcryptd_io",
1678 				       WQ_NON_REENTRANT|
1679 				       WQ_MEM_RECLAIM,
1680 				       1);
1681 	if (!cc->io_queue) {
1682 		ti->error = "Couldn't create kcryptd io queue";
1683 		goto bad;
1684 	}
1685 
1686 	cc->crypt_queue = alloc_workqueue("kcryptd",
1687 					  WQ_NON_REENTRANT|
1688 					  WQ_CPU_INTENSIVE|
1689 					  WQ_MEM_RECLAIM,
1690 					  1);
1691 	if (!cc->crypt_queue) {
1692 		ti->error = "Couldn't create kcryptd queue";
1693 		goto bad;
1694 	}
1695 
1696 	ti->num_flush_requests = 1;
1697 	ti->discard_zeroes_data_unsupported = true;
1698 
1699 	return 0;
1700 
1701 bad:
1702 	crypt_dtr(ti);
1703 	return ret;
1704 }
1705 
1706 static int crypt_map(struct dm_target *ti, struct bio *bio,
1707 		     union map_info *map_context)
1708 {
1709 	struct dm_crypt_io *io;
1710 	struct crypt_config *cc = ti->private;
1711 
1712 	/*
1713 	 * If bio is REQ_FLUSH or REQ_DISCARD, just bypass crypt queues.
1714 	 * - for REQ_FLUSH device-mapper core ensures that no IO is in-flight
1715 	 * - for REQ_DISCARD caller must use flush if IO ordering matters
1716 	 */
1717 	if (unlikely(bio->bi_rw & (REQ_FLUSH | REQ_DISCARD))) {
1718 		bio->bi_bdev = cc->dev->bdev;
1719 		if (bio_sectors(bio))
1720 			bio->bi_sector = cc->start + dm_target_offset(ti, bio->bi_sector);
1721 		return DM_MAPIO_REMAPPED;
1722 	}
1723 
1724 	io = crypt_io_alloc(cc, bio, dm_target_offset(ti, bio->bi_sector));
1725 
1726 	if (bio_data_dir(io->base_bio) == READ) {
1727 		if (kcryptd_io_read(io, GFP_NOWAIT))
1728 			kcryptd_queue_io(io);
1729 	} else
1730 		kcryptd_queue_crypt(io);
1731 
1732 	return DM_MAPIO_SUBMITTED;
1733 }
1734 
1735 static int crypt_status(struct dm_target *ti, status_type_t type,
1736 			unsigned status_flags, char *result, unsigned maxlen)
1737 {
1738 	struct crypt_config *cc = ti->private;
1739 	unsigned int sz = 0;
1740 
1741 	switch (type) {
1742 	case STATUSTYPE_INFO:
1743 		result[0] = '\0';
1744 		break;
1745 
1746 	case STATUSTYPE_TABLE:
1747 		DMEMIT("%s ", cc->cipher_string);
1748 
1749 		if (cc->key_size > 0) {
1750 			if ((maxlen - sz) < ((cc->key_size << 1) + 1))
1751 				return -ENOMEM;
1752 
1753 			crypt_encode_key(result + sz, cc->key, cc->key_size);
1754 			sz += cc->key_size << 1;
1755 		} else {
1756 			if (sz >= maxlen)
1757 				return -ENOMEM;
1758 			result[sz++] = '-';
1759 		}
1760 
1761 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
1762 				cc->dev->name, (unsigned long long)cc->start);
1763 
1764 		if (ti->num_discard_requests)
1765 			DMEMIT(" 1 allow_discards");
1766 
1767 		break;
1768 	}
1769 	return 0;
1770 }
1771 
1772 static void crypt_postsuspend(struct dm_target *ti)
1773 {
1774 	struct crypt_config *cc = ti->private;
1775 
1776 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
1777 }
1778 
1779 static int crypt_preresume(struct dm_target *ti)
1780 {
1781 	struct crypt_config *cc = ti->private;
1782 
1783 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
1784 		DMERR("aborting resume - crypt key is not set.");
1785 		return -EAGAIN;
1786 	}
1787 
1788 	return 0;
1789 }
1790 
1791 static void crypt_resume(struct dm_target *ti)
1792 {
1793 	struct crypt_config *cc = ti->private;
1794 
1795 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
1796 }
1797 
1798 /* Message interface
1799  *	key set <key>
1800  *	key wipe
1801  */
1802 static int crypt_message(struct dm_target *ti, unsigned argc, char **argv)
1803 {
1804 	struct crypt_config *cc = ti->private;
1805 	int ret = -EINVAL;
1806 
1807 	if (argc < 2)
1808 		goto error;
1809 
1810 	if (!strcasecmp(argv[0], "key")) {
1811 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
1812 			DMWARN("not suspended during key manipulation.");
1813 			return -EINVAL;
1814 		}
1815 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
1816 			ret = crypt_set_key(cc, argv[2]);
1817 			if (ret)
1818 				return ret;
1819 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
1820 				ret = cc->iv_gen_ops->init(cc);
1821 			return ret;
1822 		}
1823 		if (argc == 2 && !strcasecmp(argv[1], "wipe")) {
1824 			if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
1825 				ret = cc->iv_gen_ops->wipe(cc);
1826 				if (ret)
1827 					return ret;
1828 			}
1829 			return crypt_wipe_key(cc);
1830 		}
1831 	}
1832 
1833 error:
1834 	DMWARN("unrecognised message received.");
1835 	return -EINVAL;
1836 }
1837 
1838 static int crypt_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
1839 		       struct bio_vec *biovec, int max_size)
1840 {
1841 	struct crypt_config *cc = ti->private;
1842 	struct request_queue *q = bdev_get_queue(cc->dev->bdev);
1843 
1844 	if (!q->merge_bvec_fn)
1845 		return max_size;
1846 
1847 	bvm->bi_bdev = cc->dev->bdev;
1848 	bvm->bi_sector = cc->start + dm_target_offset(ti, bvm->bi_sector);
1849 
1850 	return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
1851 }
1852 
1853 static int crypt_iterate_devices(struct dm_target *ti,
1854 				 iterate_devices_callout_fn fn, void *data)
1855 {
1856 	struct crypt_config *cc = ti->private;
1857 
1858 	return fn(ti, cc->dev, cc->start, ti->len, data);
1859 }
1860 
1861 static struct target_type crypt_target = {
1862 	.name   = "crypt",
1863 	.version = {1, 11, 0},
1864 	.module = THIS_MODULE,
1865 	.ctr    = crypt_ctr,
1866 	.dtr    = crypt_dtr,
1867 	.map    = crypt_map,
1868 	.status = crypt_status,
1869 	.postsuspend = crypt_postsuspend,
1870 	.preresume = crypt_preresume,
1871 	.resume = crypt_resume,
1872 	.message = crypt_message,
1873 	.merge  = crypt_merge,
1874 	.iterate_devices = crypt_iterate_devices,
1875 };
1876 
1877 static int __init dm_crypt_init(void)
1878 {
1879 	int r;
1880 
1881 	_crypt_io_pool = KMEM_CACHE(dm_crypt_io, 0);
1882 	if (!_crypt_io_pool)
1883 		return -ENOMEM;
1884 
1885 	r = dm_register_target(&crypt_target);
1886 	if (r < 0) {
1887 		DMERR("register failed %d", r);
1888 		kmem_cache_destroy(_crypt_io_pool);
1889 	}
1890 
1891 	return r;
1892 }
1893 
1894 static void __exit dm_crypt_exit(void)
1895 {
1896 	dm_unregister_target(&crypt_target);
1897 	kmem_cache_destroy(_crypt_io_pool);
1898 }
1899 
1900 module_init(dm_crypt_init);
1901 module_exit(dm_crypt_exit);
1902 
1903 MODULE_AUTHOR("Christophe Saout <christophe@saout.de>");
1904 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
1905 MODULE_LICENSE("GPL");
1906