xref: /openbmc/linux/drivers/md/dm-crypt.c (revision 565d76cb)
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 <asm/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 #define MESG_STR(x) x, sizeof(x)
34 
35 /*
36  * context holding the current state of a multi-part conversion
37  */
38 struct convert_context {
39 	struct completion restart;
40 	struct bio *bio_in;
41 	struct bio *bio_out;
42 	unsigned int offset_in;
43 	unsigned int offset_out;
44 	unsigned int idx_in;
45 	unsigned int idx_out;
46 	sector_t sector;
47 	atomic_t pending;
48 };
49 
50 /*
51  * per bio private data
52  */
53 struct dm_crypt_io {
54 	struct dm_target *target;
55 	struct bio *base_bio;
56 	struct work_struct work;
57 
58 	struct convert_context ctx;
59 
60 	atomic_t pending;
61 	int error;
62 	sector_t sector;
63 	struct dm_crypt_io *base_io;
64 };
65 
66 struct dm_crypt_request {
67 	struct convert_context *ctx;
68 	struct scatterlist sg_in;
69 	struct scatterlist sg_out;
70 	sector_t iv_sector;
71 };
72 
73 struct crypt_config;
74 
75 struct crypt_iv_operations {
76 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
77 		   const char *opts);
78 	void (*dtr)(struct crypt_config *cc);
79 	int (*init)(struct crypt_config *cc);
80 	int (*wipe)(struct crypt_config *cc);
81 	int (*generator)(struct crypt_config *cc, u8 *iv,
82 			 struct dm_crypt_request *dmreq);
83 	int (*post)(struct crypt_config *cc, u8 *iv,
84 		    struct dm_crypt_request *dmreq);
85 };
86 
87 struct iv_essiv_private {
88 	struct crypto_hash *hash_tfm;
89 	u8 *salt;
90 };
91 
92 struct iv_benbi_private {
93 	int shift;
94 };
95 
96 #define LMK_SEED_SIZE 64 /* hash + 0 */
97 struct iv_lmk_private {
98 	struct crypto_shash *hash_tfm;
99 	u8 *seed;
100 };
101 
102 /*
103  * Crypt: maps a linear range of a block device
104  * and encrypts / decrypts at the same time.
105  */
106 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID };
107 
108 /*
109  * Duplicated per-CPU state for cipher.
110  */
111 struct crypt_cpu {
112 	struct ablkcipher_request *req;
113 	/* ESSIV: struct crypto_cipher *essiv_tfm */
114 	void *iv_private;
115 	struct crypto_ablkcipher *tfms[0];
116 };
117 
118 /*
119  * The fields in here must be read only after initialization,
120  * changing state should be in crypt_cpu.
121  */
122 struct crypt_config {
123 	struct dm_dev *dev;
124 	sector_t start;
125 
126 	/*
127 	 * pool for per bio private data, crypto requests and
128 	 * encryption requeusts/buffer pages
129 	 */
130 	mempool_t *io_pool;
131 	mempool_t *req_pool;
132 	mempool_t *page_pool;
133 	struct bio_set *bs;
134 
135 	struct workqueue_struct *io_queue;
136 	struct workqueue_struct *crypt_queue;
137 
138 	char *cipher;
139 	char *cipher_string;
140 
141 	struct crypt_iv_operations *iv_gen_ops;
142 	union {
143 		struct iv_essiv_private essiv;
144 		struct iv_benbi_private benbi;
145 		struct iv_lmk_private lmk;
146 	} iv_gen_private;
147 	sector_t iv_offset;
148 	unsigned int iv_size;
149 
150 	/*
151 	 * Duplicated per cpu state. Access through
152 	 * per_cpu_ptr() only.
153 	 */
154 	struct crypt_cpu __percpu *cpu;
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 #define MIN_BIO_PAGES  8
181 
182 static struct kmem_cache *_crypt_io_pool;
183 
184 static void clone_init(struct dm_crypt_io *, struct bio *);
185 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
186 static u8 *iv_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq);
187 
188 static struct crypt_cpu *this_crypt_config(struct crypt_config *cc)
189 {
190 	return this_cpu_ptr(cc->cpu);
191 }
192 
193 /*
194  * Use this to access cipher attributes that are the same for each CPU.
195  */
196 static struct crypto_ablkcipher *any_tfm(struct crypt_config *cc)
197 {
198 	return __this_cpu_ptr(cc->cpu)->tfms[0];
199 }
200 
201 /*
202  * Different IV generation algorithms:
203  *
204  * plain: the initial vector is the 32-bit little-endian version of the sector
205  *        number, padded with zeros if necessary.
206  *
207  * plain64: the initial vector is the 64-bit little-endian version of the sector
208  *        number, padded with zeros if necessary.
209  *
210  * essiv: "encrypted sector|salt initial vector", the sector number is
211  *        encrypted with the bulk cipher using a salt as key. The salt
212  *        should be derived from the bulk cipher's key via hashing.
213  *
214  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
215  *        (needed for LRW-32-AES and possible other narrow block modes)
216  *
217  * null: the initial vector is always zero.  Provides compatibility with
218  *       obsolete loop_fish2 devices.  Do not use for new devices.
219  *
220  * lmk:  Compatible implementation of the block chaining mode used
221  *       by the Loop-AES block device encryption system
222  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
223  *       It operates on full 512 byte sectors and uses CBC
224  *       with an IV derived from the sector number, the data and
225  *       optionally extra IV seed.
226  *       This means that after decryption the first block
227  *       of sector must be tweaked according to decrypted data.
228  *       Loop-AES can use three encryption schemes:
229  *         version 1: is plain aes-cbc mode
230  *         version 2: uses 64 multikey scheme with lmk IV generator
231  *         version 3: the same as version 2 with additional IV seed
232  *                   (it uses 65 keys, last key is used as IV seed)
233  *
234  * plumb: unimplemented, see:
235  * http://article.gmane.org/gmane.linux.kernel.device-mapper.dm-crypt/454
236  */
237 
238 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
239 			      struct dm_crypt_request *dmreq)
240 {
241 	memset(iv, 0, cc->iv_size);
242 	*(u32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
243 
244 	return 0;
245 }
246 
247 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
248 				struct dm_crypt_request *dmreq)
249 {
250 	memset(iv, 0, cc->iv_size);
251 	*(u64 *)iv = cpu_to_le64(dmreq->iv_sector);
252 
253 	return 0;
254 }
255 
256 /* Initialise ESSIV - compute salt but no local memory allocations */
257 static int crypt_iv_essiv_init(struct crypt_config *cc)
258 {
259 	struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv;
260 	struct hash_desc desc;
261 	struct scatterlist sg;
262 	struct crypto_cipher *essiv_tfm;
263 	int err, cpu;
264 
265 	sg_init_one(&sg, cc->key, cc->key_size);
266 	desc.tfm = essiv->hash_tfm;
267 	desc.flags = CRYPTO_TFM_REQ_MAY_SLEEP;
268 
269 	err = crypto_hash_digest(&desc, &sg, cc->key_size, essiv->salt);
270 	if (err)
271 		return err;
272 
273 	for_each_possible_cpu(cpu) {
274 		essiv_tfm = per_cpu_ptr(cc->cpu, cpu)->iv_private,
275 
276 		err = crypto_cipher_setkey(essiv_tfm, essiv->salt,
277 				    crypto_hash_digestsize(essiv->hash_tfm));
278 		if (err)
279 			return err;
280 	}
281 
282 	return 0;
283 }
284 
285 /* Wipe salt and reset key derived from volume key */
286 static int crypt_iv_essiv_wipe(struct crypt_config *cc)
287 {
288 	struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv;
289 	unsigned salt_size = crypto_hash_digestsize(essiv->hash_tfm);
290 	struct crypto_cipher *essiv_tfm;
291 	int cpu, r, err = 0;
292 
293 	memset(essiv->salt, 0, salt_size);
294 
295 	for_each_possible_cpu(cpu) {
296 		essiv_tfm = per_cpu_ptr(cc->cpu, cpu)->iv_private;
297 		r = crypto_cipher_setkey(essiv_tfm, essiv->salt, salt_size);
298 		if (r)
299 			err = r;
300 	}
301 
302 	return err;
303 }
304 
305 /* Set up per cpu cipher state */
306 static struct crypto_cipher *setup_essiv_cpu(struct crypt_config *cc,
307 					     struct dm_target *ti,
308 					     u8 *salt, unsigned saltsize)
309 {
310 	struct crypto_cipher *essiv_tfm;
311 	int err;
312 
313 	/* Setup the essiv_tfm with the given salt */
314 	essiv_tfm = crypto_alloc_cipher(cc->cipher, 0, CRYPTO_ALG_ASYNC);
315 	if (IS_ERR(essiv_tfm)) {
316 		ti->error = "Error allocating crypto tfm for ESSIV";
317 		return essiv_tfm;
318 	}
319 
320 	if (crypto_cipher_blocksize(essiv_tfm) !=
321 	    crypto_ablkcipher_ivsize(any_tfm(cc))) {
322 		ti->error = "Block size of ESSIV cipher does "
323 			    "not match IV size of block cipher";
324 		crypto_free_cipher(essiv_tfm);
325 		return ERR_PTR(-EINVAL);
326 	}
327 
328 	err = crypto_cipher_setkey(essiv_tfm, salt, saltsize);
329 	if (err) {
330 		ti->error = "Failed to set key for ESSIV cipher";
331 		crypto_free_cipher(essiv_tfm);
332 		return ERR_PTR(err);
333 	}
334 
335 	return essiv_tfm;
336 }
337 
338 static void crypt_iv_essiv_dtr(struct crypt_config *cc)
339 {
340 	int cpu;
341 	struct crypt_cpu *cpu_cc;
342 	struct crypto_cipher *essiv_tfm;
343 	struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv;
344 
345 	crypto_free_hash(essiv->hash_tfm);
346 	essiv->hash_tfm = NULL;
347 
348 	kzfree(essiv->salt);
349 	essiv->salt = NULL;
350 
351 	for_each_possible_cpu(cpu) {
352 		cpu_cc = per_cpu_ptr(cc->cpu, cpu);
353 		essiv_tfm = cpu_cc->iv_private;
354 
355 		if (essiv_tfm)
356 			crypto_free_cipher(essiv_tfm);
357 
358 		cpu_cc->iv_private = NULL;
359 	}
360 }
361 
362 static int crypt_iv_essiv_ctr(struct crypt_config *cc, struct dm_target *ti,
363 			      const char *opts)
364 {
365 	struct crypto_cipher *essiv_tfm = NULL;
366 	struct crypto_hash *hash_tfm = NULL;
367 	u8 *salt = NULL;
368 	int err, cpu;
369 
370 	if (!opts) {
371 		ti->error = "Digest algorithm missing for ESSIV mode";
372 		return -EINVAL;
373 	}
374 
375 	/* Allocate hash algorithm */
376 	hash_tfm = crypto_alloc_hash(opts, 0, CRYPTO_ALG_ASYNC);
377 	if (IS_ERR(hash_tfm)) {
378 		ti->error = "Error initializing ESSIV hash";
379 		err = PTR_ERR(hash_tfm);
380 		goto bad;
381 	}
382 
383 	salt = kzalloc(crypto_hash_digestsize(hash_tfm), GFP_KERNEL);
384 	if (!salt) {
385 		ti->error = "Error kmallocing salt storage in ESSIV";
386 		err = -ENOMEM;
387 		goto bad;
388 	}
389 
390 	cc->iv_gen_private.essiv.salt = salt;
391 	cc->iv_gen_private.essiv.hash_tfm = hash_tfm;
392 
393 	for_each_possible_cpu(cpu) {
394 		essiv_tfm = setup_essiv_cpu(cc, ti, salt,
395 					crypto_hash_digestsize(hash_tfm));
396 		if (IS_ERR(essiv_tfm)) {
397 			crypt_iv_essiv_dtr(cc);
398 			return PTR_ERR(essiv_tfm);
399 		}
400 		per_cpu_ptr(cc->cpu, cpu)->iv_private = essiv_tfm;
401 	}
402 
403 	return 0;
404 
405 bad:
406 	if (hash_tfm && !IS_ERR(hash_tfm))
407 		crypto_free_hash(hash_tfm);
408 	kfree(salt);
409 	return err;
410 }
411 
412 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
413 			      struct dm_crypt_request *dmreq)
414 {
415 	struct crypto_cipher *essiv_tfm = this_crypt_config(cc)->iv_private;
416 
417 	memset(iv, 0, cc->iv_size);
418 	*(u64 *)iv = cpu_to_le64(dmreq->iv_sector);
419 	crypto_cipher_encrypt_one(essiv_tfm, iv, iv);
420 
421 	return 0;
422 }
423 
424 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
425 			      const char *opts)
426 {
427 	unsigned bs = crypto_ablkcipher_blocksize(any_tfm(cc));
428 	int log = ilog2(bs);
429 
430 	/* we need to calculate how far we must shift the sector count
431 	 * to get the cipher block count, we use this shift in _gen */
432 
433 	if (1 << log != bs) {
434 		ti->error = "cypher blocksize is not a power of 2";
435 		return -EINVAL;
436 	}
437 
438 	if (log > 9) {
439 		ti->error = "cypher blocksize is > 512";
440 		return -EINVAL;
441 	}
442 
443 	cc->iv_gen_private.benbi.shift = 9 - log;
444 
445 	return 0;
446 }
447 
448 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
449 {
450 }
451 
452 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
453 			      struct dm_crypt_request *dmreq)
454 {
455 	__be64 val;
456 
457 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
458 
459 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
460 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
461 
462 	return 0;
463 }
464 
465 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
466 			     struct dm_crypt_request *dmreq)
467 {
468 	memset(iv, 0, cc->iv_size);
469 
470 	return 0;
471 }
472 
473 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
474 {
475 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
476 
477 	if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
478 		crypto_free_shash(lmk->hash_tfm);
479 	lmk->hash_tfm = NULL;
480 
481 	kzfree(lmk->seed);
482 	lmk->seed = NULL;
483 }
484 
485 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
486 			    const char *opts)
487 {
488 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
489 
490 	lmk->hash_tfm = crypto_alloc_shash("md5", 0, 0);
491 	if (IS_ERR(lmk->hash_tfm)) {
492 		ti->error = "Error initializing LMK hash";
493 		return PTR_ERR(lmk->hash_tfm);
494 	}
495 
496 	/* No seed in LMK version 2 */
497 	if (cc->key_parts == cc->tfms_count) {
498 		lmk->seed = NULL;
499 		return 0;
500 	}
501 
502 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
503 	if (!lmk->seed) {
504 		crypt_iv_lmk_dtr(cc);
505 		ti->error = "Error kmallocing seed storage in LMK";
506 		return -ENOMEM;
507 	}
508 
509 	return 0;
510 }
511 
512 static int crypt_iv_lmk_init(struct crypt_config *cc)
513 {
514 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
515 	int subkey_size = cc->key_size / cc->key_parts;
516 
517 	/* LMK seed is on the position of LMK_KEYS + 1 key */
518 	if (lmk->seed)
519 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
520 		       crypto_shash_digestsize(lmk->hash_tfm));
521 
522 	return 0;
523 }
524 
525 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
526 {
527 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
528 
529 	if (lmk->seed)
530 		memset(lmk->seed, 0, LMK_SEED_SIZE);
531 
532 	return 0;
533 }
534 
535 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
536 			    struct dm_crypt_request *dmreq,
537 			    u8 *data)
538 {
539 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
540 	struct {
541 		struct shash_desc desc;
542 		char ctx[crypto_shash_descsize(lmk->hash_tfm)];
543 	} sdesc;
544 	struct md5_state md5state;
545 	u32 buf[4];
546 	int i, r;
547 
548 	sdesc.desc.tfm = lmk->hash_tfm;
549 	sdesc.desc.flags = CRYPTO_TFM_REQ_MAY_SLEEP;
550 
551 	r = crypto_shash_init(&sdesc.desc);
552 	if (r)
553 		return r;
554 
555 	if (lmk->seed) {
556 		r = crypto_shash_update(&sdesc.desc, lmk->seed, LMK_SEED_SIZE);
557 		if (r)
558 			return r;
559 	}
560 
561 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
562 	r = crypto_shash_update(&sdesc.desc, data + 16, 16 * 31);
563 	if (r)
564 		return r;
565 
566 	/* Sector is cropped to 56 bits here */
567 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
568 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
569 	buf[2] = cpu_to_le32(4024);
570 	buf[3] = 0;
571 	r = crypto_shash_update(&sdesc.desc, (u8 *)buf, sizeof(buf));
572 	if (r)
573 		return r;
574 
575 	/* No MD5 padding here */
576 	r = crypto_shash_export(&sdesc.desc, &md5state);
577 	if (r)
578 		return r;
579 
580 	for (i = 0; i < MD5_HASH_WORDS; i++)
581 		__cpu_to_le32s(&md5state.hash[i]);
582 	memcpy(iv, &md5state.hash, cc->iv_size);
583 
584 	return 0;
585 }
586 
587 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
588 			    struct dm_crypt_request *dmreq)
589 {
590 	u8 *src;
591 	int r = 0;
592 
593 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
594 		src = kmap_atomic(sg_page(&dmreq->sg_in), KM_USER0);
595 		r = crypt_iv_lmk_one(cc, iv, dmreq, src + dmreq->sg_in.offset);
596 		kunmap_atomic(src, KM_USER0);
597 	} else
598 		memset(iv, 0, cc->iv_size);
599 
600 	return r;
601 }
602 
603 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
604 			     struct dm_crypt_request *dmreq)
605 {
606 	u8 *dst;
607 	int r;
608 
609 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
610 		return 0;
611 
612 	dst = kmap_atomic(sg_page(&dmreq->sg_out), KM_USER0);
613 	r = crypt_iv_lmk_one(cc, iv, dmreq, dst + dmreq->sg_out.offset);
614 
615 	/* Tweak the first block of plaintext sector */
616 	if (!r)
617 		crypto_xor(dst + dmreq->sg_out.offset, iv, cc->iv_size);
618 
619 	kunmap_atomic(dst, KM_USER0);
620 	return r;
621 }
622 
623 static struct crypt_iv_operations crypt_iv_plain_ops = {
624 	.generator = crypt_iv_plain_gen
625 };
626 
627 static struct crypt_iv_operations crypt_iv_plain64_ops = {
628 	.generator = crypt_iv_plain64_gen
629 };
630 
631 static struct crypt_iv_operations crypt_iv_essiv_ops = {
632 	.ctr       = crypt_iv_essiv_ctr,
633 	.dtr       = crypt_iv_essiv_dtr,
634 	.init      = crypt_iv_essiv_init,
635 	.wipe      = crypt_iv_essiv_wipe,
636 	.generator = crypt_iv_essiv_gen
637 };
638 
639 static struct crypt_iv_operations crypt_iv_benbi_ops = {
640 	.ctr	   = crypt_iv_benbi_ctr,
641 	.dtr	   = crypt_iv_benbi_dtr,
642 	.generator = crypt_iv_benbi_gen
643 };
644 
645 static struct crypt_iv_operations crypt_iv_null_ops = {
646 	.generator = crypt_iv_null_gen
647 };
648 
649 static struct crypt_iv_operations crypt_iv_lmk_ops = {
650 	.ctr	   = crypt_iv_lmk_ctr,
651 	.dtr	   = crypt_iv_lmk_dtr,
652 	.init	   = crypt_iv_lmk_init,
653 	.wipe	   = crypt_iv_lmk_wipe,
654 	.generator = crypt_iv_lmk_gen,
655 	.post	   = crypt_iv_lmk_post
656 };
657 
658 static void crypt_convert_init(struct crypt_config *cc,
659 			       struct convert_context *ctx,
660 			       struct bio *bio_out, struct bio *bio_in,
661 			       sector_t sector)
662 {
663 	ctx->bio_in = bio_in;
664 	ctx->bio_out = bio_out;
665 	ctx->offset_in = 0;
666 	ctx->offset_out = 0;
667 	ctx->idx_in = bio_in ? bio_in->bi_idx : 0;
668 	ctx->idx_out = bio_out ? bio_out->bi_idx : 0;
669 	ctx->sector = sector + cc->iv_offset;
670 	init_completion(&ctx->restart);
671 }
672 
673 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
674 					     struct ablkcipher_request *req)
675 {
676 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
677 }
678 
679 static struct ablkcipher_request *req_of_dmreq(struct crypt_config *cc,
680 					       struct dm_crypt_request *dmreq)
681 {
682 	return (struct ablkcipher_request *)((char *)dmreq - cc->dmreq_start);
683 }
684 
685 static u8 *iv_of_dmreq(struct crypt_config *cc,
686 		       struct dm_crypt_request *dmreq)
687 {
688 	return (u8 *)ALIGN((unsigned long)(dmreq + 1),
689 		crypto_ablkcipher_alignmask(any_tfm(cc)) + 1);
690 }
691 
692 static int crypt_convert_block(struct crypt_config *cc,
693 			       struct convert_context *ctx,
694 			       struct ablkcipher_request *req)
695 {
696 	struct bio_vec *bv_in = bio_iovec_idx(ctx->bio_in, ctx->idx_in);
697 	struct bio_vec *bv_out = bio_iovec_idx(ctx->bio_out, ctx->idx_out);
698 	struct dm_crypt_request *dmreq;
699 	u8 *iv;
700 	int r = 0;
701 
702 	dmreq = dmreq_of_req(cc, req);
703 	iv = iv_of_dmreq(cc, dmreq);
704 
705 	dmreq->iv_sector = ctx->sector;
706 	dmreq->ctx = ctx;
707 	sg_init_table(&dmreq->sg_in, 1);
708 	sg_set_page(&dmreq->sg_in, bv_in->bv_page, 1 << SECTOR_SHIFT,
709 		    bv_in->bv_offset + ctx->offset_in);
710 
711 	sg_init_table(&dmreq->sg_out, 1);
712 	sg_set_page(&dmreq->sg_out, bv_out->bv_page, 1 << SECTOR_SHIFT,
713 		    bv_out->bv_offset + ctx->offset_out);
714 
715 	ctx->offset_in += 1 << SECTOR_SHIFT;
716 	if (ctx->offset_in >= bv_in->bv_len) {
717 		ctx->offset_in = 0;
718 		ctx->idx_in++;
719 	}
720 
721 	ctx->offset_out += 1 << SECTOR_SHIFT;
722 	if (ctx->offset_out >= bv_out->bv_len) {
723 		ctx->offset_out = 0;
724 		ctx->idx_out++;
725 	}
726 
727 	if (cc->iv_gen_ops) {
728 		r = cc->iv_gen_ops->generator(cc, iv, dmreq);
729 		if (r < 0)
730 			return r;
731 	}
732 
733 	ablkcipher_request_set_crypt(req, &dmreq->sg_in, &dmreq->sg_out,
734 				     1 << SECTOR_SHIFT, iv);
735 
736 	if (bio_data_dir(ctx->bio_in) == WRITE)
737 		r = crypto_ablkcipher_encrypt(req);
738 	else
739 		r = crypto_ablkcipher_decrypt(req);
740 
741 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
742 		r = cc->iv_gen_ops->post(cc, iv, dmreq);
743 
744 	return r;
745 }
746 
747 static void kcryptd_async_done(struct crypto_async_request *async_req,
748 			       int error);
749 
750 static void crypt_alloc_req(struct crypt_config *cc,
751 			    struct convert_context *ctx)
752 {
753 	struct crypt_cpu *this_cc = this_crypt_config(cc);
754 	unsigned key_index = ctx->sector & (cc->tfms_count - 1);
755 
756 	if (!this_cc->req)
757 		this_cc->req = mempool_alloc(cc->req_pool, GFP_NOIO);
758 
759 	ablkcipher_request_set_tfm(this_cc->req, this_cc->tfms[key_index]);
760 	ablkcipher_request_set_callback(this_cc->req,
761 	    CRYPTO_TFM_REQ_MAY_BACKLOG | CRYPTO_TFM_REQ_MAY_SLEEP,
762 	    kcryptd_async_done, dmreq_of_req(cc, this_cc->req));
763 }
764 
765 /*
766  * Encrypt / decrypt data from one bio to another one (can be the same one)
767  */
768 static int crypt_convert(struct crypt_config *cc,
769 			 struct convert_context *ctx)
770 {
771 	struct crypt_cpu *this_cc = this_crypt_config(cc);
772 	int r;
773 
774 	atomic_set(&ctx->pending, 1);
775 
776 	while(ctx->idx_in < ctx->bio_in->bi_vcnt &&
777 	      ctx->idx_out < ctx->bio_out->bi_vcnt) {
778 
779 		crypt_alloc_req(cc, ctx);
780 
781 		atomic_inc(&ctx->pending);
782 
783 		r = crypt_convert_block(cc, ctx, this_cc->req);
784 
785 		switch (r) {
786 		/* async */
787 		case -EBUSY:
788 			wait_for_completion(&ctx->restart);
789 			INIT_COMPLETION(ctx->restart);
790 			/* fall through*/
791 		case -EINPROGRESS:
792 			this_cc->req = NULL;
793 			ctx->sector++;
794 			continue;
795 
796 		/* sync */
797 		case 0:
798 			atomic_dec(&ctx->pending);
799 			ctx->sector++;
800 			cond_resched();
801 			continue;
802 
803 		/* error */
804 		default:
805 			atomic_dec(&ctx->pending);
806 			return r;
807 		}
808 	}
809 
810 	return 0;
811 }
812 
813 static void dm_crypt_bio_destructor(struct bio *bio)
814 {
815 	struct dm_crypt_io *io = bio->bi_private;
816 	struct crypt_config *cc = io->target->private;
817 
818 	bio_free(bio, cc->bs);
819 }
820 
821 /*
822  * Generate a new unfragmented bio with the given size
823  * This should never violate the device limitations
824  * May return a smaller bio when running out of pages, indicated by
825  * *out_of_pages set to 1.
826  */
827 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size,
828 				      unsigned *out_of_pages)
829 {
830 	struct crypt_config *cc = io->target->private;
831 	struct bio *clone;
832 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
833 	gfp_t gfp_mask = GFP_NOIO | __GFP_HIGHMEM;
834 	unsigned i, len;
835 	struct page *page;
836 
837 	clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, cc->bs);
838 	if (!clone)
839 		return NULL;
840 
841 	clone_init(io, clone);
842 	*out_of_pages = 0;
843 
844 	for (i = 0; i < nr_iovecs; i++) {
845 		page = mempool_alloc(cc->page_pool, gfp_mask);
846 		if (!page) {
847 			*out_of_pages = 1;
848 			break;
849 		}
850 
851 		/*
852 		 * if additional pages cannot be allocated without waiting,
853 		 * return a partially allocated bio, the caller will then try
854 		 * to allocate additional bios while submitting this partial bio
855 		 */
856 		if (i == (MIN_BIO_PAGES - 1))
857 			gfp_mask = (gfp_mask | __GFP_NOWARN) & ~__GFP_WAIT;
858 
859 		len = (size > PAGE_SIZE) ? PAGE_SIZE : size;
860 
861 		if (!bio_add_page(clone, page, len, 0)) {
862 			mempool_free(page, cc->page_pool);
863 			break;
864 		}
865 
866 		size -= len;
867 	}
868 
869 	if (!clone->bi_size) {
870 		bio_put(clone);
871 		return NULL;
872 	}
873 
874 	return clone;
875 }
876 
877 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
878 {
879 	unsigned int i;
880 	struct bio_vec *bv;
881 
882 	for (i = 0; i < clone->bi_vcnt; i++) {
883 		bv = bio_iovec_idx(clone, i);
884 		BUG_ON(!bv->bv_page);
885 		mempool_free(bv->bv_page, cc->page_pool);
886 		bv->bv_page = NULL;
887 	}
888 }
889 
890 static struct dm_crypt_io *crypt_io_alloc(struct dm_target *ti,
891 					  struct bio *bio, sector_t sector)
892 {
893 	struct crypt_config *cc = ti->private;
894 	struct dm_crypt_io *io;
895 
896 	io = mempool_alloc(cc->io_pool, GFP_NOIO);
897 	io->target = ti;
898 	io->base_bio = bio;
899 	io->sector = sector;
900 	io->error = 0;
901 	io->base_io = NULL;
902 	atomic_set(&io->pending, 0);
903 
904 	return io;
905 }
906 
907 static void crypt_inc_pending(struct dm_crypt_io *io)
908 {
909 	atomic_inc(&io->pending);
910 }
911 
912 /*
913  * One of the bios was finished. Check for completion of
914  * the whole request and correctly clean up the buffer.
915  * If base_io is set, wait for the last fragment to complete.
916  */
917 static void crypt_dec_pending(struct dm_crypt_io *io)
918 {
919 	struct crypt_config *cc = io->target->private;
920 	struct bio *base_bio = io->base_bio;
921 	struct dm_crypt_io *base_io = io->base_io;
922 	int error = io->error;
923 
924 	if (!atomic_dec_and_test(&io->pending))
925 		return;
926 
927 	mempool_free(io, cc->io_pool);
928 
929 	if (likely(!base_io))
930 		bio_endio(base_bio, error);
931 	else {
932 		if (error && !base_io->error)
933 			base_io->error = error;
934 		crypt_dec_pending(base_io);
935 	}
936 }
937 
938 /*
939  * kcryptd/kcryptd_io:
940  *
941  * Needed because it would be very unwise to do decryption in an
942  * interrupt context.
943  *
944  * kcryptd performs the actual encryption or decryption.
945  *
946  * kcryptd_io performs the IO submission.
947  *
948  * They must be separated as otherwise the final stages could be
949  * starved by new requests which can block in the first stages due
950  * to memory allocation.
951  *
952  * The work is done per CPU global for all dm-crypt instances.
953  * They should not depend on each other and do not block.
954  */
955 static void crypt_endio(struct bio *clone, int error)
956 {
957 	struct dm_crypt_io *io = clone->bi_private;
958 	struct crypt_config *cc = io->target->private;
959 	unsigned rw = bio_data_dir(clone);
960 
961 	if (unlikely(!bio_flagged(clone, BIO_UPTODATE) && !error))
962 		error = -EIO;
963 
964 	/*
965 	 * free the processed pages
966 	 */
967 	if (rw == WRITE)
968 		crypt_free_buffer_pages(cc, clone);
969 
970 	bio_put(clone);
971 
972 	if (rw == READ && !error) {
973 		kcryptd_queue_crypt(io);
974 		return;
975 	}
976 
977 	if (unlikely(error))
978 		io->error = error;
979 
980 	crypt_dec_pending(io);
981 }
982 
983 static void clone_init(struct dm_crypt_io *io, struct bio *clone)
984 {
985 	struct crypt_config *cc = io->target->private;
986 
987 	clone->bi_private = io;
988 	clone->bi_end_io  = crypt_endio;
989 	clone->bi_bdev    = cc->dev->bdev;
990 	clone->bi_rw      = io->base_bio->bi_rw;
991 	clone->bi_destructor = dm_crypt_bio_destructor;
992 }
993 
994 static void kcryptd_unplug(struct crypt_config *cc)
995 {
996 	blk_unplug(bdev_get_queue(cc->dev->bdev));
997 }
998 
999 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1000 {
1001 	struct crypt_config *cc = io->target->private;
1002 	struct bio *base_bio = io->base_bio;
1003 	struct bio *clone;
1004 
1005 	/*
1006 	 * The block layer might modify the bvec array, so always
1007 	 * copy the required bvecs because we need the original
1008 	 * one in order to decrypt the whole bio data *afterwards*.
1009 	 */
1010 	clone = bio_alloc_bioset(gfp, bio_segments(base_bio), cc->bs);
1011 	if (!clone) {
1012 		kcryptd_unplug(cc);
1013 		return 1;
1014 	}
1015 
1016 	crypt_inc_pending(io);
1017 
1018 	clone_init(io, clone);
1019 	clone->bi_idx = 0;
1020 	clone->bi_vcnt = bio_segments(base_bio);
1021 	clone->bi_size = base_bio->bi_size;
1022 	clone->bi_sector = cc->start + io->sector;
1023 	memcpy(clone->bi_io_vec, bio_iovec(base_bio),
1024 	       sizeof(struct bio_vec) * clone->bi_vcnt);
1025 
1026 	generic_make_request(clone);
1027 	return 0;
1028 }
1029 
1030 static void kcryptd_io_write(struct dm_crypt_io *io)
1031 {
1032 	struct bio *clone = io->ctx.bio_out;
1033 	generic_make_request(clone);
1034 }
1035 
1036 static void kcryptd_io(struct work_struct *work)
1037 {
1038 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1039 
1040 	if (bio_data_dir(io->base_bio) == READ) {
1041 		crypt_inc_pending(io);
1042 		if (kcryptd_io_read(io, GFP_NOIO))
1043 			io->error = -ENOMEM;
1044 		crypt_dec_pending(io);
1045 	} else
1046 		kcryptd_io_write(io);
1047 }
1048 
1049 static void kcryptd_queue_io(struct dm_crypt_io *io)
1050 {
1051 	struct crypt_config *cc = io->target->private;
1052 
1053 	INIT_WORK(&io->work, kcryptd_io);
1054 	queue_work(cc->io_queue, &io->work);
1055 }
1056 
1057 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io,
1058 					  int error, int async)
1059 {
1060 	struct bio *clone = io->ctx.bio_out;
1061 	struct crypt_config *cc = io->target->private;
1062 
1063 	if (unlikely(error < 0)) {
1064 		crypt_free_buffer_pages(cc, clone);
1065 		bio_put(clone);
1066 		io->error = -EIO;
1067 		crypt_dec_pending(io);
1068 		return;
1069 	}
1070 
1071 	/* crypt_convert should have filled the clone bio */
1072 	BUG_ON(io->ctx.idx_out < clone->bi_vcnt);
1073 
1074 	clone->bi_sector = cc->start + io->sector;
1075 
1076 	if (async)
1077 		kcryptd_queue_io(io);
1078 	else
1079 		generic_make_request(clone);
1080 }
1081 
1082 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
1083 {
1084 	struct crypt_config *cc = io->target->private;
1085 	struct bio *clone;
1086 	struct dm_crypt_io *new_io;
1087 	int crypt_finished;
1088 	unsigned out_of_pages = 0;
1089 	unsigned remaining = io->base_bio->bi_size;
1090 	sector_t sector = io->sector;
1091 	int r;
1092 
1093 	/*
1094 	 * Prevent io from disappearing until this function completes.
1095 	 */
1096 	crypt_inc_pending(io);
1097 	crypt_convert_init(cc, &io->ctx, NULL, io->base_bio, sector);
1098 
1099 	/*
1100 	 * The allocated buffers can be smaller than the whole bio,
1101 	 * so repeat the whole process until all the data can be handled.
1102 	 */
1103 	while (remaining) {
1104 		clone = crypt_alloc_buffer(io, remaining, &out_of_pages);
1105 		if (unlikely(!clone)) {
1106 			io->error = -ENOMEM;
1107 			break;
1108 		}
1109 
1110 		io->ctx.bio_out = clone;
1111 		io->ctx.idx_out = 0;
1112 
1113 		remaining -= clone->bi_size;
1114 		sector += bio_sectors(clone);
1115 
1116 		crypt_inc_pending(io);
1117 		r = crypt_convert(cc, &io->ctx);
1118 		crypt_finished = atomic_dec_and_test(&io->ctx.pending);
1119 
1120 		/* Encryption was already finished, submit io now */
1121 		if (crypt_finished) {
1122 			kcryptd_crypt_write_io_submit(io, r, 0);
1123 
1124 			/*
1125 			 * If there was an error, do not try next fragments.
1126 			 * For async, error is processed in async handler.
1127 			 */
1128 			if (unlikely(r < 0))
1129 				break;
1130 
1131 			io->sector = sector;
1132 		}
1133 
1134 		/*
1135 		 * Out of memory -> run queues
1136 		 * But don't wait if split was due to the io size restriction
1137 		 */
1138 		if (unlikely(out_of_pages))
1139 			congestion_wait(BLK_RW_ASYNC, HZ/100);
1140 
1141 		/*
1142 		 * With async crypto it is unsafe to share the crypto context
1143 		 * between fragments, so switch to a new dm_crypt_io structure.
1144 		 */
1145 		if (unlikely(!crypt_finished && remaining)) {
1146 			new_io = crypt_io_alloc(io->target, io->base_bio,
1147 						sector);
1148 			crypt_inc_pending(new_io);
1149 			crypt_convert_init(cc, &new_io->ctx, NULL,
1150 					   io->base_bio, sector);
1151 			new_io->ctx.idx_in = io->ctx.idx_in;
1152 			new_io->ctx.offset_in = io->ctx.offset_in;
1153 
1154 			/*
1155 			 * Fragments after the first use the base_io
1156 			 * pending count.
1157 			 */
1158 			if (!io->base_io)
1159 				new_io->base_io = io;
1160 			else {
1161 				new_io->base_io = io->base_io;
1162 				crypt_inc_pending(io->base_io);
1163 				crypt_dec_pending(io);
1164 			}
1165 
1166 			io = new_io;
1167 		}
1168 	}
1169 
1170 	crypt_dec_pending(io);
1171 }
1172 
1173 static void kcryptd_crypt_read_done(struct dm_crypt_io *io, int error)
1174 {
1175 	if (unlikely(error < 0))
1176 		io->error = -EIO;
1177 
1178 	crypt_dec_pending(io);
1179 }
1180 
1181 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
1182 {
1183 	struct crypt_config *cc = io->target->private;
1184 	int r = 0;
1185 
1186 	crypt_inc_pending(io);
1187 
1188 	crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
1189 			   io->sector);
1190 
1191 	r = crypt_convert(cc, &io->ctx);
1192 
1193 	if (atomic_dec_and_test(&io->ctx.pending))
1194 		kcryptd_crypt_read_done(io, r);
1195 
1196 	crypt_dec_pending(io);
1197 }
1198 
1199 static void kcryptd_async_done(struct crypto_async_request *async_req,
1200 			       int error)
1201 {
1202 	struct dm_crypt_request *dmreq = async_req->data;
1203 	struct convert_context *ctx = dmreq->ctx;
1204 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1205 	struct crypt_config *cc = io->target->private;
1206 
1207 	if (error == -EINPROGRESS) {
1208 		complete(&ctx->restart);
1209 		return;
1210 	}
1211 
1212 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
1213 		error = cc->iv_gen_ops->post(cc, iv_of_dmreq(cc, dmreq), dmreq);
1214 
1215 	mempool_free(req_of_dmreq(cc, dmreq), cc->req_pool);
1216 
1217 	if (!atomic_dec_and_test(&ctx->pending))
1218 		return;
1219 
1220 	if (bio_data_dir(io->base_bio) == READ)
1221 		kcryptd_crypt_read_done(io, error);
1222 	else
1223 		kcryptd_crypt_write_io_submit(io, error, 1);
1224 }
1225 
1226 static void kcryptd_crypt(struct work_struct *work)
1227 {
1228 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1229 
1230 	if (bio_data_dir(io->base_bio) == READ)
1231 		kcryptd_crypt_read_convert(io);
1232 	else
1233 		kcryptd_crypt_write_convert(io);
1234 }
1235 
1236 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
1237 {
1238 	struct crypt_config *cc = io->target->private;
1239 
1240 	INIT_WORK(&io->work, kcryptd_crypt);
1241 	queue_work(cc->crypt_queue, &io->work);
1242 }
1243 
1244 /*
1245  * Decode key from its hex representation
1246  */
1247 static int crypt_decode_key(u8 *key, char *hex, unsigned int size)
1248 {
1249 	char buffer[3];
1250 	char *endp;
1251 	unsigned int i;
1252 
1253 	buffer[2] = '\0';
1254 
1255 	for (i = 0; i < size; i++) {
1256 		buffer[0] = *hex++;
1257 		buffer[1] = *hex++;
1258 
1259 		key[i] = (u8)simple_strtoul(buffer, &endp, 16);
1260 
1261 		if (endp != &buffer[2])
1262 			return -EINVAL;
1263 	}
1264 
1265 	if (*hex != '\0')
1266 		return -EINVAL;
1267 
1268 	return 0;
1269 }
1270 
1271 /*
1272  * Encode key into its hex representation
1273  */
1274 static void crypt_encode_key(char *hex, u8 *key, unsigned int size)
1275 {
1276 	unsigned int i;
1277 
1278 	for (i = 0; i < size; i++) {
1279 		sprintf(hex, "%02x", *key);
1280 		hex += 2;
1281 		key++;
1282 	}
1283 }
1284 
1285 static void crypt_free_tfms(struct crypt_config *cc, int cpu)
1286 {
1287 	struct crypt_cpu *cpu_cc = per_cpu_ptr(cc->cpu, cpu);
1288 	unsigned i;
1289 
1290 	for (i = 0; i < cc->tfms_count; i++)
1291 		if (cpu_cc->tfms[i] && !IS_ERR(cpu_cc->tfms[i])) {
1292 			crypto_free_ablkcipher(cpu_cc->tfms[i]);
1293 			cpu_cc->tfms[i] = NULL;
1294 		}
1295 }
1296 
1297 static int crypt_alloc_tfms(struct crypt_config *cc, int cpu, char *ciphermode)
1298 {
1299 	struct crypt_cpu *cpu_cc = per_cpu_ptr(cc->cpu, cpu);
1300 	unsigned i;
1301 	int err;
1302 
1303 	for (i = 0; i < cc->tfms_count; i++) {
1304 		cpu_cc->tfms[i] = crypto_alloc_ablkcipher(ciphermode, 0, 0);
1305 		if (IS_ERR(cpu_cc->tfms[i])) {
1306 			err = PTR_ERR(cpu_cc->tfms[i]);
1307 			crypt_free_tfms(cc, cpu);
1308 			return err;
1309 		}
1310 	}
1311 
1312 	return 0;
1313 }
1314 
1315 static int crypt_setkey_allcpus(struct crypt_config *cc)
1316 {
1317 	unsigned subkey_size = cc->key_size >> ilog2(cc->tfms_count);
1318 	int cpu, err = 0, i, r;
1319 
1320 	for_each_possible_cpu(cpu) {
1321 		for (i = 0; i < cc->tfms_count; i++) {
1322 			r = crypto_ablkcipher_setkey(per_cpu_ptr(cc->cpu, cpu)->tfms[i],
1323 						     cc->key + (i * subkey_size), subkey_size);
1324 			if (r)
1325 				err = r;
1326 		}
1327 	}
1328 
1329 	return err;
1330 }
1331 
1332 static int crypt_set_key(struct crypt_config *cc, char *key)
1333 {
1334 	/* The key size may not be changed. */
1335 	if (cc->key_size != (strlen(key) >> 1))
1336 		return -EINVAL;
1337 
1338 	/* Hyphen (which gives a key_size of zero) means there is no key. */
1339 	if (!cc->key_size && strcmp(key, "-"))
1340 		return -EINVAL;
1341 
1342 	if (cc->key_size && crypt_decode_key(cc->key, key, cc->key_size) < 0)
1343 		return -EINVAL;
1344 
1345 	set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
1346 
1347 	return crypt_setkey_allcpus(cc);
1348 }
1349 
1350 static int crypt_wipe_key(struct crypt_config *cc)
1351 {
1352 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
1353 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
1354 
1355 	return crypt_setkey_allcpus(cc);
1356 }
1357 
1358 static void crypt_dtr(struct dm_target *ti)
1359 {
1360 	struct crypt_config *cc = ti->private;
1361 	struct crypt_cpu *cpu_cc;
1362 	int cpu;
1363 
1364 	ti->private = NULL;
1365 
1366 	if (!cc)
1367 		return;
1368 
1369 	if (cc->io_queue)
1370 		destroy_workqueue(cc->io_queue);
1371 	if (cc->crypt_queue)
1372 		destroy_workqueue(cc->crypt_queue);
1373 
1374 	if (cc->cpu)
1375 		for_each_possible_cpu(cpu) {
1376 			cpu_cc = per_cpu_ptr(cc->cpu, cpu);
1377 			if (cpu_cc->req)
1378 				mempool_free(cpu_cc->req, cc->req_pool);
1379 			crypt_free_tfms(cc, cpu);
1380 		}
1381 
1382 	if (cc->bs)
1383 		bioset_free(cc->bs);
1384 
1385 	if (cc->page_pool)
1386 		mempool_destroy(cc->page_pool);
1387 	if (cc->req_pool)
1388 		mempool_destroy(cc->req_pool);
1389 	if (cc->io_pool)
1390 		mempool_destroy(cc->io_pool);
1391 
1392 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
1393 		cc->iv_gen_ops->dtr(cc);
1394 
1395 	if (cc->dev)
1396 		dm_put_device(ti, cc->dev);
1397 
1398 	if (cc->cpu)
1399 		free_percpu(cc->cpu);
1400 
1401 	kzfree(cc->cipher);
1402 	kzfree(cc->cipher_string);
1403 
1404 	/* Must zero key material before freeing */
1405 	kzfree(cc);
1406 }
1407 
1408 static int crypt_ctr_cipher(struct dm_target *ti,
1409 			    char *cipher_in, char *key)
1410 {
1411 	struct crypt_config *cc = ti->private;
1412 	char *tmp, *cipher, *chainmode, *ivmode, *ivopts, *keycount;
1413 	char *cipher_api = NULL;
1414 	int cpu, ret = -EINVAL;
1415 
1416 	/* Convert to crypto api definition? */
1417 	if (strchr(cipher_in, '(')) {
1418 		ti->error = "Bad cipher specification";
1419 		return -EINVAL;
1420 	}
1421 
1422 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
1423 	if (!cc->cipher_string)
1424 		goto bad_mem;
1425 
1426 	/*
1427 	 * Legacy dm-crypt cipher specification
1428 	 * cipher[:keycount]-mode-iv:ivopts
1429 	 */
1430 	tmp = cipher_in;
1431 	keycount = strsep(&tmp, "-");
1432 	cipher = strsep(&keycount, ":");
1433 
1434 	if (!keycount)
1435 		cc->tfms_count = 1;
1436 	else if (sscanf(keycount, "%u", &cc->tfms_count) != 1 ||
1437 		 !is_power_of_2(cc->tfms_count)) {
1438 		ti->error = "Bad cipher key count specification";
1439 		return -EINVAL;
1440 	}
1441 	cc->key_parts = cc->tfms_count;
1442 
1443 	cc->cipher = kstrdup(cipher, GFP_KERNEL);
1444 	if (!cc->cipher)
1445 		goto bad_mem;
1446 
1447 	chainmode = strsep(&tmp, "-");
1448 	ivopts = strsep(&tmp, "-");
1449 	ivmode = strsep(&ivopts, ":");
1450 
1451 	if (tmp)
1452 		DMWARN("Ignoring unexpected additional cipher options");
1453 
1454 	cc->cpu = __alloc_percpu(sizeof(*(cc->cpu)) +
1455 				 cc->tfms_count * sizeof(*(cc->cpu->tfms)),
1456 				 __alignof__(struct crypt_cpu));
1457 	if (!cc->cpu) {
1458 		ti->error = "Cannot allocate per cpu state";
1459 		goto bad_mem;
1460 	}
1461 
1462 	/*
1463 	 * For compatibility with the original dm-crypt mapping format, if
1464 	 * only the cipher name is supplied, use cbc-plain.
1465 	 */
1466 	if (!chainmode || (!strcmp(chainmode, "plain") && !ivmode)) {
1467 		chainmode = "cbc";
1468 		ivmode = "plain";
1469 	}
1470 
1471 	if (strcmp(chainmode, "ecb") && !ivmode) {
1472 		ti->error = "IV mechanism required";
1473 		return -EINVAL;
1474 	}
1475 
1476 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
1477 	if (!cipher_api)
1478 		goto bad_mem;
1479 
1480 	ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
1481 		       "%s(%s)", chainmode, cipher);
1482 	if (ret < 0) {
1483 		kfree(cipher_api);
1484 		goto bad_mem;
1485 	}
1486 
1487 	/* Allocate cipher */
1488 	for_each_possible_cpu(cpu) {
1489 		ret = crypt_alloc_tfms(cc, cpu, cipher_api);
1490 		if (ret < 0) {
1491 			ti->error = "Error allocating crypto tfm";
1492 			goto bad;
1493 		}
1494 	}
1495 
1496 	/* Initialize and set key */
1497 	ret = crypt_set_key(cc, key);
1498 	if (ret < 0) {
1499 		ti->error = "Error decoding and setting key";
1500 		goto bad;
1501 	}
1502 
1503 	/* Initialize IV */
1504 	cc->iv_size = crypto_ablkcipher_ivsize(any_tfm(cc));
1505 	if (cc->iv_size)
1506 		/* at least a 64 bit sector number should fit in our buffer */
1507 		cc->iv_size = max(cc->iv_size,
1508 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
1509 	else if (ivmode) {
1510 		DMWARN("Selected cipher does not support IVs");
1511 		ivmode = NULL;
1512 	}
1513 
1514 	/* Choose ivmode, see comments at iv code. */
1515 	if (ivmode == NULL)
1516 		cc->iv_gen_ops = NULL;
1517 	else if (strcmp(ivmode, "plain") == 0)
1518 		cc->iv_gen_ops = &crypt_iv_plain_ops;
1519 	else if (strcmp(ivmode, "plain64") == 0)
1520 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
1521 	else if (strcmp(ivmode, "essiv") == 0)
1522 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
1523 	else if (strcmp(ivmode, "benbi") == 0)
1524 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
1525 	else if (strcmp(ivmode, "null") == 0)
1526 		cc->iv_gen_ops = &crypt_iv_null_ops;
1527 	else if (strcmp(ivmode, "lmk") == 0) {
1528 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
1529 		/* Version 2 and 3 is recognised according
1530 		 * to length of provided multi-key string.
1531 		 * If present (version 3), last key is used as IV seed.
1532 		 */
1533 		if (cc->key_size % cc->key_parts)
1534 			cc->key_parts++;
1535 	} else {
1536 		ret = -EINVAL;
1537 		ti->error = "Invalid IV mode";
1538 		goto bad;
1539 	}
1540 
1541 	/* Allocate IV */
1542 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
1543 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
1544 		if (ret < 0) {
1545 			ti->error = "Error creating IV";
1546 			goto bad;
1547 		}
1548 	}
1549 
1550 	/* Initialize IV (set keys for ESSIV etc) */
1551 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
1552 		ret = cc->iv_gen_ops->init(cc);
1553 		if (ret < 0) {
1554 			ti->error = "Error initialising IV";
1555 			goto bad;
1556 		}
1557 	}
1558 
1559 	ret = 0;
1560 bad:
1561 	kfree(cipher_api);
1562 	return ret;
1563 
1564 bad_mem:
1565 	ti->error = "Cannot allocate cipher strings";
1566 	return -ENOMEM;
1567 }
1568 
1569 /*
1570  * Construct an encryption mapping:
1571  * <cipher> <key> <iv_offset> <dev_path> <start>
1572  */
1573 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1574 {
1575 	struct crypt_config *cc;
1576 	unsigned int key_size;
1577 	unsigned long long tmpll;
1578 	int ret;
1579 
1580 	if (argc != 5) {
1581 		ti->error = "Not enough arguments";
1582 		return -EINVAL;
1583 	}
1584 
1585 	key_size = strlen(argv[1]) >> 1;
1586 
1587 	cc = kzalloc(sizeof(*cc) + key_size * sizeof(u8), GFP_KERNEL);
1588 	if (!cc) {
1589 		ti->error = "Cannot allocate encryption context";
1590 		return -ENOMEM;
1591 	}
1592 	cc->key_size = key_size;
1593 
1594 	ti->private = cc;
1595 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
1596 	if (ret < 0)
1597 		goto bad;
1598 
1599 	ret = -ENOMEM;
1600 	cc->io_pool = mempool_create_slab_pool(MIN_IOS, _crypt_io_pool);
1601 	if (!cc->io_pool) {
1602 		ti->error = "Cannot allocate crypt io mempool";
1603 		goto bad;
1604 	}
1605 
1606 	cc->dmreq_start = sizeof(struct ablkcipher_request);
1607 	cc->dmreq_start += crypto_ablkcipher_reqsize(any_tfm(cc));
1608 	cc->dmreq_start = ALIGN(cc->dmreq_start, crypto_tfm_ctx_alignment());
1609 	cc->dmreq_start += crypto_ablkcipher_alignmask(any_tfm(cc)) &
1610 			   ~(crypto_tfm_ctx_alignment() - 1);
1611 
1612 	cc->req_pool = mempool_create_kmalloc_pool(MIN_IOS, cc->dmreq_start +
1613 			sizeof(struct dm_crypt_request) + cc->iv_size);
1614 	if (!cc->req_pool) {
1615 		ti->error = "Cannot allocate crypt request mempool";
1616 		goto bad;
1617 	}
1618 
1619 	cc->page_pool = mempool_create_page_pool(MIN_POOL_PAGES, 0);
1620 	if (!cc->page_pool) {
1621 		ti->error = "Cannot allocate page mempool";
1622 		goto bad;
1623 	}
1624 
1625 	cc->bs = bioset_create(MIN_IOS, 0);
1626 	if (!cc->bs) {
1627 		ti->error = "Cannot allocate crypt bioset";
1628 		goto bad;
1629 	}
1630 
1631 	ret = -EINVAL;
1632 	if (sscanf(argv[2], "%llu", &tmpll) != 1) {
1633 		ti->error = "Invalid iv_offset sector";
1634 		goto bad;
1635 	}
1636 	cc->iv_offset = tmpll;
1637 
1638 	if (dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev)) {
1639 		ti->error = "Device lookup failed";
1640 		goto bad;
1641 	}
1642 
1643 	if (sscanf(argv[4], "%llu", &tmpll) != 1) {
1644 		ti->error = "Invalid device sector";
1645 		goto bad;
1646 	}
1647 	cc->start = tmpll;
1648 
1649 	ret = -ENOMEM;
1650 	cc->io_queue = alloc_workqueue("kcryptd_io",
1651 				       WQ_NON_REENTRANT|
1652 				       WQ_MEM_RECLAIM,
1653 				       1);
1654 	if (!cc->io_queue) {
1655 		ti->error = "Couldn't create kcryptd io queue";
1656 		goto bad;
1657 	}
1658 
1659 	cc->crypt_queue = alloc_workqueue("kcryptd",
1660 					  WQ_NON_REENTRANT|
1661 					  WQ_CPU_INTENSIVE|
1662 					  WQ_MEM_RECLAIM,
1663 					  1);
1664 	if (!cc->crypt_queue) {
1665 		ti->error = "Couldn't create kcryptd queue";
1666 		goto bad;
1667 	}
1668 
1669 	ti->num_flush_requests = 1;
1670 	return 0;
1671 
1672 bad:
1673 	crypt_dtr(ti);
1674 	return ret;
1675 }
1676 
1677 static int crypt_map(struct dm_target *ti, struct bio *bio,
1678 		     union map_info *map_context)
1679 {
1680 	struct dm_crypt_io *io;
1681 	struct crypt_config *cc;
1682 
1683 	if (bio->bi_rw & REQ_FLUSH) {
1684 		cc = ti->private;
1685 		bio->bi_bdev = cc->dev->bdev;
1686 		return DM_MAPIO_REMAPPED;
1687 	}
1688 
1689 	io = crypt_io_alloc(ti, bio, dm_target_offset(ti, bio->bi_sector));
1690 
1691 	if (bio_data_dir(io->base_bio) == READ) {
1692 		if (kcryptd_io_read(io, GFP_NOWAIT))
1693 			kcryptd_queue_io(io);
1694 	} else
1695 		kcryptd_queue_crypt(io);
1696 
1697 	return DM_MAPIO_SUBMITTED;
1698 }
1699 
1700 static int crypt_status(struct dm_target *ti, status_type_t type,
1701 			char *result, unsigned int maxlen)
1702 {
1703 	struct crypt_config *cc = ti->private;
1704 	unsigned int sz = 0;
1705 
1706 	switch (type) {
1707 	case STATUSTYPE_INFO:
1708 		result[0] = '\0';
1709 		break;
1710 
1711 	case STATUSTYPE_TABLE:
1712 		DMEMIT("%s ", cc->cipher_string);
1713 
1714 		if (cc->key_size > 0) {
1715 			if ((maxlen - sz) < ((cc->key_size << 1) + 1))
1716 				return -ENOMEM;
1717 
1718 			crypt_encode_key(result + sz, cc->key, cc->key_size);
1719 			sz += cc->key_size << 1;
1720 		} else {
1721 			if (sz >= maxlen)
1722 				return -ENOMEM;
1723 			result[sz++] = '-';
1724 		}
1725 
1726 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
1727 				cc->dev->name, (unsigned long long)cc->start);
1728 		break;
1729 	}
1730 	return 0;
1731 }
1732 
1733 static void crypt_postsuspend(struct dm_target *ti)
1734 {
1735 	struct crypt_config *cc = ti->private;
1736 
1737 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
1738 }
1739 
1740 static int crypt_preresume(struct dm_target *ti)
1741 {
1742 	struct crypt_config *cc = ti->private;
1743 
1744 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
1745 		DMERR("aborting resume - crypt key is not set.");
1746 		return -EAGAIN;
1747 	}
1748 
1749 	return 0;
1750 }
1751 
1752 static void crypt_resume(struct dm_target *ti)
1753 {
1754 	struct crypt_config *cc = ti->private;
1755 
1756 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
1757 }
1758 
1759 /* Message interface
1760  *	key set <key>
1761  *	key wipe
1762  */
1763 static int crypt_message(struct dm_target *ti, unsigned argc, char **argv)
1764 {
1765 	struct crypt_config *cc = ti->private;
1766 	int ret = -EINVAL;
1767 
1768 	if (argc < 2)
1769 		goto error;
1770 
1771 	if (!strnicmp(argv[0], MESG_STR("key"))) {
1772 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
1773 			DMWARN("not suspended during key manipulation.");
1774 			return -EINVAL;
1775 		}
1776 		if (argc == 3 && !strnicmp(argv[1], MESG_STR("set"))) {
1777 			ret = crypt_set_key(cc, argv[2]);
1778 			if (ret)
1779 				return ret;
1780 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
1781 				ret = cc->iv_gen_ops->init(cc);
1782 			return ret;
1783 		}
1784 		if (argc == 2 && !strnicmp(argv[1], MESG_STR("wipe"))) {
1785 			if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
1786 				ret = cc->iv_gen_ops->wipe(cc);
1787 				if (ret)
1788 					return ret;
1789 			}
1790 			return crypt_wipe_key(cc);
1791 		}
1792 	}
1793 
1794 error:
1795 	DMWARN("unrecognised message received.");
1796 	return -EINVAL;
1797 }
1798 
1799 static int crypt_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
1800 		       struct bio_vec *biovec, int max_size)
1801 {
1802 	struct crypt_config *cc = ti->private;
1803 	struct request_queue *q = bdev_get_queue(cc->dev->bdev);
1804 
1805 	if (!q->merge_bvec_fn)
1806 		return max_size;
1807 
1808 	bvm->bi_bdev = cc->dev->bdev;
1809 	bvm->bi_sector = cc->start + dm_target_offset(ti, bvm->bi_sector);
1810 
1811 	return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
1812 }
1813 
1814 static int crypt_iterate_devices(struct dm_target *ti,
1815 				 iterate_devices_callout_fn fn, void *data)
1816 {
1817 	struct crypt_config *cc = ti->private;
1818 
1819 	return fn(ti, cc->dev, cc->start, ti->len, data);
1820 }
1821 
1822 static struct target_type crypt_target = {
1823 	.name   = "crypt",
1824 	.version = {1, 10, 0},
1825 	.module = THIS_MODULE,
1826 	.ctr    = crypt_ctr,
1827 	.dtr    = crypt_dtr,
1828 	.map    = crypt_map,
1829 	.status = crypt_status,
1830 	.postsuspend = crypt_postsuspend,
1831 	.preresume = crypt_preresume,
1832 	.resume = crypt_resume,
1833 	.message = crypt_message,
1834 	.merge  = crypt_merge,
1835 	.iterate_devices = crypt_iterate_devices,
1836 };
1837 
1838 static int __init dm_crypt_init(void)
1839 {
1840 	int r;
1841 
1842 	_crypt_io_pool = KMEM_CACHE(dm_crypt_io, 0);
1843 	if (!_crypt_io_pool)
1844 		return -ENOMEM;
1845 
1846 	r = dm_register_target(&crypt_target);
1847 	if (r < 0) {
1848 		DMERR("register failed %d", r);
1849 		kmem_cache_destroy(_crypt_io_pool);
1850 	}
1851 
1852 	return r;
1853 }
1854 
1855 static void __exit dm_crypt_exit(void)
1856 {
1857 	dm_unregister_target(&crypt_target);
1858 	kmem_cache_destroy(_crypt_io_pool);
1859 }
1860 
1861 module_init(dm_crypt_init);
1862 module_exit(dm_crypt_exit);
1863 
1864 MODULE_AUTHOR("Christophe Saout <christophe@saout.de>");
1865 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
1866 MODULE_LICENSE("GPL");
1867