xref: /openbmc/linux/drivers/md/dm-crypt.c (revision 6c33a6f4)
1 /*
2  * Copyright (C) 2003 Jana Saout <jana@saout.de>
3  * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
4  * Copyright (C) 2006-2020 Red Hat, Inc. All rights reserved.
5  * Copyright (C) 2013-2020 Milan Broz <gmazyland@gmail.com>
6  *
7  * This file is released under the GPL.
8  */
9 
10 #include <linux/completion.h>
11 #include <linux/err.h>
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/kernel.h>
15 #include <linux/key.h>
16 #include <linux/bio.h>
17 #include <linux/blkdev.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/crypto.h>
21 #include <linux/workqueue.h>
22 #include <linux/kthread.h>
23 #include <linux/backing-dev.h>
24 #include <linux/atomic.h>
25 #include <linux/scatterlist.h>
26 #include <linux/rbtree.h>
27 #include <linux/ctype.h>
28 #include <asm/page.h>
29 #include <asm/unaligned.h>
30 #include <crypto/hash.h>
31 #include <crypto/md5.h>
32 #include <crypto/algapi.h>
33 #include <crypto/skcipher.h>
34 #include <crypto/aead.h>
35 #include <crypto/authenc.h>
36 #include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
37 #include <keys/user-type.h>
38 
39 #include <linux/device-mapper.h>
40 
41 #define DM_MSG_PREFIX "crypt"
42 
43 /*
44  * context holding the current state of a multi-part conversion
45  */
46 struct convert_context {
47 	struct completion restart;
48 	struct bio *bio_in;
49 	struct bio *bio_out;
50 	struct bvec_iter iter_in;
51 	struct bvec_iter iter_out;
52 	u64 cc_sector;
53 	atomic_t cc_pending;
54 	union {
55 		struct skcipher_request *req;
56 		struct aead_request *req_aead;
57 	} r;
58 
59 };
60 
61 /*
62  * per bio private data
63  */
64 struct dm_crypt_io {
65 	struct crypt_config *cc;
66 	struct bio *base_bio;
67 	u8 *integrity_metadata;
68 	bool integrity_metadata_from_pool;
69 	struct work_struct work;
70 
71 	struct convert_context ctx;
72 
73 	atomic_t io_pending;
74 	blk_status_t error;
75 	sector_t sector;
76 
77 	struct rb_node rb_node;
78 } CRYPTO_MINALIGN_ATTR;
79 
80 struct dm_crypt_request {
81 	struct convert_context *ctx;
82 	struct scatterlist sg_in[4];
83 	struct scatterlist sg_out[4];
84 	u64 iv_sector;
85 };
86 
87 struct crypt_config;
88 
89 struct crypt_iv_operations {
90 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
91 		   const char *opts);
92 	void (*dtr)(struct crypt_config *cc);
93 	int (*init)(struct crypt_config *cc);
94 	int (*wipe)(struct crypt_config *cc);
95 	int (*generator)(struct crypt_config *cc, u8 *iv,
96 			 struct dm_crypt_request *dmreq);
97 	int (*post)(struct crypt_config *cc, u8 *iv,
98 		    struct dm_crypt_request *dmreq);
99 };
100 
101 struct iv_benbi_private {
102 	int shift;
103 };
104 
105 #define LMK_SEED_SIZE 64 /* hash + 0 */
106 struct iv_lmk_private {
107 	struct crypto_shash *hash_tfm;
108 	u8 *seed;
109 };
110 
111 #define TCW_WHITENING_SIZE 16
112 struct iv_tcw_private {
113 	struct crypto_shash *crc32_tfm;
114 	u8 *iv_seed;
115 	u8 *whitening;
116 };
117 
118 #define ELEPHANT_MAX_KEY_SIZE 32
119 struct iv_elephant_private {
120 	struct crypto_skcipher *tfm;
121 };
122 
123 /*
124  * Crypt: maps a linear range of a block device
125  * and encrypts / decrypts at the same time.
126  */
127 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
128 	     DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD };
129 
130 enum cipher_flags {
131 	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cihper */
132 	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
133 	CRYPT_ENCRYPT_PREPROCESS,	/* Must preprocess data for encryption (elephant) */
134 };
135 
136 /*
137  * The fields in here must be read only after initialization.
138  */
139 struct crypt_config {
140 	struct dm_dev *dev;
141 	sector_t start;
142 
143 	struct percpu_counter n_allocated_pages;
144 
145 	struct workqueue_struct *io_queue;
146 	struct workqueue_struct *crypt_queue;
147 
148 	spinlock_t write_thread_lock;
149 	struct task_struct *write_thread;
150 	struct rb_root write_tree;
151 
152 	char *cipher_string;
153 	char *cipher_auth;
154 	char *key_string;
155 
156 	const struct crypt_iv_operations *iv_gen_ops;
157 	union {
158 		struct iv_benbi_private benbi;
159 		struct iv_lmk_private lmk;
160 		struct iv_tcw_private tcw;
161 		struct iv_elephant_private elephant;
162 	} iv_gen_private;
163 	u64 iv_offset;
164 	unsigned int iv_size;
165 	unsigned short int sector_size;
166 	unsigned char sector_shift;
167 
168 	union {
169 		struct crypto_skcipher **tfms;
170 		struct crypto_aead **tfms_aead;
171 	} cipher_tfm;
172 	unsigned tfms_count;
173 	unsigned long cipher_flags;
174 
175 	/*
176 	 * Layout of each crypto request:
177 	 *
178 	 *   struct skcipher_request
179 	 *      context
180 	 *      padding
181 	 *   struct dm_crypt_request
182 	 *      padding
183 	 *   IV
184 	 *
185 	 * The padding is added so that dm_crypt_request and the IV are
186 	 * correctly aligned.
187 	 */
188 	unsigned int dmreq_start;
189 
190 	unsigned int per_bio_data_size;
191 
192 	unsigned long flags;
193 	unsigned int key_size;
194 	unsigned int key_parts;      /* independent parts in key buffer */
195 	unsigned int key_extra_size; /* additional keys length */
196 	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
197 
198 	unsigned int integrity_tag_size;
199 	unsigned int integrity_iv_size;
200 	unsigned int on_disk_tag_size;
201 
202 	/*
203 	 * pool for per bio private data, crypto requests,
204 	 * encryption requeusts/buffer pages and integrity tags
205 	 */
206 	unsigned tag_pool_max_sectors;
207 	mempool_t tag_pool;
208 	mempool_t req_pool;
209 	mempool_t page_pool;
210 
211 	struct bio_set bs;
212 	struct mutex bio_alloc_lock;
213 
214 	u8 *authenc_key; /* space for keys in authenc() format (if used) */
215 	u8 key[0];
216 };
217 
218 #define MIN_IOS		64
219 #define MAX_TAG_SIZE	480
220 #define POOL_ENTRY_SIZE	512
221 
222 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
223 static unsigned dm_crypt_clients_n = 0;
224 static volatile unsigned long dm_crypt_pages_per_client;
225 #define DM_CRYPT_MEMORY_PERCENT			2
226 #define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_PAGES * 16)
227 
228 static void clone_init(struct dm_crypt_io *, struct bio *);
229 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
230 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
231 					     struct scatterlist *sg);
232 
233 /*
234  * Use this to access cipher attributes that are independent of the key.
235  */
236 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
237 {
238 	return cc->cipher_tfm.tfms[0];
239 }
240 
241 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
242 {
243 	return cc->cipher_tfm.tfms_aead[0];
244 }
245 
246 /*
247  * Different IV generation algorithms:
248  *
249  * plain: the initial vector is the 32-bit little-endian version of the sector
250  *        number, padded with zeros if necessary.
251  *
252  * plain64: the initial vector is the 64-bit little-endian version of the sector
253  *        number, padded with zeros if necessary.
254  *
255  * plain64be: the initial vector is the 64-bit big-endian version of the sector
256  *        number, padded with zeros if necessary.
257  *
258  * essiv: "encrypted sector|salt initial vector", the sector number is
259  *        encrypted with the bulk cipher using a salt as key. The salt
260  *        should be derived from the bulk cipher's key via hashing.
261  *
262  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
263  *        (needed for LRW-32-AES and possible other narrow block modes)
264  *
265  * null: the initial vector is always zero.  Provides compatibility with
266  *       obsolete loop_fish2 devices.  Do not use for new devices.
267  *
268  * lmk:  Compatible implementation of the block chaining mode used
269  *       by the Loop-AES block device encryption system
270  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
271  *       It operates on full 512 byte sectors and uses CBC
272  *       with an IV derived from the sector number, the data and
273  *       optionally extra IV seed.
274  *       This means that after decryption the first block
275  *       of sector must be tweaked according to decrypted data.
276  *       Loop-AES can use three encryption schemes:
277  *         version 1: is plain aes-cbc mode
278  *         version 2: uses 64 multikey scheme with lmk IV generator
279  *         version 3: the same as version 2 with additional IV seed
280  *                   (it uses 65 keys, last key is used as IV seed)
281  *
282  * tcw:  Compatible implementation of the block chaining mode used
283  *       by the TrueCrypt device encryption system (prior to version 4.1).
284  *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
285  *       It operates on full 512 byte sectors and uses CBC
286  *       with an IV derived from initial key and the sector number.
287  *       In addition, whitening value is applied on every sector, whitening
288  *       is calculated from initial key, sector number and mixed using CRC32.
289  *       Note that this encryption scheme is vulnerable to watermarking attacks
290  *       and should be used for old compatible containers access only.
291  *
292  * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
293  *        The IV is encrypted little-endian byte-offset (with the same key
294  *        and cipher as the volume).
295  *
296  * elephant: The extended version of eboiv with additional Elephant diffuser
297  *           used with Bitlocker CBC mode.
298  *           This mode was used in older Windows systems
299  *           http://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
300  */
301 
302 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
303 			      struct dm_crypt_request *dmreq)
304 {
305 	memset(iv, 0, cc->iv_size);
306 	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
307 
308 	return 0;
309 }
310 
311 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
312 				struct dm_crypt_request *dmreq)
313 {
314 	memset(iv, 0, cc->iv_size);
315 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
316 
317 	return 0;
318 }
319 
320 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
321 				  struct dm_crypt_request *dmreq)
322 {
323 	memset(iv, 0, cc->iv_size);
324 	/* iv_size is at least of size u64; usually it is 16 bytes */
325 	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
326 
327 	return 0;
328 }
329 
330 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
331 			      struct dm_crypt_request *dmreq)
332 {
333 	/*
334 	 * ESSIV encryption of the IV is now handled by the crypto API,
335 	 * so just pass the plain sector number here.
336 	 */
337 	memset(iv, 0, cc->iv_size);
338 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
339 
340 	return 0;
341 }
342 
343 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
344 			      const char *opts)
345 {
346 	unsigned bs;
347 	int log;
348 
349 	if (test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags))
350 		bs = crypto_aead_blocksize(any_tfm_aead(cc));
351 	else
352 		bs = crypto_skcipher_blocksize(any_tfm(cc));
353 	log = ilog2(bs);
354 
355 	/* we need to calculate how far we must shift the sector count
356 	 * to get the cipher block count, we use this shift in _gen */
357 
358 	if (1 << log != bs) {
359 		ti->error = "cypher blocksize is not a power of 2";
360 		return -EINVAL;
361 	}
362 
363 	if (log > 9) {
364 		ti->error = "cypher blocksize is > 512";
365 		return -EINVAL;
366 	}
367 
368 	cc->iv_gen_private.benbi.shift = 9 - log;
369 
370 	return 0;
371 }
372 
373 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
374 {
375 }
376 
377 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
378 			      struct dm_crypt_request *dmreq)
379 {
380 	__be64 val;
381 
382 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
383 
384 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
385 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
386 
387 	return 0;
388 }
389 
390 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
391 			     struct dm_crypt_request *dmreq)
392 {
393 	memset(iv, 0, cc->iv_size);
394 
395 	return 0;
396 }
397 
398 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
399 {
400 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
401 
402 	if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
403 		crypto_free_shash(lmk->hash_tfm);
404 	lmk->hash_tfm = NULL;
405 
406 	kzfree(lmk->seed);
407 	lmk->seed = NULL;
408 }
409 
410 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
411 			    const char *opts)
412 {
413 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
414 
415 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
416 		ti->error = "Unsupported sector size for LMK";
417 		return -EINVAL;
418 	}
419 
420 	lmk->hash_tfm = crypto_alloc_shash("md5", 0, 0);
421 	if (IS_ERR(lmk->hash_tfm)) {
422 		ti->error = "Error initializing LMK hash";
423 		return PTR_ERR(lmk->hash_tfm);
424 	}
425 
426 	/* No seed in LMK version 2 */
427 	if (cc->key_parts == cc->tfms_count) {
428 		lmk->seed = NULL;
429 		return 0;
430 	}
431 
432 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
433 	if (!lmk->seed) {
434 		crypt_iv_lmk_dtr(cc);
435 		ti->error = "Error kmallocing seed storage in LMK";
436 		return -ENOMEM;
437 	}
438 
439 	return 0;
440 }
441 
442 static int crypt_iv_lmk_init(struct crypt_config *cc)
443 {
444 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
445 	int subkey_size = cc->key_size / cc->key_parts;
446 
447 	/* LMK seed is on the position of LMK_KEYS + 1 key */
448 	if (lmk->seed)
449 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
450 		       crypto_shash_digestsize(lmk->hash_tfm));
451 
452 	return 0;
453 }
454 
455 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
456 {
457 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
458 
459 	if (lmk->seed)
460 		memset(lmk->seed, 0, LMK_SEED_SIZE);
461 
462 	return 0;
463 }
464 
465 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
466 			    struct dm_crypt_request *dmreq,
467 			    u8 *data)
468 {
469 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
470 	SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
471 	struct md5_state md5state;
472 	__le32 buf[4];
473 	int i, r;
474 
475 	desc->tfm = lmk->hash_tfm;
476 
477 	r = crypto_shash_init(desc);
478 	if (r)
479 		return r;
480 
481 	if (lmk->seed) {
482 		r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
483 		if (r)
484 			return r;
485 	}
486 
487 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
488 	r = crypto_shash_update(desc, data + 16, 16 * 31);
489 	if (r)
490 		return r;
491 
492 	/* Sector is cropped to 56 bits here */
493 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
494 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
495 	buf[2] = cpu_to_le32(4024);
496 	buf[3] = 0;
497 	r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
498 	if (r)
499 		return r;
500 
501 	/* No MD5 padding here */
502 	r = crypto_shash_export(desc, &md5state);
503 	if (r)
504 		return r;
505 
506 	for (i = 0; i < MD5_HASH_WORDS; i++)
507 		__cpu_to_le32s(&md5state.hash[i]);
508 	memcpy(iv, &md5state.hash, cc->iv_size);
509 
510 	return 0;
511 }
512 
513 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
514 			    struct dm_crypt_request *dmreq)
515 {
516 	struct scatterlist *sg;
517 	u8 *src;
518 	int r = 0;
519 
520 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
521 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
522 		src = kmap_atomic(sg_page(sg));
523 		r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
524 		kunmap_atomic(src);
525 	} else
526 		memset(iv, 0, cc->iv_size);
527 
528 	return r;
529 }
530 
531 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
532 			     struct dm_crypt_request *dmreq)
533 {
534 	struct scatterlist *sg;
535 	u8 *dst;
536 	int r;
537 
538 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
539 		return 0;
540 
541 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
542 	dst = kmap_atomic(sg_page(sg));
543 	r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
544 
545 	/* Tweak the first block of plaintext sector */
546 	if (!r)
547 		crypto_xor(dst + sg->offset, iv, cc->iv_size);
548 
549 	kunmap_atomic(dst);
550 	return r;
551 }
552 
553 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
554 {
555 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
556 
557 	kzfree(tcw->iv_seed);
558 	tcw->iv_seed = NULL;
559 	kzfree(tcw->whitening);
560 	tcw->whitening = NULL;
561 
562 	if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
563 		crypto_free_shash(tcw->crc32_tfm);
564 	tcw->crc32_tfm = NULL;
565 }
566 
567 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
568 			    const char *opts)
569 {
570 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
571 
572 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
573 		ti->error = "Unsupported sector size for TCW";
574 		return -EINVAL;
575 	}
576 
577 	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
578 		ti->error = "Wrong key size for TCW";
579 		return -EINVAL;
580 	}
581 
582 	tcw->crc32_tfm = crypto_alloc_shash("crc32", 0, 0);
583 	if (IS_ERR(tcw->crc32_tfm)) {
584 		ti->error = "Error initializing CRC32 in TCW";
585 		return PTR_ERR(tcw->crc32_tfm);
586 	}
587 
588 	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
589 	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
590 	if (!tcw->iv_seed || !tcw->whitening) {
591 		crypt_iv_tcw_dtr(cc);
592 		ti->error = "Error allocating seed storage in TCW";
593 		return -ENOMEM;
594 	}
595 
596 	return 0;
597 }
598 
599 static int crypt_iv_tcw_init(struct crypt_config *cc)
600 {
601 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
602 	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
603 
604 	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
605 	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
606 	       TCW_WHITENING_SIZE);
607 
608 	return 0;
609 }
610 
611 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
612 {
613 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
614 
615 	memset(tcw->iv_seed, 0, cc->iv_size);
616 	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
617 
618 	return 0;
619 }
620 
621 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
622 				  struct dm_crypt_request *dmreq,
623 				  u8 *data)
624 {
625 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
626 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
627 	u8 buf[TCW_WHITENING_SIZE];
628 	SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
629 	int i, r;
630 
631 	/* xor whitening with sector number */
632 	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
633 	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
634 
635 	/* calculate crc32 for every 32bit part and xor it */
636 	desc->tfm = tcw->crc32_tfm;
637 	for (i = 0; i < 4; i++) {
638 		r = crypto_shash_init(desc);
639 		if (r)
640 			goto out;
641 		r = crypto_shash_update(desc, &buf[i * 4], 4);
642 		if (r)
643 			goto out;
644 		r = crypto_shash_final(desc, &buf[i * 4]);
645 		if (r)
646 			goto out;
647 	}
648 	crypto_xor(&buf[0], &buf[12], 4);
649 	crypto_xor(&buf[4], &buf[8], 4);
650 
651 	/* apply whitening (8 bytes) to whole sector */
652 	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
653 		crypto_xor(data + i * 8, buf, 8);
654 out:
655 	memzero_explicit(buf, sizeof(buf));
656 	return r;
657 }
658 
659 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
660 			    struct dm_crypt_request *dmreq)
661 {
662 	struct scatterlist *sg;
663 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
664 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
665 	u8 *src;
666 	int r = 0;
667 
668 	/* Remove whitening from ciphertext */
669 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
670 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
671 		src = kmap_atomic(sg_page(sg));
672 		r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
673 		kunmap_atomic(src);
674 	}
675 
676 	/* Calculate IV */
677 	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
678 	if (cc->iv_size > 8)
679 		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
680 			       cc->iv_size - 8);
681 
682 	return r;
683 }
684 
685 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
686 			     struct dm_crypt_request *dmreq)
687 {
688 	struct scatterlist *sg;
689 	u8 *dst;
690 	int r;
691 
692 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
693 		return 0;
694 
695 	/* Apply whitening on ciphertext */
696 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
697 	dst = kmap_atomic(sg_page(sg));
698 	r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
699 	kunmap_atomic(dst);
700 
701 	return r;
702 }
703 
704 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
705 				struct dm_crypt_request *dmreq)
706 {
707 	/* Used only for writes, there must be an additional space to store IV */
708 	get_random_bytes(iv, cc->iv_size);
709 	return 0;
710 }
711 
712 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
713 			    const char *opts)
714 {
715 	if (test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags)) {
716 		ti->error = "AEAD transforms not supported for EBOIV";
717 		return -EINVAL;
718 	}
719 
720 	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
721 		ti->error = "Block size of EBOIV cipher does "
722 			    "not match IV size of block cipher";
723 		return -EINVAL;
724 	}
725 
726 	return 0;
727 }
728 
729 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
730 			    struct dm_crypt_request *dmreq)
731 {
732 	u8 buf[MAX_CIPHER_BLOCKSIZE] __aligned(__alignof__(__le64));
733 	struct skcipher_request *req;
734 	struct scatterlist src, dst;
735 	struct crypto_wait wait;
736 	int err;
737 
738 	req = skcipher_request_alloc(any_tfm(cc), GFP_NOIO);
739 	if (!req)
740 		return -ENOMEM;
741 
742 	memset(buf, 0, cc->iv_size);
743 	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
744 
745 	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
746 	sg_init_one(&dst, iv, cc->iv_size);
747 	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
748 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
749 	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
750 	skcipher_request_free(req);
751 
752 	return err;
753 }
754 
755 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
756 {
757 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
758 
759 	crypto_free_skcipher(elephant->tfm);
760 	elephant->tfm = NULL;
761 }
762 
763 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
764 			    const char *opts)
765 {
766 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
767 	int r;
768 
769 	elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0, 0);
770 	if (IS_ERR(elephant->tfm)) {
771 		r = PTR_ERR(elephant->tfm);
772 		elephant->tfm = NULL;
773 		return r;
774 	}
775 
776 	r = crypt_iv_eboiv_ctr(cc, ti, NULL);
777 	if (r)
778 		crypt_iv_elephant_dtr(cc);
779 	return r;
780 }
781 
782 static void diffuser_disk_to_cpu(u32 *d, size_t n)
783 {
784 #ifndef __LITTLE_ENDIAN
785 	int i;
786 
787 	for (i = 0; i < n; i++)
788 		d[i] = le32_to_cpu((__le32)d[i]);
789 #endif
790 }
791 
792 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
793 {
794 #ifndef __LITTLE_ENDIAN
795 	int i;
796 
797 	for (i = 0; i < n; i++)
798 		d[i] = cpu_to_le32((u32)d[i]);
799 #endif
800 }
801 
802 static void diffuser_a_decrypt(u32 *d, size_t n)
803 {
804 	int i, i1, i2, i3;
805 
806 	for (i = 0; i < 5; i++) {
807 		i1 = 0;
808 		i2 = n - 2;
809 		i3 = n - 5;
810 
811 		while (i1 < (n - 1)) {
812 			d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
813 			i1++; i2++; i3++;
814 
815 			if (i3 >= n)
816 				i3 -= n;
817 
818 			d[i1] += d[i2] ^ d[i3];
819 			i1++; i2++; i3++;
820 
821 			if (i2 >= n)
822 				i2 -= n;
823 
824 			d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
825 			i1++; i2++; i3++;
826 
827 			d[i1] += d[i2] ^ d[i3];
828 			i1++; i2++; i3++;
829 		}
830 	}
831 }
832 
833 static void diffuser_a_encrypt(u32 *d, size_t n)
834 {
835 	int i, i1, i2, i3;
836 
837 	for (i = 0; i < 5; i++) {
838 		i1 = n - 1;
839 		i2 = n - 2 - 1;
840 		i3 = n - 5 - 1;
841 
842 		while (i1 > 0) {
843 			d[i1] -= d[i2] ^ d[i3];
844 			i1--; i2--; i3--;
845 
846 			d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
847 			i1--; i2--; i3--;
848 
849 			if (i2 < 0)
850 				i2 += n;
851 
852 			d[i1] -= d[i2] ^ d[i3];
853 			i1--; i2--; i3--;
854 
855 			if (i3 < 0)
856 				i3 += n;
857 
858 			d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
859 			i1--; i2--; i3--;
860 		}
861 	}
862 }
863 
864 static void diffuser_b_decrypt(u32 *d, size_t n)
865 {
866 	int i, i1, i2, i3;
867 
868 	for (i = 0; i < 3; i++) {
869 		i1 = 0;
870 		i2 = 2;
871 		i3 = 5;
872 
873 		while (i1 < (n - 1)) {
874 			d[i1] += d[i2] ^ d[i3];
875 			i1++; i2++; i3++;
876 
877 			d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
878 			i1++; i2++; i3++;
879 
880 			if (i2 >= n)
881 				i2 -= n;
882 
883 			d[i1] += d[i2] ^ d[i3];
884 			i1++; i2++; i3++;
885 
886 			if (i3 >= n)
887 				i3 -= n;
888 
889 			d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
890 			i1++; i2++; i3++;
891 		}
892 	}
893 }
894 
895 static void diffuser_b_encrypt(u32 *d, size_t n)
896 {
897 	int i, i1, i2, i3;
898 
899 	for (i = 0; i < 3; i++) {
900 		i1 = n - 1;
901 		i2 = 2 - 1;
902 		i3 = 5 - 1;
903 
904 		while (i1 > 0) {
905 			d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
906 			i1--; i2--; i3--;
907 
908 			if (i3 < 0)
909 				i3 += n;
910 
911 			d[i1] -= d[i2] ^ d[i3];
912 			i1--; i2--; i3--;
913 
914 			if (i2 < 0)
915 				i2 += n;
916 
917 			d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
918 			i1--; i2--; i3--;
919 
920 			d[i1] -= d[i2] ^ d[i3];
921 			i1--; i2--; i3--;
922 		}
923 	}
924 }
925 
926 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
927 {
928 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
929 	u8 *es, *ks, *data, *data2, *data_offset;
930 	struct skcipher_request *req;
931 	struct scatterlist *sg, *sg2, src, dst;
932 	struct crypto_wait wait;
933 	int i, r;
934 
935 	req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
936 	es = kzalloc(16, GFP_NOIO); /* Key for AES */
937 	ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
938 
939 	if (!req || !es || !ks) {
940 		r = -ENOMEM;
941 		goto out;
942 	}
943 
944 	*(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
945 
946 	/* E(Ks, e(s)) */
947 	sg_init_one(&src, es, 16);
948 	sg_init_one(&dst, ks, 16);
949 	skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
950 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
951 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
952 	if (r)
953 		goto out;
954 
955 	/* E(Ks, e'(s)) */
956 	es[15] = 0x80;
957 	sg_init_one(&dst, &ks[16], 16);
958 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
959 	if (r)
960 		goto out;
961 
962 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
963 	data = kmap_atomic(sg_page(sg));
964 	data_offset = data + sg->offset;
965 
966 	/* Cannot modify original bio, copy to sg_out and apply Elephant to it */
967 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
968 		sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
969 		data2 = kmap_atomic(sg_page(sg2));
970 		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
971 		kunmap_atomic(data2);
972 	}
973 
974 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
975 		diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
976 		diffuser_b_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
977 		diffuser_a_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
978 		diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
979 	}
980 
981 	for (i = 0; i < (cc->sector_size / 32); i++)
982 		crypto_xor(data_offset + i * 32, ks, 32);
983 
984 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
985 		diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
986 		diffuser_a_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
987 		diffuser_b_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
988 		diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
989 	}
990 
991 	kunmap_atomic(data);
992 out:
993 	kzfree(ks);
994 	kzfree(es);
995 	skcipher_request_free(req);
996 	return r;
997 }
998 
999 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1000 			    struct dm_crypt_request *dmreq)
1001 {
1002 	int r;
1003 
1004 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1005 		r = crypt_iv_elephant(cc, dmreq);
1006 		if (r)
1007 			return r;
1008 	}
1009 
1010 	return crypt_iv_eboiv_gen(cc, iv, dmreq);
1011 }
1012 
1013 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1014 				  struct dm_crypt_request *dmreq)
1015 {
1016 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1017 		return crypt_iv_elephant(cc, dmreq);
1018 
1019 	return 0;
1020 }
1021 
1022 static int crypt_iv_elephant_init(struct crypt_config *cc)
1023 {
1024 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1025 	int key_offset = cc->key_size - cc->key_extra_size;
1026 
1027 	return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1028 }
1029 
1030 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1031 {
1032 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1033 	u8 key[ELEPHANT_MAX_KEY_SIZE];
1034 
1035 	memset(key, 0, cc->key_extra_size);
1036 	return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1037 }
1038 
1039 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1040 	.generator = crypt_iv_plain_gen
1041 };
1042 
1043 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1044 	.generator = crypt_iv_plain64_gen
1045 };
1046 
1047 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1048 	.generator = crypt_iv_plain64be_gen
1049 };
1050 
1051 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1052 	.generator = crypt_iv_essiv_gen
1053 };
1054 
1055 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1056 	.ctr	   = crypt_iv_benbi_ctr,
1057 	.dtr	   = crypt_iv_benbi_dtr,
1058 	.generator = crypt_iv_benbi_gen
1059 };
1060 
1061 static const struct crypt_iv_operations crypt_iv_null_ops = {
1062 	.generator = crypt_iv_null_gen
1063 };
1064 
1065 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1066 	.ctr	   = crypt_iv_lmk_ctr,
1067 	.dtr	   = crypt_iv_lmk_dtr,
1068 	.init	   = crypt_iv_lmk_init,
1069 	.wipe	   = crypt_iv_lmk_wipe,
1070 	.generator = crypt_iv_lmk_gen,
1071 	.post	   = crypt_iv_lmk_post
1072 };
1073 
1074 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1075 	.ctr	   = crypt_iv_tcw_ctr,
1076 	.dtr	   = crypt_iv_tcw_dtr,
1077 	.init	   = crypt_iv_tcw_init,
1078 	.wipe	   = crypt_iv_tcw_wipe,
1079 	.generator = crypt_iv_tcw_gen,
1080 	.post	   = crypt_iv_tcw_post
1081 };
1082 
1083 static struct crypt_iv_operations crypt_iv_random_ops = {
1084 	.generator = crypt_iv_random_gen
1085 };
1086 
1087 static struct crypt_iv_operations crypt_iv_eboiv_ops = {
1088 	.ctr	   = crypt_iv_eboiv_ctr,
1089 	.generator = crypt_iv_eboiv_gen
1090 };
1091 
1092 static struct crypt_iv_operations crypt_iv_elephant_ops = {
1093 	.ctr	   = crypt_iv_elephant_ctr,
1094 	.dtr	   = crypt_iv_elephant_dtr,
1095 	.init	   = crypt_iv_elephant_init,
1096 	.wipe	   = crypt_iv_elephant_wipe,
1097 	.generator = crypt_iv_elephant_gen,
1098 	.post	   = crypt_iv_elephant_post
1099 };
1100 
1101 /*
1102  * Integrity extensions
1103  */
1104 static bool crypt_integrity_aead(struct crypt_config *cc)
1105 {
1106 	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1107 }
1108 
1109 static bool crypt_integrity_hmac(struct crypt_config *cc)
1110 {
1111 	return crypt_integrity_aead(cc) && cc->key_mac_size;
1112 }
1113 
1114 /* Get sg containing data */
1115 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1116 					     struct scatterlist *sg)
1117 {
1118 	if (unlikely(crypt_integrity_aead(cc)))
1119 		return &sg[2];
1120 
1121 	return sg;
1122 }
1123 
1124 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1125 {
1126 	struct bio_integrity_payload *bip;
1127 	unsigned int tag_len;
1128 	int ret;
1129 
1130 	if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1131 		return 0;
1132 
1133 	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1134 	if (IS_ERR(bip))
1135 		return PTR_ERR(bip);
1136 
1137 	tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1138 
1139 	bip->bip_iter.bi_size = tag_len;
1140 	bip->bip_iter.bi_sector = io->cc->start + io->sector;
1141 
1142 	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1143 				     tag_len, offset_in_page(io->integrity_metadata));
1144 	if (unlikely(ret != tag_len))
1145 		return -ENOMEM;
1146 
1147 	return 0;
1148 }
1149 
1150 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1151 {
1152 #ifdef CONFIG_BLK_DEV_INTEGRITY
1153 	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1154 	struct mapped_device *md = dm_table_get_md(ti->table);
1155 
1156 	/* From now we require underlying device with our integrity profile */
1157 	if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1158 		ti->error = "Integrity profile not supported.";
1159 		return -EINVAL;
1160 	}
1161 
1162 	if (bi->tag_size != cc->on_disk_tag_size ||
1163 	    bi->tuple_size != cc->on_disk_tag_size) {
1164 		ti->error = "Integrity profile tag size mismatch.";
1165 		return -EINVAL;
1166 	}
1167 	if (1 << bi->interval_exp != cc->sector_size) {
1168 		ti->error = "Integrity profile sector size mismatch.";
1169 		return -EINVAL;
1170 	}
1171 
1172 	if (crypt_integrity_aead(cc)) {
1173 		cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1174 		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1175 		       cc->integrity_tag_size, cc->integrity_iv_size);
1176 
1177 		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1178 			ti->error = "Integrity AEAD auth tag size is not supported.";
1179 			return -EINVAL;
1180 		}
1181 	} else if (cc->integrity_iv_size)
1182 		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1183 		       cc->integrity_iv_size);
1184 
1185 	if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1186 		ti->error = "Not enough space for integrity tag in the profile.";
1187 		return -EINVAL;
1188 	}
1189 
1190 	return 0;
1191 #else
1192 	ti->error = "Integrity profile not supported.";
1193 	return -EINVAL;
1194 #endif
1195 }
1196 
1197 static void crypt_convert_init(struct crypt_config *cc,
1198 			       struct convert_context *ctx,
1199 			       struct bio *bio_out, struct bio *bio_in,
1200 			       sector_t sector)
1201 {
1202 	ctx->bio_in = bio_in;
1203 	ctx->bio_out = bio_out;
1204 	if (bio_in)
1205 		ctx->iter_in = bio_in->bi_iter;
1206 	if (bio_out)
1207 		ctx->iter_out = bio_out->bi_iter;
1208 	ctx->cc_sector = sector + cc->iv_offset;
1209 	init_completion(&ctx->restart);
1210 }
1211 
1212 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1213 					     void *req)
1214 {
1215 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1216 }
1217 
1218 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1219 {
1220 	return (void *)((char *)dmreq - cc->dmreq_start);
1221 }
1222 
1223 static u8 *iv_of_dmreq(struct crypt_config *cc,
1224 		       struct dm_crypt_request *dmreq)
1225 {
1226 	if (crypt_integrity_aead(cc))
1227 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1228 			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1229 	else
1230 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1231 			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1232 }
1233 
1234 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1235 		       struct dm_crypt_request *dmreq)
1236 {
1237 	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1238 }
1239 
1240 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1241 		       struct dm_crypt_request *dmreq)
1242 {
1243 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1244 	return (__le64 *) ptr;
1245 }
1246 
1247 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1248 		       struct dm_crypt_request *dmreq)
1249 {
1250 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1251 		  cc->iv_size + sizeof(uint64_t);
1252 	return (unsigned int*)ptr;
1253 }
1254 
1255 static void *tag_from_dmreq(struct crypt_config *cc,
1256 				struct dm_crypt_request *dmreq)
1257 {
1258 	struct convert_context *ctx = dmreq->ctx;
1259 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1260 
1261 	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1262 		cc->on_disk_tag_size];
1263 }
1264 
1265 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1266 			       struct dm_crypt_request *dmreq)
1267 {
1268 	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1269 }
1270 
1271 static int crypt_convert_block_aead(struct crypt_config *cc,
1272 				     struct convert_context *ctx,
1273 				     struct aead_request *req,
1274 				     unsigned int tag_offset)
1275 {
1276 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1277 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1278 	struct dm_crypt_request *dmreq;
1279 	u8 *iv, *org_iv, *tag_iv, *tag;
1280 	__le64 *sector;
1281 	int r = 0;
1282 
1283 	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1284 
1285 	/* Reject unexpected unaligned bio. */
1286 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1287 		return -EIO;
1288 
1289 	dmreq = dmreq_of_req(cc, req);
1290 	dmreq->iv_sector = ctx->cc_sector;
1291 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1292 		dmreq->iv_sector >>= cc->sector_shift;
1293 	dmreq->ctx = ctx;
1294 
1295 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1296 
1297 	sector = org_sector_of_dmreq(cc, dmreq);
1298 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1299 
1300 	iv = iv_of_dmreq(cc, dmreq);
1301 	org_iv = org_iv_of_dmreq(cc, dmreq);
1302 	tag = tag_from_dmreq(cc, dmreq);
1303 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1304 
1305 	/* AEAD request:
1306 	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1307 	 *  | (authenticated) | (auth+encryption) |              |
1308 	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1309 	 */
1310 	sg_init_table(dmreq->sg_in, 4);
1311 	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1312 	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1313 	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1314 	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1315 
1316 	sg_init_table(dmreq->sg_out, 4);
1317 	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1318 	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1319 	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1320 	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1321 
1322 	if (cc->iv_gen_ops) {
1323 		/* For READs use IV stored in integrity metadata */
1324 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1325 			memcpy(org_iv, tag_iv, cc->iv_size);
1326 		} else {
1327 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1328 			if (r < 0)
1329 				return r;
1330 			/* Store generated IV in integrity metadata */
1331 			if (cc->integrity_iv_size)
1332 				memcpy(tag_iv, org_iv, cc->iv_size);
1333 		}
1334 		/* Working copy of IV, to be modified in crypto API */
1335 		memcpy(iv, org_iv, cc->iv_size);
1336 	}
1337 
1338 	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1339 	if (bio_data_dir(ctx->bio_in) == WRITE) {
1340 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1341 				       cc->sector_size, iv);
1342 		r = crypto_aead_encrypt(req);
1343 		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1344 			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1345 			       cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1346 	} else {
1347 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1348 				       cc->sector_size + cc->integrity_tag_size, iv);
1349 		r = crypto_aead_decrypt(req);
1350 	}
1351 
1352 	if (r == -EBADMSG) {
1353 		char b[BDEVNAME_SIZE];
1354 		DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", bio_devname(ctx->bio_in, b),
1355 			    (unsigned long long)le64_to_cpu(*sector));
1356 	}
1357 
1358 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1359 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1360 
1361 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1362 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1363 
1364 	return r;
1365 }
1366 
1367 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1368 					struct convert_context *ctx,
1369 					struct skcipher_request *req,
1370 					unsigned int tag_offset)
1371 {
1372 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1373 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1374 	struct scatterlist *sg_in, *sg_out;
1375 	struct dm_crypt_request *dmreq;
1376 	u8 *iv, *org_iv, *tag_iv;
1377 	__le64 *sector;
1378 	int r = 0;
1379 
1380 	/* Reject unexpected unaligned bio. */
1381 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1382 		return -EIO;
1383 
1384 	dmreq = dmreq_of_req(cc, req);
1385 	dmreq->iv_sector = ctx->cc_sector;
1386 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1387 		dmreq->iv_sector >>= cc->sector_shift;
1388 	dmreq->ctx = ctx;
1389 
1390 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1391 
1392 	iv = iv_of_dmreq(cc, dmreq);
1393 	org_iv = org_iv_of_dmreq(cc, dmreq);
1394 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1395 
1396 	sector = org_sector_of_dmreq(cc, dmreq);
1397 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1398 
1399 	/* For skcipher we use only the first sg item */
1400 	sg_in  = &dmreq->sg_in[0];
1401 	sg_out = &dmreq->sg_out[0];
1402 
1403 	sg_init_table(sg_in, 1);
1404 	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1405 
1406 	sg_init_table(sg_out, 1);
1407 	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1408 
1409 	if (cc->iv_gen_ops) {
1410 		/* For READs use IV stored in integrity metadata */
1411 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1412 			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1413 		} else {
1414 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1415 			if (r < 0)
1416 				return r;
1417 			/* Data can be already preprocessed in generator */
1418 			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1419 				sg_in = sg_out;
1420 			/* Store generated IV in integrity metadata */
1421 			if (cc->integrity_iv_size)
1422 				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1423 		}
1424 		/* Working copy of IV, to be modified in crypto API */
1425 		memcpy(iv, org_iv, cc->iv_size);
1426 	}
1427 
1428 	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1429 
1430 	if (bio_data_dir(ctx->bio_in) == WRITE)
1431 		r = crypto_skcipher_encrypt(req);
1432 	else
1433 		r = crypto_skcipher_decrypt(req);
1434 
1435 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1436 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1437 
1438 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1439 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1440 
1441 	return r;
1442 }
1443 
1444 static void kcryptd_async_done(struct crypto_async_request *async_req,
1445 			       int error);
1446 
1447 static void crypt_alloc_req_skcipher(struct crypt_config *cc,
1448 				     struct convert_context *ctx)
1449 {
1450 	unsigned key_index = ctx->cc_sector & (cc->tfms_count - 1);
1451 
1452 	if (!ctx->r.req)
1453 		ctx->r.req = mempool_alloc(&cc->req_pool, GFP_NOIO);
1454 
1455 	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1456 
1457 	/*
1458 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1459 	 * requests if driver request queue is full.
1460 	 */
1461 	skcipher_request_set_callback(ctx->r.req,
1462 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1463 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1464 }
1465 
1466 static void crypt_alloc_req_aead(struct crypt_config *cc,
1467 				 struct convert_context *ctx)
1468 {
1469 	if (!ctx->r.req_aead)
1470 		ctx->r.req_aead = mempool_alloc(&cc->req_pool, GFP_NOIO);
1471 
1472 	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1473 
1474 	/*
1475 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1476 	 * requests if driver request queue is full.
1477 	 */
1478 	aead_request_set_callback(ctx->r.req_aead,
1479 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1480 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1481 }
1482 
1483 static void crypt_alloc_req(struct crypt_config *cc,
1484 			    struct convert_context *ctx)
1485 {
1486 	if (crypt_integrity_aead(cc))
1487 		crypt_alloc_req_aead(cc, ctx);
1488 	else
1489 		crypt_alloc_req_skcipher(cc, ctx);
1490 }
1491 
1492 static void crypt_free_req_skcipher(struct crypt_config *cc,
1493 				    struct skcipher_request *req, struct bio *base_bio)
1494 {
1495 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1496 
1497 	if ((struct skcipher_request *)(io + 1) != req)
1498 		mempool_free(req, &cc->req_pool);
1499 }
1500 
1501 static void crypt_free_req_aead(struct crypt_config *cc,
1502 				struct aead_request *req, struct bio *base_bio)
1503 {
1504 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1505 
1506 	if ((struct aead_request *)(io + 1) != req)
1507 		mempool_free(req, &cc->req_pool);
1508 }
1509 
1510 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1511 {
1512 	if (crypt_integrity_aead(cc))
1513 		crypt_free_req_aead(cc, req, base_bio);
1514 	else
1515 		crypt_free_req_skcipher(cc, req, base_bio);
1516 }
1517 
1518 /*
1519  * Encrypt / decrypt data from one bio to another one (can be the same one)
1520  */
1521 static blk_status_t crypt_convert(struct crypt_config *cc,
1522 			 struct convert_context *ctx)
1523 {
1524 	unsigned int tag_offset = 0;
1525 	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1526 	int r;
1527 
1528 	atomic_set(&ctx->cc_pending, 1);
1529 
1530 	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1531 
1532 		crypt_alloc_req(cc, ctx);
1533 		atomic_inc(&ctx->cc_pending);
1534 
1535 		if (crypt_integrity_aead(cc))
1536 			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1537 		else
1538 			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1539 
1540 		switch (r) {
1541 		/*
1542 		 * The request was queued by a crypto driver
1543 		 * but the driver request queue is full, let's wait.
1544 		 */
1545 		case -EBUSY:
1546 			wait_for_completion(&ctx->restart);
1547 			reinit_completion(&ctx->restart);
1548 			/* fall through */
1549 		/*
1550 		 * The request is queued and processed asynchronously,
1551 		 * completion function kcryptd_async_done() will be called.
1552 		 */
1553 		case -EINPROGRESS:
1554 			ctx->r.req = NULL;
1555 			ctx->cc_sector += sector_step;
1556 			tag_offset++;
1557 			continue;
1558 		/*
1559 		 * The request was already processed (synchronously).
1560 		 */
1561 		case 0:
1562 			atomic_dec(&ctx->cc_pending);
1563 			ctx->cc_sector += sector_step;
1564 			tag_offset++;
1565 			cond_resched();
1566 			continue;
1567 		/*
1568 		 * There was a data integrity error.
1569 		 */
1570 		case -EBADMSG:
1571 			atomic_dec(&ctx->cc_pending);
1572 			return BLK_STS_PROTECTION;
1573 		/*
1574 		 * There was an error while processing the request.
1575 		 */
1576 		default:
1577 			atomic_dec(&ctx->cc_pending);
1578 			return BLK_STS_IOERR;
1579 		}
1580 	}
1581 
1582 	return 0;
1583 }
1584 
1585 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1586 
1587 /*
1588  * Generate a new unfragmented bio with the given size
1589  * This should never violate the device limitations (but only because
1590  * max_segment_size is being constrained to PAGE_SIZE).
1591  *
1592  * This function may be called concurrently. If we allocate from the mempool
1593  * concurrently, there is a possibility of deadlock. For example, if we have
1594  * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1595  * the mempool concurrently, it may deadlock in a situation where both processes
1596  * have allocated 128 pages and the mempool is exhausted.
1597  *
1598  * In order to avoid this scenario we allocate the pages under a mutex.
1599  *
1600  * In order to not degrade performance with excessive locking, we try
1601  * non-blocking allocations without a mutex first but on failure we fallback
1602  * to blocking allocations with a mutex.
1603  */
1604 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size)
1605 {
1606 	struct crypt_config *cc = io->cc;
1607 	struct bio *clone;
1608 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1609 	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1610 	unsigned i, len, remaining_size;
1611 	struct page *page;
1612 
1613 retry:
1614 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1615 		mutex_lock(&cc->bio_alloc_lock);
1616 
1617 	clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, &cc->bs);
1618 	if (!clone)
1619 		goto out;
1620 
1621 	clone_init(io, clone);
1622 
1623 	remaining_size = size;
1624 
1625 	for (i = 0; i < nr_iovecs; i++) {
1626 		page = mempool_alloc(&cc->page_pool, gfp_mask);
1627 		if (!page) {
1628 			crypt_free_buffer_pages(cc, clone);
1629 			bio_put(clone);
1630 			gfp_mask |= __GFP_DIRECT_RECLAIM;
1631 			goto retry;
1632 		}
1633 
1634 		len = (remaining_size > PAGE_SIZE) ? PAGE_SIZE : remaining_size;
1635 
1636 		bio_add_page(clone, page, len, 0);
1637 
1638 		remaining_size -= len;
1639 	}
1640 
1641 	/* Allocate space for integrity tags */
1642 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1643 		crypt_free_buffer_pages(cc, clone);
1644 		bio_put(clone);
1645 		clone = NULL;
1646 	}
1647 out:
1648 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1649 		mutex_unlock(&cc->bio_alloc_lock);
1650 
1651 	return clone;
1652 }
1653 
1654 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1655 {
1656 	struct bio_vec *bv;
1657 	struct bvec_iter_all iter_all;
1658 
1659 	bio_for_each_segment_all(bv, clone, iter_all) {
1660 		BUG_ON(!bv->bv_page);
1661 		mempool_free(bv->bv_page, &cc->page_pool);
1662 	}
1663 }
1664 
1665 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1666 			  struct bio *bio, sector_t sector)
1667 {
1668 	io->cc = cc;
1669 	io->base_bio = bio;
1670 	io->sector = sector;
1671 	io->error = 0;
1672 	io->ctx.r.req = NULL;
1673 	io->integrity_metadata = NULL;
1674 	io->integrity_metadata_from_pool = false;
1675 	atomic_set(&io->io_pending, 0);
1676 }
1677 
1678 static void crypt_inc_pending(struct dm_crypt_io *io)
1679 {
1680 	atomic_inc(&io->io_pending);
1681 }
1682 
1683 /*
1684  * One of the bios was finished. Check for completion of
1685  * the whole request and correctly clean up the buffer.
1686  */
1687 static void crypt_dec_pending(struct dm_crypt_io *io)
1688 {
1689 	struct crypt_config *cc = io->cc;
1690 	struct bio *base_bio = io->base_bio;
1691 	blk_status_t error = io->error;
1692 
1693 	if (!atomic_dec_and_test(&io->io_pending))
1694 		return;
1695 
1696 	if (io->ctx.r.req)
1697 		crypt_free_req(cc, io->ctx.r.req, base_bio);
1698 
1699 	if (unlikely(io->integrity_metadata_from_pool))
1700 		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1701 	else
1702 		kfree(io->integrity_metadata);
1703 
1704 	base_bio->bi_status = error;
1705 	bio_endio(base_bio);
1706 }
1707 
1708 /*
1709  * kcryptd/kcryptd_io:
1710  *
1711  * Needed because it would be very unwise to do decryption in an
1712  * interrupt context.
1713  *
1714  * kcryptd performs the actual encryption or decryption.
1715  *
1716  * kcryptd_io performs the IO submission.
1717  *
1718  * They must be separated as otherwise the final stages could be
1719  * starved by new requests which can block in the first stages due
1720  * to memory allocation.
1721  *
1722  * The work is done per CPU global for all dm-crypt instances.
1723  * They should not depend on each other and do not block.
1724  */
1725 static void crypt_endio(struct bio *clone)
1726 {
1727 	struct dm_crypt_io *io = clone->bi_private;
1728 	struct crypt_config *cc = io->cc;
1729 	unsigned rw = bio_data_dir(clone);
1730 	blk_status_t error;
1731 
1732 	/*
1733 	 * free the processed pages
1734 	 */
1735 	if (rw == WRITE)
1736 		crypt_free_buffer_pages(cc, clone);
1737 
1738 	error = clone->bi_status;
1739 	bio_put(clone);
1740 
1741 	if (rw == READ && !error) {
1742 		kcryptd_queue_crypt(io);
1743 		return;
1744 	}
1745 
1746 	if (unlikely(error))
1747 		io->error = error;
1748 
1749 	crypt_dec_pending(io);
1750 }
1751 
1752 static void clone_init(struct dm_crypt_io *io, struct bio *clone)
1753 {
1754 	struct crypt_config *cc = io->cc;
1755 
1756 	clone->bi_private = io;
1757 	clone->bi_end_io  = crypt_endio;
1758 	bio_set_dev(clone, cc->dev->bdev);
1759 	clone->bi_opf	  = io->base_bio->bi_opf;
1760 }
1761 
1762 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1763 {
1764 	struct crypt_config *cc = io->cc;
1765 	struct bio *clone;
1766 
1767 	/*
1768 	 * We need the original biovec array in order to decrypt
1769 	 * the whole bio data *afterwards* -- thanks to immutable
1770 	 * biovecs we don't need to worry about the block layer
1771 	 * modifying the biovec array; so leverage bio_clone_fast().
1772 	 */
1773 	clone = bio_clone_fast(io->base_bio, gfp, &cc->bs);
1774 	if (!clone)
1775 		return 1;
1776 
1777 	crypt_inc_pending(io);
1778 
1779 	clone_init(io, clone);
1780 	clone->bi_iter.bi_sector = cc->start + io->sector;
1781 
1782 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1783 		crypt_dec_pending(io);
1784 		bio_put(clone);
1785 		return 1;
1786 	}
1787 
1788 	generic_make_request(clone);
1789 	return 0;
1790 }
1791 
1792 static void kcryptd_io_read_work(struct work_struct *work)
1793 {
1794 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1795 
1796 	crypt_inc_pending(io);
1797 	if (kcryptd_io_read(io, GFP_NOIO))
1798 		io->error = BLK_STS_RESOURCE;
1799 	crypt_dec_pending(io);
1800 }
1801 
1802 static void kcryptd_queue_read(struct dm_crypt_io *io)
1803 {
1804 	struct crypt_config *cc = io->cc;
1805 
1806 	INIT_WORK(&io->work, kcryptd_io_read_work);
1807 	queue_work(cc->io_queue, &io->work);
1808 }
1809 
1810 static void kcryptd_io_write(struct dm_crypt_io *io)
1811 {
1812 	struct bio *clone = io->ctx.bio_out;
1813 
1814 	generic_make_request(clone);
1815 }
1816 
1817 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1818 
1819 static int dmcrypt_write(void *data)
1820 {
1821 	struct crypt_config *cc = data;
1822 	struct dm_crypt_io *io;
1823 
1824 	while (1) {
1825 		struct rb_root write_tree;
1826 		struct blk_plug plug;
1827 
1828 		spin_lock_irq(&cc->write_thread_lock);
1829 continue_locked:
1830 
1831 		if (!RB_EMPTY_ROOT(&cc->write_tree))
1832 			goto pop_from_list;
1833 
1834 		set_current_state(TASK_INTERRUPTIBLE);
1835 
1836 		spin_unlock_irq(&cc->write_thread_lock);
1837 
1838 		if (unlikely(kthread_should_stop())) {
1839 			set_current_state(TASK_RUNNING);
1840 			break;
1841 		}
1842 
1843 		schedule();
1844 
1845 		set_current_state(TASK_RUNNING);
1846 		spin_lock_irq(&cc->write_thread_lock);
1847 		goto continue_locked;
1848 
1849 pop_from_list:
1850 		write_tree = cc->write_tree;
1851 		cc->write_tree = RB_ROOT;
1852 		spin_unlock_irq(&cc->write_thread_lock);
1853 
1854 		BUG_ON(rb_parent(write_tree.rb_node));
1855 
1856 		/*
1857 		 * Note: we cannot walk the tree here with rb_next because
1858 		 * the structures may be freed when kcryptd_io_write is called.
1859 		 */
1860 		blk_start_plug(&plug);
1861 		do {
1862 			io = crypt_io_from_node(rb_first(&write_tree));
1863 			rb_erase(&io->rb_node, &write_tree);
1864 			kcryptd_io_write(io);
1865 		} while (!RB_EMPTY_ROOT(&write_tree));
1866 		blk_finish_plug(&plug);
1867 	}
1868 	return 0;
1869 }
1870 
1871 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1872 {
1873 	struct bio *clone = io->ctx.bio_out;
1874 	struct crypt_config *cc = io->cc;
1875 	unsigned long flags;
1876 	sector_t sector;
1877 	struct rb_node **rbp, *parent;
1878 
1879 	if (unlikely(io->error)) {
1880 		crypt_free_buffer_pages(cc, clone);
1881 		bio_put(clone);
1882 		crypt_dec_pending(io);
1883 		return;
1884 	}
1885 
1886 	/* crypt_convert should have filled the clone bio */
1887 	BUG_ON(io->ctx.iter_out.bi_size);
1888 
1889 	clone->bi_iter.bi_sector = cc->start + io->sector;
1890 
1891 	if (likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) {
1892 		generic_make_request(clone);
1893 		return;
1894 	}
1895 
1896 	spin_lock_irqsave(&cc->write_thread_lock, flags);
1897 	if (RB_EMPTY_ROOT(&cc->write_tree))
1898 		wake_up_process(cc->write_thread);
1899 	rbp = &cc->write_tree.rb_node;
1900 	parent = NULL;
1901 	sector = io->sector;
1902 	while (*rbp) {
1903 		parent = *rbp;
1904 		if (sector < crypt_io_from_node(parent)->sector)
1905 			rbp = &(*rbp)->rb_left;
1906 		else
1907 			rbp = &(*rbp)->rb_right;
1908 	}
1909 	rb_link_node(&io->rb_node, parent, rbp);
1910 	rb_insert_color(&io->rb_node, &cc->write_tree);
1911 	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
1912 }
1913 
1914 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
1915 {
1916 	struct crypt_config *cc = io->cc;
1917 	struct bio *clone;
1918 	int crypt_finished;
1919 	sector_t sector = io->sector;
1920 	blk_status_t r;
1921 
1922 	/*
1923 	 * Prevent io from disappearing until this function completes.
1924 	 */
1925 	crypt_inc_pending(io);
1926 	crypt_convert_init(cc, &io->ctx, NULL, io->base_bio, sector);
1927 
1928 	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
1929 	if (unlikely(!clone)) {
1930 		io->error = BLK_STS_IOERR;
1931 		goto dec;
1932 	}
1933 
1934 	io->ctx.bio_out = clone;
1935 	io->ctx.iter_out = clone->bi_iter;
1936 
1937 	sector += bio_sectors(clone);
1938 
1939 	crypt_inc_pending(io);
1940 	r = crypt_convert(cc, &io->ctx);
1941 	if (r)
1942 		io->error = r;
1943 	crypt_finished = atomic_dec_and_test(&io->ctx.cc_pending);
1944 
1945 	/* Encryption was already finished, submit io now */
1946 	if (crypt_finished) {
1947 		kcryptd_crypt_write_io_submit(io, 0);
1948 		io->sector = sector;
1949 	}
1950 
1951 dec:
1952 	crypt_dec_pending(io);
1953 }
1954 
1955 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
1956 {
1957 	crypt_dec_pending(io);
1958 }
1959 
1960 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
1961 {
1962 	struct crypt_config *cc = io->cc;
1963 	blk_status_t r;
1964 
1965 	crypt_inc_pending(io);
1966 
1967 	crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
1968 			   io->sector);
1969 
1970 	r = crypt_convert(cc, &io->ctx);
1971 	if (r)
1972 		io->error = r;
1973 
1974 	if (atomic_dec_and_test(&io->ctx.cc_pending))
1975 		kcryptd_crypt_read_done(io);
1976 
1977 	crypt_dec_pending(io);
1978 }
1979 
1980 static void kcryptd_async_done(struct crypto_async_request *async_req,
1981 			       int error)
1982 {
1983 	struct dm_crypt_request *dmreq = async_req->data;
1984 	struct convert_context *ctx = dmreq->ctx;
1985 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1986 	struct crypt_config *cc = io->cc;
1987 
1988 	/*
1989 	 * A request from crypto driver backlog is going to be processed now,
1990 	 * finish the completion and continue in crypt_convert().
1991 	 * (Callback will be called for the second time for this request.)
1992 	 */
1993 	if (error == -EINPROGRESS) {
1994 		complete(&ctx->restart);
1995 		return;
1996 	}
1997 
1998 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
1999 		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2000 
2001 	if (error == -EBADMSG) {
2002 		char b[BDEVNAME_SIZE];
2003 		DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", bio_devname(ctx->bio_in, b),
2004 			    (unsigned long long)le64_to_cpu(*org_sector_of_dmreq(cc, dmreq)));
2005 		io->error = BLK_STS_PROTECTION;
2006 	} else if (error < 0)
2007 		io->error = BLK_STS_IOERR;
2008 
2009 	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2010 
2011 	if (!atomic_dec_and_test(&ctx->cc_pending))
2012 		return;
2013 
2014 	if (bio_data_dir(io->base_bio) == READ)
2015 		kcryptd_crypt_read_done(io);
2016 	else
2017 		kcryptd_crypt_write_io_submit(io, 1);
2018 }
2019 
2020 static void kcryptd_crypt(struct work_struct *work)
2021 {
2022 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2023 
2024 	if (bio_data_dir(io->base_bio) == READ)
2025 		kcryptd_crypt_read_convert(io);
2026 	else
2027 		kcryptd_crypt_write_convert(io);
2028 }
2029 
2030 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2031 {
2032 	struct crypt_config *cc = io->cc;
2033 
2034 	INIT_WORK(&io->work, kcryptd_crypt);
2035 	queue_work(cc->crypt_queue, &io->work);
2036 }
2037 
2038 static void crypt_free_tfms_aead(struct crypt_config *cc)
2039 {
2040 	if (!cc->cipher_tfm.tfms_aead)
2041 		return;
2042 
2043 	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2044 		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2045 		cc->cipher_tfm.tfms_aead[0] = NULL;
2046 	}
2047 
2048 	kfree(cc->cipher_tfm.tfms_aead);
2049 	cc->cipher_tfm.tfms_aead = NULL;
2050 }
2051 
2052 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2053 {
2054 	unsigned i;
2055 
2056 	if (!cc->cipher_tfm.tfms)
2057 		return;
2058 
2059 	for (i = 0; i < cc->tfms_count; i++)
2060 		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2061 			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2062 			cc->cipher_tfm.tfms[i] = NULL;
2063 		}
2064 
2065 	kfree(cc->cipher_tfm.tfms);
2066 	cc->cipher_tfm.tfms = NULL;
2067 }
2068 
2069 static void crypt_free_tfms(struct crypt_config *cc)
2070 {
2071 	if (crypt_integrity_aead(cc))
2072 		crypt_free_tfms_aead(cc);
2073 	else
2074 		crypt_free_tfms_skcipher(cc);
2075 }
2076 
2077 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2078 {
2079 	unsigned i;
2080 	int err;
2081 
2082 	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2083 				      sizeof(struct crypto_skcipher *),
2084 				      GFP_KERNEL);
2085 	if (!cc->cipher_tfm.tfms)
2086 		return -ENOMEM;
2087 
2088 	for (i = 0; i < cc->tfms_count; i++) {
2089 		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0, 0);
2090 		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2091 			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2092 			crypt_free_tfms(cc);
2093 			return err;
2094 		}
2095 	}
2096 
2097 	/*
2098 	 * dm-crypt performance can vary greatly depending on which crypto
2099 	 * algorithm implementation is used.  Help people debug performance
2100 	 * problems by logging the ->cra_driver_name.
2101 	 */
2102 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2103 	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2104 	return 0;
2105 }
2106 
2107 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2108 {
2109 	int err;
2110 
2111 	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2112 	if (!cc->cipher_tfm.tfms)
2113 		return -ENOMEM;
2114 
2115 	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0, 0);
2116 	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2117 		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2118 		crypt_free_tfms(cc);
2119 		return err;
2120 	}
2121 
2122 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2123 	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2124 	return 0;
2125 }
2126 
2127 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2128 {
2129 	if (crypt_integrity_aead(cc))
2130 		return crypt_alloc_tfms_aead(cc, ciphermode);
2131 	else
2132 		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2133 }
2134 
2135 static unsigned crypt_subkey_size(struct crypt_config *cc)
2136 {
2137 	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2138 }
2139 
2140 static unsigned crypt_authenckey_size(struct crypt_config *cc)
2141 {
2142 	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2143 }
2144 
2145 /*
2146  * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2147  * the key must be for some reason in special format.
2148  * This funcion converts cc->key to this special format.
2149  */
2150 static void crypt_copy_authenckey(char *p, const void *key,
2151 				  unsigned enckeylen, unsigned authkeylen)
2152 {
2153 	struct crypto_authenc_key_param *param;
2154 	struct rtattr *rta;
2155 
2156 	rta = (struct rtattr *)p;
2157 	param = RTA_DATA(rta);
2158 	param->enckeylen = cpu_to_be32(enckeylen);
2159 	rta->rta_len = RTA_LENGTH(sizeof(*param));
2160 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2161 	p += RTA_SPACE(sizeof(*param));
2162 	memcpy(p, key + enckeylen, authkeylen);
2163 	p += authkeylen;
2164 	memcpy(p, key, enckeylen);
2165 }
2166 
2167 static int crypt_setkey(struct crypt_config *cc)
2168 {
2169 	unsigned subkey_size;
2170 	int err = 0, i, r;
2171 
2172 	/* Ignore extra keys (which are used for IV etc) */
2173 	subkey_size = crypt_subkey_size(cc);
2174 
2175 	if (crypt_integrity_hmac(cc)) {
2176 		if (subkey_size < cc->key_mac_size)
2177 			return -EINVAL;
2178 
2179 		crypt_copy_authenckey(cc->authenc_key, cc->key,
2180 				      subkey_size - cc->key_mac_size,
2181 				      cc->key_mac_size);
2182 	}
2183 
2184 	for (i = 0; i < cc->tfms_count; i++) {
2185 		if (crypt_integrity_hmac(cc))
2186 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2187 				cc->authenc_key, crypt_authenckey_size(cc));
2188 		else if (crypt_integrity_aead(cc))
2189 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2190 					       cc->key + (i * subkey_size),
2191 					       subkey_size);
2192 		else
2193 			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2194 						   cc->key + (i * subkey_size),
2195 						   subkey_size);
2196 		if (r)
2197 			err = r;
2198 	}
2199 
2200 	if (crypt_integrity_hmac(cc))
2201 		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2202 
2203 	return err;
2204 }
2205 
2206 #ifdef CONFIG_KEYS
2207 
2208 static bool contains_whitespace(const char *str)
2209 {
2210 	while (*str)
2211 		if (isspace(*str++))
2212 			return true;
2213 	return false;
2214 }
2215 
2216 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2217 {
2218 	char *new_key_string, *key_desc;
2219 	int ret;
2220 	struct key *key;
2221 	const struct user_key_payload *ukp;
2222 
2223 	/*
2224 	 * Reject key_string with whitespace. dm core currently lacks code for
2225 	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2226 	 */
2227 	if (contains_whitespace(key_string)) {
2228 		DMERR("whitespace chars not allowed in key string");
2229 		return -EINVAL;
2230 	}
2231 
2232 	/* look for next ':' separating key_type from key_description */
2233 	key_desc = strpbrk(key_string, ":");
2234 	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2235 		return -EINVAL;
2236 
2237 	if (strncmp(key_string, "logon:", key_desc - key_string + 1) &&
2238 	    strncmp(key_string, "user:", key_desc - key_string + 1))
2239 		return -EINVAL;
2240 
2241 	new_key_string = kstrdup(key_string, GFP_KERNEL);
2242 	if (!new_key_string)
2243 		return -ENOMEM;
2244 
2245 	key = request_key(key_string[0] == 'l' ? &key_type_logon : &key_type_user,
2246 			  key_desc + 1, NULL);
2247 	if (IS_ERR(key)) {
2248 		kzfree(new_key_string);
2249 		return PTR_ERR(key);
2250 	}
2251 
2252 	down_read(&key->sem);
2253 
2254 	ukp = user_key_payload_locked(key);
2255 	if (!ukp) {
2256 		up_read(&key->sem);
2257 		key_put(key);
2258 		kzfree(new_key_string);
2259 		return -EKEYREVOKED;
2260 	}
2261 
2262 	if (cc->key_size != ukp->datalen) {
2263 		up_read(&key->sem);
2264 		key_put(key);
2265 		kzfree(new_key_string);
2266 		return -EINVAL;
2267 	}
2268 
2269 	memcpy(cc->key, ukp->data, cc->key_size);
2270 
2271 	up_read(&key->sem);
2272 	key_put(key);
2273 
2274 	/* clear the flag since following operations may invalidate previously valid key */
2275 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2276 
2277 	ret = crypt_setkey(cc);
2278 
2279 	if (!ret) {
2280 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2281 		kzfree(cc->key_string);
2282 		cc->key_string = new_key_string;
2283 	} else
2284 		kzfree(new_key_string);
2285 
2286 	return ret;
2287 }
2288 
2289 static int get_key_size(char **key_string)
2290 {
2291 	char *colon, dummy;
2292 	int ret;
2293 
2294 	if (*key_string[0] != ':')
2295 		return strlen(*key_string) >> 1;
2296 
2297 	/* look for next ':' in key string */
2298 	colon = strpbrk(*key_string + 1, ":");
2299 	if (!colon)
2300 		return -EINVAL;
2301 
2302 	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2303 		return -EINVAL;
2304 
2305 	*key_string = colon;
2306 
2307 	/* remaining key string should be :<logon|user>:<key_desc> */
2308 
2309 	return ret;
2310 }
2311 
2312 #else
2313 
2314 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2315 {
2316 	return -EINVAL;
2317 }
2318 
2319 static int get_key_size(char **key_string)
2320 {
2321 	return (*key_string[0] == ':') ? -EINVAL : strlen(*key_string) >> 1;
2322 }
2323 
2324 #endif
2325 
2326 static int crypt_set_key(struct crypt_config *cc, char *key)
2327 {
2328 	int r = -EINVAL;
2329 	int key_string_len = strlen(key);
2330 
2331 	/* Hyphen (which gives a key_size of zero) means there is no key. */
2332 	if (!cc->key_size && strcmp(key, "-"))
2333 		goto out;
2334 
2335 	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2336 	if (key[0] == ':') {
2337 		r = crypt_set_keyring_key(cc, key + 1);
2338 		goto out;
2339 	}
2340 
2341 	/* clear the flag since following operations may invalidate previously valid key */
2342 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2343 
2344 	/* wipe references to any kernel keyring key */
2345 	kzfree(cc->key_string);
2346 	cc->key_string = NULL;
2347 
2348 	/* Decode key from its hex representation. */
2349 	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2350 		goto out;
2351 
2352 	r = crypt_setkey(cc);
2353 	if (!r)
2354 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2355 
2356 out:
2357 	/* Hex key string not needed after here, so wipe it. */
2358 	memset(key, '0', key_string_len);
2359 
2360 	return r;
2361 }
2362 
2363 static int crypt_wipe_key(struct crypt_config *cc)
2364 {
2365 	int r;
2366 
2367 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2368 	get_random_bytes(&cc->key, cc->key_size);
2369 
2370 	/* Wipe IV private keys */
2371 	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2372 		r = cc->iv_gen_ops->wipe(cc);
2373 		if (r)
2374 			return r;
2375 	}
2376 
2377 	kzfree(cc->key_string);
2378 	cc->key_string = NULL;
2379 	r = crypt_setkey(cc);
2380 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2381 
2382 	return r;
2383 }
2384 
2385 static void crypt_calculate_pages_per_client(void)
2386 {
2387 	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2388 
2389 	if (!dm_crypt_clients_n)
2390 		return;
2391 
2392 	pages /= dm_crypt_clients_n;
2393 	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2394 		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2395 	dm_crypt_pages_per_client = pages;
2396 }
2397 
2398 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2399 {
2400 	struct crypt_config *cc = pool_data;
2401 	struct page *page;
2402 
2403 	if (unlikely(percpu_counter_compare(&cc->n_allocated_pages, dm_crypt_pages_per_client) >= 0) &&
2404 	    likely(gfp_mask & __GFP_NORETRY))
2405 		return NULL;
2406 
2407 	page = alloc_page(gfp_mask);
2408 	if (likely(page != NULL))
2409 		percpu_counter_add(&cc->n_allocated_pages, 1);
2410 
2411 	return page;
2412 }
2413 
2414 static void crypt_page_free(void *page, void *pool_data)
2415 {
2416 	struct crypt_config *cc = pool_data;
2417 
2418 	__free_page(page);
2419 	percpu_counter_sub(&cc->n_allocated_pages, 1);
2420 }
2421 
2422 static void crypt_dtr(struct dm_target *ti)
2423 {
2424 	struct crypt_config *cc = ti->private;
2425 
2426 	ti->private = NULL;
2427 
2428 	if (!cc)
2429 		return;
2430 
2431 	if (cc->write_thread)
2432 		kthread_stop(cc->write_thread);
2433 
2434 	if (cc->io_queue)
2435 		destroy_workqueue(cc->io_queue);
2436 	if (cc->crypt_queue)
2437 		destroy_workqueue(cc->crypt_queue);
2438 
2439 	crypt_free_tfms(cc);
2440 
2441 	bioset_exit(&cc->bs);
2442 
2443 	mempool_exit(&cc->page_pool);
2444 	mempool_exit(&cc->req_pool);
2445 	mempool_exit(&cc->tag_pool);
2446 
2447 	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2448 	percpu_counter_destroy(&cc->n_allocated_pages);
2449 
2450 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2451 		cc->iv_gen_ops->dtr(cc);
2452 
2453 	if (cc->dev)
2454 		dm_put_device(ti, cc->dev);
2455 
2456 	kzfree(cc->cipher_string);
2457 	kzfree(cc->key_string);
2458 	kzfree(cc->cipher_auth);
2459 	kzfree(cc->authenc_key);
2460 
2461 	mutex_destroy(&cc->bio_alloc_lock);
2462 
2463 	/* Must zero key material before freeing */
2464 	kzfree(cc);
2465 
2466 	spin_lock(&dm_crypt_clients_lock);
2467 	WARN_ON(!dm_crypt_clients_n);
2468 	dm_crypt_clients_n--;
2469 	crypt_calculate_pages_per_client();
2470 	spin_unlock(&dm_crypt_clients_lock);
2471 }
2472 
2473 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2474 {
2475 	struct crypt_config *cc = ti->private;
2476 
2477 	if (crypt_integrity_aead(cc))
2478 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2479 	else
2480 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2481 
2482 	if (cc->iv_size)
2483 		/* at least a 64 bit sector number should fit in our buffer */
2484 		cc->iv_size = max(cc->iv_size,
2485 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2486 	else if (ivmode) {
2487 		DMWARN("Selected cipher does not support IVs");
2488 		ivmode = NULL;
2489 	}
2490 
2491 	/* Choose ivmode, see comments at iv code. */
2492 	if (ivmode == NULL)
2493 		cc->iv_gen_ops = NULL;
2494 	else if (strcmp(ivmode, "plain") == 0)
2495 		cc->iv_gen_ops = &crypt_iv_plain_ops;
2496 	else if (strcmp(ivmode, "plain64") == 0)
2497 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2498 	else if (strcmp(ivmode, "plain64be") == 0)
2499 		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2500 	else if (strcmp(ivmode, "essiv") == 0)
2501 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2502 	else if (strcmp(ivmode, "benbi") == 0)
2503 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2504 	else if (strcmp(ivmode, "null") == 0)
2505 		cc->iv_gen_ops = &crypt_iv_null_ops;
2506 	else if (strcmp(ivmode, "eboiv") == 0)
2507 		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2508 	else if (strcmp(ivmode, "elephant") == 0) {
2509 		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2510 		cc->key_parts = 2;
2511 		cc->key_extra_size = cc->key_size / 2;
2512 		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2513 			return -EINVAL;
2514 		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2515 	} else if (strcmp(ivmode, "lmk") == 0) {
2516 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2517 		/*
2518 		 * Version 2 and 3 is recognised according
2519 		 * to length of provided multi-key string.
2520 		 * If present (version 3), last key is used as IV seed.
2521 		 * All keys (including IV seed) are always the same size.
2522 		 */
2523 		if (cc->key_size % cc->key_parts) {
2524 			cc->key_parts++;
2525 			cc->key_extra_size = cc->key_size / cc->key_parts;
2526 		}
2527 	} else if (strcmp(ivmode, "tcw") == 0) {
2528 		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2529 		cc->key_parts += 2; /* IV + whitening */
2530 		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2531 	} else if (strcmp(ivmode, "random") == 0) {
2532 		cc->iv_gen_ops = &crypt_iv_random_ops;
2533 		/* Need storage space in integrity fields. */
2534 		cc->integrity_iv_size = cc->iv_size;
2535 	} else {
2536 		ti->error = "Invalid IV mode";
2537 		return -EINVAL;
2538 	}
2539 
2540 	return 0;
2541 }
2542 
2543 /*
2544  * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2545  * The HMAC is needed to calculate tag size (HMAC digest size).
2546  * This should be probably done by crypto-api calls (once available...)
2547  */
2548 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2549 {
2550 	char *start, *end, *mac_alg = NULL;
2551 	struct crypto_ahash *mac;
2552 
2553 	if (!strstarts(cipher_api, "authenc("))
2554 		return 0;
2555 
2556 	start = strchr(cipher_api, '(');
2557 	end = strchr(cipher_api, ',');
2558 	if (!start || !end || ++start > end)
2559 		return -EINVAL;
2560 
2561 	mac_alg = kzalloc(end - start + 1, GFP_KERNEL);
2562 	if (!mac_alg)
2563 		return -ENOMEM;
2564 	strncpy(mac_alg, start, end - start);
2565 
2566 	mac = crypto_alloc_ahash(mac_alg, 0, 0);
2567 	kfree(mac_alg);
2568 
2569 	if (IS_ERR(mac))
2570 		return PTR_ERR(mac);
2571 
2572 	cc->key_mac_size = crypto_ahash_digestsize(mac);
2573 	crypto_free_ahash(mac);
2574 
2575 	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2576 	if (!cc->authenc_key)
2577 		return -ENOMEM;
2578 
2579 	return 0;
2580 }
2581 
2582 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2583 				char **ivmode, char **ivopts)
2584 {
2585 	struct crypt_config *cc = ti->private;
2586 	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2587 	int ret = -EINVAL;
2588 
2589 	cc->tfms_count = 1;
2590 
2591 	/*
2592 	 * New format (capi: prefix)
2593 	 * capi:cipher_api_spec-iv:ivopts
2594 	 */
2595 	tmp = &cipher_in[strlen("capi:")];
2596 
2597 	/* Separate IV options if present, it can contain another '-' in hash name */
2598 	*ivopts = strrchr(tmp, ':');
2599 	if (*ivopts) {
2600 		**ivopts = '\0';
2601 		(*ivopts)++;
2602 	}
2603 	/* Parse IV mode */
2604 	*ivmode = strrchr(tmp, '-');
2605 	if (*ivmode) {
2606 		**ivmode = '\0';
2607 		(*ivmode)++;
2608 	}
2609 	/* The rest is crypto API spec */
2610 	cipher_api = tmp;
2611 
2612 	/* Alloc AEAD, can be used only in new format. */
2613 	if (crypt_integrity_aead(cc)) {
2614 		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2615 		if (ret < 0) {
2616 			ti->error = "Invalid AEAD cipher spec";
2617 			return -ENOMEM;
2618 		}
2619 	}
2620 
2621 	if (*ivmode && !strcmp(*ivmode, "lmk"))
2622 		cc->tfms_count = 64;
2623 
2624 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2625 		if (!*ivopts) {
2626 			ti->error = "Digest algorithm missing for ESSIV mode";
2627 			return -EINVAL;
2628 		}
2629 		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2630 			       cipher_api, *ivopts);
2631 		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2632 			ti->error = "Cannot allocate cipher string";
2633 			return -ENOMEM;
2634 		}
2635 		cipher_api = buf;
2636 	}
2637 
2638 	cc->key_parts = cc->tfms_count;
2639 
2640 	/* Allocate cipher */
2641 	ret = crypt_alloc_tfms(cc, cipher_api);
2642 	if (ret < 0) {
2643 		ti->error = "Error allocating crypto tfm";
2644 		return ret;
2645 	}
2646 
2647 	if (crypt_integrity_aead(cc))
2648 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2649 	else
2650 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2651 
2652 	return 0;
2653 }
2654 
2655 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2656 				char **ivmode, char **ivopts)
2657 {
2658 	struct crypt_config *cc = ti->private;
2659 	char *tmp, *cipher, *chainmode, *keycount;
2660 	char *cipher_api = NULL;
2661 	int ret = -EINVAL;
2662 	char dummy;
2663 
2664 	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2665 		ti->error = "Bad cipher specification";
2666 		return -EINVAL;
2667 	}
2668 
2669 	/*
2670 	 * Legacy dm-crypt cipher specification
2671 	 * cipher[:keycount]-mode-iv:ivopts
2672 	 */
2673 	tmp = cipher_in;
2674 	keycount = strsep(&tmp, "-");
2675 	cipher = strsep(&keycount, ":");
2676 
2677 	if (!keycount)
2678 		cc->tfms_count = 1;
2679 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2680 		 !is_power_of_2(cc->tfms_count)) {
2681 		ti->error = "Bad cipher key count specification";
2682 		return -EINVAL;
2683 	}
2684 	cc->key_parts = cc->tfms_count;
2685 
2686 	chainmode = strsep(&tmp, "-");
2687 	*ivmode = strsep(&tmp, ":");
2688 	*ivopts = tmp;
2689 
2690 	/*
2691 	 * For compatibility with the original dm-crypt mapping format, if
2692 	 * only the cipher name is supplied, use cbc-plain.
2693 	 */
2694 	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2695 		chainmode = "cbc";
2696 		*ivmode = "plain";
2697 	}
2698 
2699 	if (strcmp(chainmode, "ecb") && !*ivmode) {
2700 		ti->error = "IV mechanism required";
2701 		return -EINVAL;
2702 	}
2703 
2704 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
2705 	if (!cipher_api)
2706 		goto bad_mem;
2707 
2708 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2709 		if (!*ivopts) {
2710 			ti->error = "Digest algorithm missing for ESSIV mode";
2711 			kfree(cipher_api);
2712 			return -EINVAL;
2713 		}
2714 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2715 			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
2716 	} else {
2717 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2718 			       "%s(%s)", chainmode, cipher);
2719 	}
2720 	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2721 		kfree(cipher_api);
2722 		goto bad_mem;
2723 	}
2724 
2725 	/* Allocate cipher */
2726 	ret = crypt_alloc_tfms(cc, cipher_api);
2727 	if (ret < 0) {
2728 		ti->error = "Error allocating crypto tfm";
2729 		kfree(cipher_api);
2730 		return ret;
2731 	}
2732 	kfree(cipher_api);
2733 
2734 	return 0;
2735 bad_mem:
2736 	ti->error = "Cannot allocate cipher strings";
2737 	return -ENOMEM;
2738 }
2739 
2740 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
2741 {
2742 	struct crypt_config *cc = ti->private;
2743 	char *ivmode = NULL, *ivopts = NULL;
2744 	int ret;
2745 
2746 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
2747 	if (!cc->cipher_string) {
2748 		ti->error = "Cannot allocate cipher strings";
2749 		return -ENOMEM;
2750 	}
2751 
2752 	if (strstarts(cipher_in, "capi:"))
2753 		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
2754 	else
2755 		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
2756 	if (ret)
2757 		return ret;
2758 
2759 	/* Initialize IV */
2760 	ret = crypt_ctr_ivmode(ti, ivmode);
2761 	if (ret < 0)
2762 		return ret;
2763 
2764 	/* Initialize and set key */
2765 	ret = crypt_set_key(cc, key);
2766 	if (ret < 0) {
2767 		ti->error = "Error decoding and setting key";
2768 		return ret;
2769 	}
2770 
2771 	/* Allocate IV */
2772 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
2773 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
2774 		if (ret < 0) {
2775 			ti->error = "Error creating IV";
2776 			return ret;
2777 		}
2778 	}
2779 
2780 	/* Initialize IV (set keys for ESSIV etc) */
2781 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
2782 		ret = cc->iv_gen_ops->init(cc);
2783 		if (ret < 0) {
2784 			ti->error = "Error initialising IV";
2785 			return ret;
2786 		}
2787 	}
2788 
2789 	/* wipe the kernel key payload copy */
2790 	if (cc->key_string)
2791 		memset(cc->key, 0, cc->key_size * sizeof(u8));
2792 
2793 	return ret;
2794 }
2795 
2796 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
2797 {
2798 	struct crypt_config *cc = ti->private;
2799 	struct dm_arg_set as;
2800 	static const struct dm_arg _args[] = {
2801 		{0, 6, "Invalid number of feature args"},
2802 	};
2803 	unsigned int opt_params, val;
2804 	const char *opt_string, *sval;
2805 	char dummy;
2806 	int ret;
2807 
2808 	/* Optional parameters */
2809 	as.argc = argc;
2810 	as.argv = argv;
2811 
2812 	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
2813 	if (ret)
2814 		return ret;
2815 
2816 	while (opt_params--) {
2817 		opt_string = dm_shift_arg(&as);
2818 		if (!opt_string) {
2819 			ti->error = "Not enough feature arguments";
2820 			return -EINVAL;
2821 		}
2822 
2823 		if (!strcasecmp(opt_string, "allow_discards"))
2824 			ti->num_discard_bios = 1;
2825 
2826 		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
2827 			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
2828 
2829 		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
2830 			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
2831 		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
2832 			if (val == 0 || val > MAX_TAG_SIZE) {
2833 				ti->error = "Invalid integrity arguments";
2834 				return -EINVAL;
2835 			}
2836 			cc->on_disk_tag_size = val;
2837 			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
2838 			if (!strcasecmp(sval, "aead")) {
2839 				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
2840 			} else  if (strcasecmp(sval, "none")) {
2841 				ti->error = "Unknown integrity profile";
2842 				return -EINVAL;
2843 			}
2844 
2845 			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
2846 			if (!cc->cipher_auth)
2847 				return -ENOMEM;
2848 		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
2849 			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
2850 			    cc->sector_size > 4096 ||
2851 			    (cc->sector_size & (cc->sector_size - 1))) {
2852 				ti->error = "Invalid feature value for sector_size";
2853 				return -EINVAL;
2854 			}
2855 			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
2856 				ti->error = "Device size is not multiple of sector_size feature";
2857 				return -EINVAL;
2858 			}
2859 			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
2860 		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
2861 			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
2862 		else {
2863 			ti->error = "Invalid feature arguments";
2864 			return -EINVAL;
2865 		}
2866 	}
2867 
2868 	return 0;
2869 }
2870 
2871 /*
2872  * Construct an encryption mapping:
2873  * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
2874  */
2875 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2876 {
2877 	struct crypt_config *cc;
2878 	const char *devname = dm_table_device_name(ti->table);
2879 	int key_size;
2880 	unsigned int align_mask;
2881 	unsigned long long tmpll;
2882 	int ret;
2883 	size_t iv_size_padding, additional_req_size;
2884 	char dummy;
2885 
2886 	if (argc < 5) {
2887 		ti->error = "Not enough arguments";
2888 		return -EINVAL;
2889 	}
2890 
2891 	key_size = get_key_size(&argv[1]);
2892 	if (key_size < 0) {
2893 		ti->error = "Cannot parse key size";
2894 		return -EINVAL;
2895 	}
2896 
2897 	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
2898 	if (!cc) {
2899 		ti->error = "Cannot allocate encryption context";
2900 		return -ENOMEM;
2901 	}
2902 	cc->key_size = key_size;
2903 	cc->sector_size = (1 << SECTOR_SHIFT);
2904 	cc->sector_shift = 0;
2905 
2906 	ti->private = cc;
2907 
2908 	spin_lock(&dm_crypt_clients_lock);
2909 	dm_crypt_clients_n++;
2910 	crypt_calculate_pages_per_client();
2911 	spin_unlock(&dm_crypt_clients_lock);
2912 
2913 	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
2914 	if (ret < 0)
2915 		goto bad;
2916 
2917 	/* Optional parameters need to be read before cipher constructor */
2918 	if (argc > 5) {
2919 		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
2920 		if (ret)
2921 			goto bad;
2922 	}
2923 
2924 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
2925 	if (ret < 0)
2926 		goto bad;
2927 
2928 	if (crypt_integrity_aead(cc)) {
2929 		cc->dmreq_start = sizeof(struct aead_request);
2930 		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
2931 		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
2932 	} else {
2933 		cc->dmreq_start = sizeof(struct skcipher_request);
2934 		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
2935 		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
2936 	}
2937 	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
2938 
2939 	if (align_mask < CRYPTO_MINALIGN) {
2940 		/* Allocate the padding exactly */
2941 		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
2942 				& align_mask;
2943 	} else {
2944 		/*
2945 		 * If the cipher requires greater alignment than kmalloc
2946 		 * alignment, we don't know the exact position of the
2947 		 * initialization vector. We must assume worst case.
2948 		 */
2949 		iv_size_padding = align_mask;
2950 	}
2951 
2952 	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
2953 	additional_req_size = sizeof(struct dm_crypt_request) +
2954 		iv_size_padding + cc->iv_size +
2955 		cc->iv_size +
2956 		sizeof(uint64_t) +
2957 		sizeof(unsigned int);
2958 
2959 	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
2960 	if (ret) {
2961 		ti->error = "Cannot allocate crypt request mempool";
2962 		goto bad;
2963 	}
2964 
2965 	cc->per_bio_data_size = ti->per_io_data_size =
2966 		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
2967 		      ARCH_KMALLOC_MINALIGN);
2968 
2969 	ret = mempool_init(&cc->page_pool, BIO_MAX_PAGES, crypt_page_alloc, crypt_page_free, cc);
2970 	if (ret) {
2971 		ti->error = "Cannot allocate page mempool";
2972 		goto bad;
2973 	}
2974 
2975 	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
2976 	if (ret) {
2977 		ti->error = "Cannot allocate crypt bioset";
2978 		goto bad;
2979 	}
2980 
2981 	mutex_init(&cc->bio_alloc_lock);
2982 
2983 	ret = -EINVAL;
2984 	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
2985 	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
2986 		ti->error = "Invalid iv_offset sector";
2987 		goto bad;
2988 	}
2989 	cc->iv_offset = tmpll;
2990 
2991 	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
2992 	if (ret) {
2993 		ti->error = "Device lookup failed";
2994 		goto bad;
2995 	}
2996 
2997 	ret = -EINVAL;
2998 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
2999 		ti->error = "Invalid device sector";
3000 		goto bad;
3001 	}
3002 	cc->start = tmpll;
3003 
3004 	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3005 		ret = crypt_integrity_ctr(cc, ti);
3006 		if (ret)
3007 			goto bad;
3008 
3009 		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3010 		if (!cc->tag_pool_max_sectors)
3011 			cc->tag_pool_max_sectors = 1;
3012 
3013 		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3014 			cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3015 		if (ret) {
3016 			ti->error = "Cannot allocate integrity tags mempool";
3017 			goto bad;
3018 		}
3019 
3020 		cc->tag_pool_max_sectors <<= cc->sector_shift;
3021 	}
3022 
3023 	ret = -ENOMEM;
3024 	cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3025 	if (!cc->io_queue) {
3026 		ti->error = "Couldn't create kcryptd io queue";
3027 		goto bad;
3028 	}
3029 
3030 	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3031 		cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3032 						  1, devname);
3033 	else
3034 		cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3035 						  WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3036 						  num_online_cpus(), devname);
3037 	if (!cc->crypt_queue) {
3038 		ti->error = "Couldn't create kcryptd queue";
3039 		goto bad;
3040 	}
3041 
3042 	spin_lock_init(&cc->write_thread_lock);
3043 	cc->write_tree = RB_ROOT;
3044 
3045 	cc->write_thread = kthread_create(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3046 	if (IS_ERR(cc->write_thread)) {
3047 		ret = PTR_ERR(cc->write_thread);
3048 		cc->write_thread = NULL;
3049 		ti->error = "Couldn't spawn write thread";
3050 		goto bad;
3051 	}
3052 	wake_up_process(cc->write_thread);
3053 
3054 	ti->num_flush_bios = 1;
3055 
3056 	return 0;
3057 
3058 bad:
3059 	crypt_dtr(ti);
3060 	return ret;
3061 }
3062 
3063 static int crypt_map(struct dm_target *ti, struct bio *bio)
3064 {
3065 	struct dm_crypt_io *io;
3066 	struct crypt_config *cc = ti->private;
3067 
3068 	/*
3069 	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3070 	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3071 	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3072 	 */
3073 	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3074 	    bio_op(bio) == REQ_OP_DISCARD)) {
3075 		bio_set_dev(bio, cc->dev->bdev);
3076 		if (bio_sectors(bio))
3077 			bio->bi_iter.bi_sector = cc->start +
3078 				dm_target_offset(ti, bio->bi_iter.bi_sector);
3079 		return DM_MAPIO_REMAPPED;
3080 	}
3081 
3082 	/*
3083 	 * Check if bio is too large, split as needed.
3084 	 */
3085 	if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_PAGES << PAGE_SHIFT)) &&
3086 	    (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3087 		dm_accept_partial_bio(bio, ((BIO_MAX_PAGES << PAGE_SHIFT) >> SECTOR_SHIFT));
3088 
3089 	/*
3090 	 * Ensure that bio is a multiple of internal sector encryption size
3091 	 * and is aligned to this size as defined in IO hints.
3092 	 */
3093 	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3094 		return DM_MAPIO_KILL;
3095 
3096 	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3097 		return DM_MAPIO_KILL;
3098 
3099 	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3100 	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3101 
3102 	if (cc->on_disk_tag_size) {
3103 		unsigned tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3104 
3105 		if (unlikely(tag_len > KMALLOC_MAX_SIZE) ||
3106 		    unlikely(!(io->integrity_metadata = kmalloc(tag_len,
3107 				GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN)))) {
3108 			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3109 				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3110 			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3111 			io->integrity_metadata_from_pool = true;
3112 		}
3113 	}
3114 
3115 	if (crypt_integrity_aead(cc))
3116 		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3117 	else
3118 		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3119 
3120 	if (bio_data_dir(io->base_bio) == READ) {
3121 		if (kcryptd_io_read(io, GFP_NOWAIT))
3122 			kcryptd_queue_read(io);
3123 	} else
3124 		kcryptd_queue_crypt(io);
3125 
3126 	return DM_MAPIO_SUBMITTED;
3127 }
3128 
3129 static void crypt_status(struct dm_target *ti, status_type_t type,
3130 			 unsigned status_flags, char *result, unsigned maxlen)
3131 {
3132 	struct crypt_config *cc = ti->private;
3133 	unsigned i, sz = 0;
3134 	int num_feature_args = 0;
3135 
3136 	switch (type) {
3137 	case STATUSTYPE_INFO:
3138 		result[0] = '\0';
3139 		break;
3140 
3141 	case STATUSTYPE_TABLE:
3142 		DMEMIT("%s ", cc->cipher_string);
3143 
3144 		if (cc->key_size > 0) {
3145 			if (cc->key_string)
3146 				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3147 			else
3148 				for (i = 0; i < cc->key_size; i++)
3149 					DMEMIT("%02x", cc->key[i]);
3150 		} else
3151 			DMEMIT("-");
3152 
3153 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3154 				cc->dev->name, (unsigned long long)cc->start);
3155 
3156 		num_feature_args += !!ti->num_discard_bios;
3157 		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3158 		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3159 		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3160 		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3161 		if (cc->on_disk_tag_size)
3162 			num_feature_args++;
3163 		if (num_feature_args) {
3164 			DMEMIT(" %d", num_feature_args);
3165 			if (ti->num_discard_bios)
3166 				DMEMIT(" allow_discards");
3167 			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3168 				DMEMIT(" same_cpu_crypt");
3169 			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3170 				DMEMIT(" submit_from_crypt_cpus");
3171 			if (cc->on_disk_tag_size)
3172 				DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3173 			if (cc->sector_size != (1 << SECTOR_SHIFT))
3174 				DMEMIT(" sector_size:%d", cc->sector_size);
3175 			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3176 				DMEMIT(" iv_large_sectors");
3177 		}
3178 
3179 		break;
3180 	}
3181 }
3182 
3183 static void crypt_postsuspend(struct dm_target *ti)
3184 {
3185 	struct crypt_config *cc = ti->private;
3186 
3187 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3188 }
3189 
3190 static int crypt_preresume(struct dm_target *ti)
3191 {
3192 	struct crypt_config *cc = ti->private;
3193 
3194 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3195 		DMERR("aborting resume - crypt key is not set.");
3196 		return -EAGAIN;
3197 	}
3198 
3199 	return 0;
3200 }
3201 
3202 static void crypt_resume(struct dm_target *ti)
3203 {
3204 	struct crypt_config *cc = ti->private;
3205 
3206 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3207 }
3208 
3209 /* Message interface
3210  *	key set <key>
3211  *	key wipe
3212  */
3213 static int crypt_message(struct dm_target *ti, unsigned argc, char **argv,
3214 			 char *result, unsigned maxlen)
3215 {
3216 	struct crypt_config *cc = ti->private;
3217 	int key_size, ret = -EINVAL;
3218 
3219 	if (argc < 2)
3220 		goto error;
3221 
3222 	if (!strcasecmp(argv[0], "key")) {
3223 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3224 			DMWARN("not suspended during key manipulation.");
3225 			return -EINVAL;
3226 		}
3227 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3228 			/* The key size may not be changed. */
3229 			key_size = get_key_size(&argv[2]);
3230 			if (key_size < 0 || cc->key_size != key_size) {
3231 				memset(argv[2], '0', strlen(argv[2]));
3232 				return -EINVAL;
3233 			}
3234 
3235 			ret = crypt_set_key(cc, argv[2]);
3236 			if (ret)
3237 				return ret;
3238 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3239 				ret = cc->iv_gen_ops->init(cc);
3240 			/* wipe the kernel key payload copy */
3241 			if (cc->key_string)
3242 				memset(cc->key, 0, cc->key_size * sizeof(u8));
3243 			return ret;
3244 		}
3245 		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3246 			return crypt_wipe_key(cc);
3247 	}
3248 
3249 error:
3250 	DMWARN("unrecognised message received.");
3251 	return -EINVAL;
3252 }
3253 
3254 static int crypt_iterate_devices(struct dm_target *ti,
3255 				 iterate_devices_callout_fn fn, void *data)
3256 {
3257 	struct crypt_config *cc = ti->private;
3258 
3259 	return fn(ti, cc->dev, cc->start, ti->len, data);
3260 }
3261 
3262 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3263 {
3264 	struct crypt_config *cc = ti->private;
3265 
3266 	/*
3267 	 * Unfortunate constraint that is required to avoid the potential
3268 	 * for exceeding underlying device's max_segments limits -- due to
3269 	 * crypt_alloc_buffer() possibly allocating pages for the encryption
3270 	 * bio that are not as physically contiguous as the original bio.
3271 	 */
3272 	limits->max_segment_size = PAGE_SIZE;
3273 
3274 	limits->logical_block_size =
3275 		max_t(unsigned short, limits->logical_block_size, cc->sector_size);
3276 	limits->physical_block_size =
3277 		max_t(unsigned, limits->physical_block_size, cc->sector_size);
3278 	limits->io_min = max_t(unsigned, limits->io_min, cc->sector_size);
3279 }
3280 
3281 static struct target_type crypt_target = {
3282 	.name   = "crypt",
3283 	.version = {1, 20, 0},
3284 	.module = THIS_MODULE,
3285 	.ctr    = crypt_ctr,
3286 	.dtr    = crypt_dtr,
3287 	.map    = crypt_map,
3288 	.status = crypt_status,
3289 	.postsuspend = crypt_postsuspend,
3290 	.preresume = crypt_preresume,
3291 	.resume = crypt_resume,
3292 	.message = crypt_message,
3293 	.iterate_devices = crypt_iterate_devices,
3294 	.io_hints = crypt_io_hints,
3295 };
3296 
3297 static int __init dm_crypt_init(void)
3298 {
3299 	int r;
3300 
3301 	r = dm_register_target(&crypt_target);
3302 	if (r < 0)
3303 		DMERR("register failed %d", r);
3304 
3305 	return r;
3306 }
3307 
3308 static void __exit dm_crypt_exit(void)
3309 {
3310 	dm_unregister_target(&crypt_target);
3311 }
3312 
3313 module_init(dm_crypt_init);
3314 module_exit(dm_crypt_exit);
3315 
3316 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3317 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3318 MODULE_LICENSE("GPL");
3319