xref: /openbmc/linux/drivers/md/dm-crypt.c (revision 3dc4b6fb)
1 /*
2  * Copyright (C) 2003 Jana Saout <jana@saout.de>
3  * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
4  * Copyright (C) 2006-2017 Red Hat, Inc. All rights reserved.
5  * Copyright (C) 2013-2017 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 /*
119  * Crypt: maps a linear range of a block device
120  * and encrypts / decrypts at the same time.
121  */
122 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
123 	     DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD };
124 
125 enum cipher_flags {
126 	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cihper */
127 	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
128 };
129 
130 /*
131  * The fields in here must be read only after initialization.
132  */
133 struct crypt_config {
134 	struct dm_dev *dev;
135 	sector_t start;
136 
137 	struct percpu_counter n_allocated_pages;
138 
139 	struct workqueue_struct *io_queue;
140 	struct workqueue_struct *crypt_queue;
141 
142 	spinlock_t write_thread_lock;
143 	struct task_struct *write_thread;
144 	struct rb_root write_tree;
145 
146 	char *cipher_string;
147 	char *cipher_auth;
148 	char *key_string;
149 
150 	const struct crypt_iv_operations *iv_gen_ops;
151 	union {
152 		struct iv_benbi_private benbi;
153 		struct iv_lmk_private lmk;
154 		struct iv_tcw_private tcw;
155 	} iv_gen_private;
156 	u64 iv_offset;
157 	unsigned int iv_size;
158 	unsigned short int sector_size;
159 	unsigned char sector_shift;
160 
161 	union {
162 		struct crypto_skcipher **tfms;
163 		struct crypto_aead **tfms_aead;
164 	} cipher_tfm;
165 	unsigned tfms_count;
166 	unsigned long cipher_flags;
167 
168 	/*
169 	 * Layout of each crypto request:
170 	 *
171 	 *   struct skcipher_request
172 	 *      context
173 	 *      padding
174 	 *   struct dm_crypt_request
175 	 *      padding
176 	 *   IV
177 	 *
178 	 * The padding is added so that dm_crypt_request and the IV are
179 	 * correctly aligned.
180 	 */
181 	unsigned int dmreq_start;
182 
183 	unsigned int per_bio_data_size;
184 
185 	unsigned long flags;
186 	unsigned int key_size;
187 	unsigned int key_parts;      /* independent parts in key buffer */
188 	unsigned int key_extra_size; /* additional keys length */
189 	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
190 
191 	unsigned int integrity_tag_size;
192 	unsigned int integrity_iv_size;
193 	unsigned int on_disk_tag_size;
194 
195 	/*
196 	 * pool for per bio private data, crypto requests,
197 	 * encryption requeusts/buffer pages and integrity tags
198 	 */
199 	unsigned tag_pool_max_sectors;
200 	mempool_t tag_pool;
201 	mempool_t req_pool;
202 	mempool_t page_pool;
203 
204 	struct bio_set bs;
205 	struct mutex bio_alloc_lock;
206 
207 	u8 *authenc_key; /* space for keys in authenc() format (if used) */
208 	u8 key[0];
209 };
210 
211 #define MIN_IOS		64
212 #define MAX_TAG_SIZE	480
213 #define POOL_ENTRY_SIZE	512
214 
215 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
216 static unsigned dm_crypt_clients_n = 0;
217 static volatile unsigned long dm_crypt_pages_per_client;
218 #define DM_CRYPT_MEMORY_PERCENT			2
219 #define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_PAGES * 16)
220 
221 static void clone_init(struct dm_crypt_io *, struct bio *);
222 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
223 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
224 					     struct scatterlist *sg);
225 
226 /*
227  * Use this to access cipher attributes that are independent of the key.
228  */
229 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
230 {
231 	return cc->cipher_tfm.tfms[0];
232 }
233 
234 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
235 {
236 	return cc->cipher_tfm.tfms_aead[0];
237 }
238 
239 /*
240  * Different IV generation algorithms:
241  *
242  * plain: the initial vector is the 32-bit little-endian version of the sector
243  *        number, padded with zeros if necessary.
244  *
245  * plain64: the initial vector is the 64-bit little-endian version of the sector
246  *        number, padded with zeros if necessary.
247  *
248  * plain64be: the initial vector is the 64-bit big-endian version of the sector
249  *        number, padded with zeros if necessary.
250  *
251  * essiv: "encrypted sector|salt initial vector", the sector number is
252  *        encrypted with the bulk cipher using a salt as key. The salt
253  *        should be derived from the bulk cipher's key via hashing.
254  *
255  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
256  *        (needed for LRW-32-AES and possible other narrow block modes)
257  *
258  * null: the initial vector is always zero.  Provides compatibility with
259  *       obsolete loop_fish2 devices.  Do not use for new devices.
260  *
261  * lmk:  Compatible implementation of the block chaining mode used
262  *       by the Loop-AES block device encryption system
263  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
264  *       It operates on full 512 byte sectors and uses CBC
265  *       with an IV derived from the sector number, the data and
266  *       optionally extra IV seed.
267  *       This means that after decryption the first block
268  *       of sector must be tweaked according to decrypted data.
269  *       Loop-AES can use three encryption schemes:
270  *         version 1: is plain aes-cbc mode
271  *         version 2: uses 64 multikey scheme with lmk IV generator
272  *         version 3: the same as version 2 with additional IV seed
273  *                   (it uses 65 keys, last key is used as IV seed)
274  *
275  * tcw:  Compatible implementation of the block chaining mode used
276  *       by the TrueCrypt device encryption system (prior to version 4.1).
277  *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
278  *       It operates on full 512 byte sectors and uses CBC
279  *       with an IV derived from initial key and the sector number.
280  *       In addition, whitening value is applied on every sector, whitening
281  *       is calculated from initial key, sector number and mixed using CRC32.
282  *       Note that this encryption scheme is vulnerable to watermarking attacks
283  *       and should be used for old compatible containers access only.
284  *
285  * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
286  *        The IV is encrypted little-endian byte-offset (with the same key
287  *        and cipher as the volume).
288  */
289 
290 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
291 			      struct dm_crypt_request *dmreq)
292 {
293 	memset(iv, 0, cc->iv_size);
294 	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
295 
296 	return 0;
297 }
298 
299 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
300 				struct dm_crypt_request *dmreq)
301 {
302 	memset(iv, 0, cc->iv_size);
303 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
304 
305 	return 0;
306 }
307 
308 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
309 				  struct dm_crypt_request *dmreq)
310 {
311 	memset(iv, 0, cc->iv_size);
312 	/* iv_size is at least of size u64; usually it is 16 bytes */
313 	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
314 
315 	return 0;
316 }
317 
318 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
319 			      struct dm_crypt_request *dmreq)
320 {
321 	/*
322 	 * ESSIV encryption of the IV is now handled by the crypto API,
323 	 * so just pass the plain sector number here.
324 	 */
325 	memset(iv, 0, cc->iv_size);
326 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
327 
328 	return 0;
329 }
330 
331 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
332 			      const char *opts)
333 {
334 	unsigned bs = crypto_skcipher_blocksize(any_tfm(cc));
335 	int log = ilog2(bs);
336 
337 	/* we need to calculate how far we must shift the sector count
338 	 * to get the cipher block count, we use this shift in _gen */
339 
340 	if (1 << log != bs) {
341 		ti->error = "cypher blocksize is not a power of 2";
342 		return -EINVAL;
343 	}
344 
345 	if (log > 9) {
346 		ti->error = "cypher blocksize is > 512";
347 		return -EINVAL;
348 	}
349 
350 	cc->iv_gen_private.benbi.shift = 9 - log;
351 
352 	return 0;
353 }
354 
355 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
356 {
357 }
358 
359 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
360 			      struct dm_crypt_request *dmreq)
361 {
362 	__be64 val;
363 
364 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
365 
366 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
367 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
368 
369 	return 0;
370 }
371 
372 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
373 			     struct dm_crypt_request *dmreq)
374 {
375 	memset(iv, 0, cc->iv_size);
376 
377 	return 0;
378 }
379 
380 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
381 {
382 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
383 
384 	if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
385 		crypto_free_shash(lmk->hash_tfm);
386 	lmk->hash_tfm = NULL;
387 
388 	kzfree(lmk->seed);
389 	lmk->seed = NULL;
390 }
391 
392 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
393 			    const char *opts)
394 {
395 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
396 
397 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
398 		ti->error = "Unsupported sector size for LMK";
399 		return -EINVAL;
400 	}
401 
402 	lmk->hash_tfm = crypto_alloc_shash("md5", 0, 0);
403 	if (IS_ERR(lmk->hash_tfm)) {
404 		ti->error = "Error initializing LMK hash";
405 		return PTR_ERR(lmk->hash_tfm);
406 	}
407 
408 	/* No seed in LMK version 2 */
409 	if (cc->key_parts == cc->tfms_count) {
410 		lmk->seed = NULL;
411 		return 0;
412 	}
413 
414 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
415 	if (!lmk->seed) {
416 		crypt_iv_lmk_dtr(cc);
417 		ti->error = "Error kmallocing seed storage in LMK";
418 		return -ENOMEM;
419 	}
420 
421 	return 0;
422 }
423 
424 static int crypt_iv_lmk_init(struct crypt_config *cc)
425 {
426 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
427 	int subkey_size = cc->key_size / cc->key_parts;
428 
429 	/* LMK seed is on the position of LMK_KEYS + 1 key */
430 	if (lmk->seed)
431 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
432 		       crypto_shash_digestsize(lmk->hash_tfm));
433 
434 	return 0;
435 }
436 
437 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
438 {
439 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
440 
441 	if (lmk->seed)
442 		memset(lmk->seed, 0, LMK_SEED_SIZE);
443 
444 	return 0;
445 }
446 
447 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
448 			    struct dm_crypt_request *dmreq,
449 			    u8 *data)
450 {
451 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
452 	SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
453 	struct md5_state md5state;
454 	__le32 buf[4];
455 	int i, r;
456 
457 	desc->tfm = lmk->hash_tfm;
458 
459 	r = crypto_shash_init(desc);
460 	if (r)
461 		return r;
462 
463 	if (lmk->seed) {
464 		r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
465 		if (r)
466 			return r;
467 	}
468 
469 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
470 	r = crypto_shash_update(desc, data + 16, 16 * 31);
471 	if (r)
472 		return r;
473 
474 	/* Sector is cropped to 56 bits here */
475 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
476 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
477 	buf[2] = cpu_to_le32(4024);
478 	buf[3] = 0;
479 	r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
480 	if (r)
481 		return r;
482 
483 	/* No MD5 padding here */
484 	r = crypto_shash_export(desc, &md5state);
485 	if (r)
486 		return r;
487 
488 	for (i = 0; i < MD5_HASH_WORDS; i++)
489 		__cpu_to_le32s(&md5state.hash[i]);
490 	memcpy(iv, &md5state.hash, cc->iv_size);
491 
492 	return 0;
493 }
494 
495 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
496 			    struct dm_crypt_request *dmreq)
497 {
498 	struct scatterlist *sg;
499 	u8 *src;
500 	int r = 0;
501 
502 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
503 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
504 		src = kmap_atomic(sg_page(sg));
505 		r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
506 		kunmap_atomic(src);
507 	} else
508 		memset(iv, 0, cc->iv_size);
509 
510 	return r;
511 }
512 
513 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
514 			     struct dm_crypt_request *dmreq)
515 {
516 	struct scatterlist *sg;
517 	u8 *dst;
518 	int r;
519 
520 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
521 		return 0;
522 
523 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
524 	dst = kmap_atomic(sg_page(sg));
525 	r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
526 
527 	/* Tweak the first block of plaintext sector */
528 	if (!r)
529 		crypto_xor(dst + sg->offset, iv, cc->iv_size);
530 
531 	kunmap_atomic(dst);
532 	return r;
533 }
534 
535 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
536 {
537 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
538 
539 	kzfree(tcw->iv_seed);
540 	tcw->iv_seed = NULL;
541 	kzfree(tcw->whitening);
542 	tcw->whitening = NULL;
543 
544 	if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
545 		crypto_free_shash(tcw->crc32_tfm);
546 	tcw->crc32_tfm = NULL;
547 }
548 
549 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
550 			    const char *opts)
551 {
552 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
553 
554 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
555 		ti->error = "Unsupported sector size for TCW";
556 		return -EINVAL;
557 	}
558 
559 	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
560 		ti->error = "Wrong key size for TCW";
561 		return -EINVAL;
562 	}
563 
564 	tcw->crc32_tfm = crypto_alloc_shash("crc32", 0, 0);
565 	if (IS_ERR(tcw->crc32_tfm)) {
566 		ti->error = "Error initializing CRC32 in TCW";
567 		return PTR_ERR(tcw->crc32_tfm);
568 	}
569 
570 	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
571 	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
572 	if (!tcw->iv_seed || !tcw->whitening) {
573 		crypt_iv_tcw_dtr(cc);
574 		ti->error = "Error allocating seed storage in TCW";
575 		return -ENOMEM;
576 	}
577 
578 	return 0;
579 }
580 
581 static int crypt_iv_tcw_init(struct crypt_config *cc)
582 {
583 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
584 	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
585 
586 	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
587 	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
588 	       TCW_WHITENING_SIZE);
589 
590 	return 0;
591 }
592 
593 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
594 {
595 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
596 
597 	memset(tcw->iv_seed, 0, cc->iv_size);
598 	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
599 
600 	return 0;
601 }
602 
603 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
604 				  struct dm_crypt_request *dmreq,
605 				  u8 *data)
606 {
607 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
608 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
609 	u8 buf[TCW_WHITENING_SIZE];
610 	SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
611 	int i, r;
612 
613 	/* xor whitening with sector number */
614 	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
615 	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
616 
617 	/* calculate crc32 for every 32bit part and xor it */
618 	desc->tfm = tcw->crc32_tfm;
619 	for (i = 0; i < 4; i++) {
620 		r = crypto_shash_init(desc);
621 		if (r)
622 			goto out;
623 		r = crypto_shash_update(desc, &buf[i * 4], 4);
624 		if (r)
625 			goto out;
626 		r = crypto_shash_final(desc, &buf[i * 4]);
627 		if (r)
628 			goto out;
629 	}
630 	crypto_xor(&buf[0], &buf[12], 4);
631 	crypto_xor(&buf[4], &buf[8], 4);
632 
633 	/* apply whitening (8 bytes) to whole sector */
634 	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
635 		crypto_xor(data + i * 8, buf, 8);
636 out:
637 	memzero_explicit(buf, sizeof(buf));
638 	return r;
639 }
640 
641 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
642 			    struct dm_crypt_request *dmreq)
643 {
644 	struct scatterlist *sg;
645 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
646 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
647 	u8 *src;
648 	int r = 0;
649 
650 	/* Remove whitening from ciphertext */
651 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
652 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
653 		src = kmap_atomic(sg_page(sg));
654 		r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
655 		kunmap_atomic(src);
656 	}
657 
658 	/* Calculate IV */
659 	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
660 	if (cc->iv_size > 8)
661 		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
662 			       cc->iv_size - 8);
663 
664 	return r;
665 }
666 
667 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
668 			     struct dm_crypt_request *dmreq)
669 {
670 	struct scatterlist *sg;
671 	u8 *dst;
672 	int r;
673 
674 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
675 		return 0;
676 
677 	/* Apply whitening on ciphertext */
678 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
679 	dst = kmap_atomic(sg_page(sg));
680 	r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
681 	kunmap_atomic(dst);
682 
683 	return r;
684 }
685 
686 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
687 				struct dm_crypt_request *dmreq)
688 {
689 	/* Used only for writes, there must be an additional space to store IV */
690 	get_random_bytes(iv, cc->iv_size);
691 	return 0;
692 }
693 
694 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
695 			    const char *opts)
696 {
697 	if (test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags)) {
698 		ti->error = "AEAD transforms not supported for EBOIV";
699 		return -EINVAL;
700 	}
701 
702 	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
703 		ti->error = "Block size of EBOIV cipher does "
704 			    "not match IV size of block cipher";
705 		return -EINVAL;
706 	}
707 
708 	return 0;
709 }
710 
711 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
712 			    struct dm_crypt_request *dmreq)
713 {
714 	u8 buf[MAX_CIPHER_BLOCKSIZE] __aligned(__alignof__(__le64));
715 	struct skcipher_request *req;
716 	struct scatterlist src, dst;
717 	struct crypto_wait wait;
718 	int err;
719 
720 	req = skcipher_request_alloc(any_tfm(cc), GFP_KERNEL | GFP_NOFS);
721 	if (!req)
722 		return -ENOMEM;
723 
724 	memset(buf, 0, cc->iv_size);
725 	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
726 
727 	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
728 	sg_init_one(&dst, iv, cc->iv_size);
729 	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
730 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
731 	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
732 	skcipher_request_free(req);
733 
734 	return err;
735 }
736 
737 static const struct crypt_iv_operations crypt_iv_plain_ops = {
738 	.generator = crypt_iv_plain_gen
739 };
740 
741 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
742 	.generator = crypt_iv_plain64_gen
743 };
744 
745 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
746 	.generator = crypt_iv_plain64be_gen
747 };
748 
749 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
750 	.generator = crypt_iv_essiv_gen
751 };
752 
753 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
754 	.ctr	   = crypt_iv_benbi_ctr,
755 	.dtr	   = crypt_iv_benbi_dtr,
756 	.generator = crypt_iv_benbi_gen
757 };
758 
759 static const struct crypt_iv_operations crypt_iv_null_ops = {
760 	.generator = crypt_iv_null_gen
761 };
762 
763 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
764 	.ctr	   = crypt_iv_lmk_ctr,
765 	.dtr	   = crypt_iv_lmk_dtr,
766 	.init	   = crypt_iv_lmk_init,
767 	.wipe	   = crypt_iv_lmk_wipe,
768 	.generator = crypt_iv_lmk_gen,
769 	.post	   = crypt_iv_lmk_post
770 };
771 
772 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
773 	.ctr	   = crypt_iv_tcw_ctr,
774 	.dtr	   = crypt_iv_tcw_dtr,
775 	.init	   = crypt_iv_tcw_init,
776 	.wipe	   = crypt_iv_tcw_wipe,
777 	.generator = crypt_iv_tcw_gen,
778 	.post	   = crypt_iv_tcw_post
779 };
780 
781 static struct crypt_iv_operations crypt_iv_random_ops = {
782 	.generator = crypt_iv_random_gen
783 };
784 
785 static struct crypt_iv_operations crypt_iv_eboiv_ops = {
786 	.ctr	   = crypt_iv_eboiv_ctr,
787 	.generator = crypt_iv_eboiv_gen
788 };
789 
790 /*
791  * Integrity extensions
792  */
793 static bool crypt_integrity_aead(struct crypt_config *cc)
794 {
795 	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
796 }
797 
798 static bool crypt_integrity_hmac(struct crypt_config *cc)
799 {
800 	return crypt_integrity_aead(cc) && cc->key_mac_size;
801 }
802 
803 /* Get sg containing data */
804 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
805 					     struct scatterlist *sg)
806 {
807 	if (unlikely(crypt_integrity_aead(cc)))
808 		return &sg[2];
809 
810 	return sg;
811 }
812 
813 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
814 {
815 	struct bio_integrity_payload *bip;
816 	unsigned int tag_len;
817 	int ret;
818 
819 	if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
820 		return 0;
821 
822 	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
823 	if (IS_ERR(bip))
824 		return PTR_ERR(bip);
825 
826 	tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
827 
828 	bip->bip_iter.bi_size = tag_len;
829 	bip->bip_iter.bi_sector = io->cc->start + io->sector;
830 
831 	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
832 				     tag_len, offset_in_page(io->integrity_metadata));
833 	if (unlikely(ret != tag_len))
834 		return -ENOMEM;
835 
836 	return 0;
837 }
838 
839 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
840 {
841 #ifdef CONFIG_BLK_DEV_INTEGRITY
842 	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
843 	struct mapped_device *md = dm_table_get_md(ti->table);
844 
845 	/* From now we require underlying device with our integrity profile */
846 	if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
847 		ti->error = "Integrity profile not supported.";
848 		return -EINVAL;
849 	}
850 
851 	if (bi->tag_size != cc->on_disk_tag_size ||
852 	    bi->tuple_size != cc->on_disk_tag_size) {
853 		ti->error = "Integrity profile tag size mismatch.";
854 		return -EINVAL;
855 	}
856 	if (1 << bi->interval_exp != cc->sector_size) {
857 		ti->error = "Integrity profile sector size mismatch.";
858 		return -EINVAL;
859 	}
860 
861 	if (crypt_integrity_aead(cc)) {
862 		cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
863 		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
864 		       cc->integrity_tag_size, cc->integrity_iv_size);
865 
866 		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
867 			ti->error = "Integrity AEAD auth tag size is not supported.";
868 			return -EINVAL;
869 		}
870 	} else if (cc->integrity_iv_size)
871 		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
872 		       cc->integrity_iv_size);
873 
874 	if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
875 		ti->error = "Not enough space for integrity tag in the profile.";
876 		return -EINVAL;
877 	}
878 
879 	return 0;
880 #else
881 	ti->error = "Integrity profile not supported.";
882 	return -EINVAL;
883 #endif
884 }
885 
886 static void crypt_convert_init(struct crypt_config *cc,
887 			       struct convert_context *ctx,
888 			       struct bio *bio_out, struct bio *bio_in,
889 			       sector_t sector)
890 {
891 	ctx->bio_in = bio_in;
892 	ctx->bio_out = bio_out;
893 	if (bio_in)
894 		ctx->iter_in = bio_in->bi_iter;
895 	if (bio_out)
896 		ctx->iter_out = bio_out->bi_iter;
897 	ctx->cc_sector = sector + cc->iv_offset;
898 	init_completion(&ctx->restart);
899 }
900 
901 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
902 					     void *req)
903 {
904 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
905 }
906 
907 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
908 {
909 	return (void *)((char *)dmreq - cc->dmreq_start);
910 }
911 
912 static u8 *iv_of_dmreq(struct crypt_config *cc,
913 		       struct dm_crypt_request *dmreq)
914 {
915 	if (crypt_integrity_aead(cc))
916 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
917 			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
918 	else
919 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
920 			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
921 }
922 
923 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
924 		       struct dm_crypt_request *dmreq)
925 {
926 	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
927 }
928 
929 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
930 		       struct dm_crypt_request *dmreq)
931 {
932 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
933 	return (__le64 *) ptr;
934 }
935 
936 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
937 		       struct dm_crypt_request *dmreq)
938 {
939 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
940 		  cc->iv_size + sizeof(uint64_t);
941 	return (unsigned int*)ptr;
942 }
943 
944 static void *tag_from_dmreq(struct crypt_config *cc,
945 				struct dm_crypt_request *dmreq)
946 {
947 	struct convert_context *ctx = dmreq->ctx;
948 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
949 
950 	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
951 		cc->on_disk_tag_size];
952 }
953 
954 static void *iv_tag_from_dmreq(struct crypt_config *cc,
955 			       struct dm_crypt_request *dmreq)
956 {
957 	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
958 }
959 
960 static int crypt_convert_block_aead(struct crypt_config *cc,
961 				     struct convert_context *ctx,
962 				     struct aead_request *req,
963 				     unsigned int tag_offset)
964 {
965 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
966 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
967 	struct dm_crypt_request *dmreq;
968 	u8 *iv, *org_iv, *tag_iv, *tag;
969 	__le64 *sector;
970 	int r = 0;
971 
972 	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
973 
974 	/* Reject unexpected unaligned bio. */
975 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
976 		return -EIO;
977 
978 	dmreq = dmreq_of_req(cc, req);
979 	dmreq->iv_sector = ctx->cc_sector;
980 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
981 		dmreq->iv_sector >>= cc->sector_shift;
982 	dmreq->ctx = ctx;
983 
984 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
985 
986 	sector = org_sector_of_dmreq(cc, dmreq);
987 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
988 
989 	iv = iv_of_dmreq(cc, dmreq);
990 	org_iv = org_iv_of_dmreq(cc, dmreq);
991 	tag = tag_from_dmreq(cc, dmreq);
992 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
993 
994 	/* AEAD request:
995 	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
996 	 *  | (authenticated) | (auth+encryption) |              |
997 	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
998 	 */
999 	sg_init_table(dmreq->sg_in, 4);
1000 	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1001 	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1002 	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1003 	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1004 
1005 	sg_init_table(dmreq->sg_out, 4);
1006 	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1007 	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1008 	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1009 	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1010 
1011 	if (cc->iv_gen_ops) {
1012 		/* For READs use IV stored in integrity metadata */
1013 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1014 			memcpy(org_iv, tag_iv, cc->iv_size);
1015 		} else {
1016 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1017 			if (r < 0)
1018 				return r;
1019 			/* Store generated IV in integrity metadata */
1020 			if (cc->integrity_iv_size)
1021 				memcpy(tag_iv, org_iv, cc->iv_size);
1022 		}
1023 		/* Working copy of IV, to be modified in crypto API */
1024 		memcpy(iv, org_iv, cc->iv_size);
1025 	}
1026 
1027 	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1028 	if (bio_data_dir(ctx->bio_in) == WRITE) {
1029 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1030 				       cc->sector_size, iv);
1031 		r = crypto_aead_encrypt(req);
1032 		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1033 			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1034 			       cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1035 	} else {
1036 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1037 				       cc->sector_size + cc->integrity_tag_size, iv);
1038 		r = crypto_aead_decrypt(req);
1039 	}
1040 
1041 	if (r == -EBADMSG) {
1042 		char b[BDEVNAME_SIZE];
1043 		DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", bio_devname(ctx->bio_in, b),
1044 			    (unsigned long long)le64_to_cpu(*sector));
1045 	}
1046 
1047 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1048 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1049 
1050 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1051 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1052 
1053 	return r;
1054 }
1055 
1056 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1057 					struct convert_context *ctx,
1058 					struct skcipher_request *req,
1059 					unsigned int tag_offset)
1060 {
1061 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1062 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1063 	struct scatterlist *sg_in, *sg_out;
1064 	struct dm_crypt_request *dmreq;
1065 	u8 *iv, *org_iv, *tag_iv;
1066 	__le64 *sector;
1067 	int r = 0;
1068 
1069 	/* Reject unexpected unaligned bio. */
1070 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1071 		return -EIO;
1072 
1073 	dmreq = dmreq_of_req(cc, req);
1074 	dmreq->iv_sector = ctx->cc_sector;
1075 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1076 		dmreq->iv_sector >>= cc->sector_shift;
1077 	dmreq->ctx = ctx;
1078 
1079 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1080 
1081 	iv = iv_of_dmreq(cc, dmreq);
1082 	org_iv = org_iv_of_dmreq(cc, dmreq);
1083 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1084 
1085 	sector = org_sector_of_dmreq(cc, dmreq);
1086 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1087 
1088 	/* For skcipher we use only the first sg item */
1089 	sg_in  = &dmreq->sg_in[0];
1090 	sg_out = &dmreq->sg_out[0];
1091 
1092 	sg_init_table(sg_in, 1);
1093 	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1094 
1095 	sg_init_table(sg_out, 1);
1096 	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1097 
1098 	if (cc->iv_gen_ops) {
1099 		/* For READs use IV stored in integrity metadata */
1100 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1101 			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1102 		} else {
1103 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1104 			if (r < 0)
1105 				return r;
1106 			/* Store generated IV in integrity metadata */
1107 			if (cc->integrity_iv_size)
1108 				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1109 		}
1110 		/* Working copy of IV, to be modified in crypto API */
1111 		memcpy(iv, org_iv, cc->iv_size);
1112 	}
1113 
1114 	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1115 
1116 	if (bio_data_dir(ctx->bio_in) == WRITE)
1117 		r = crypto_skcipher_encrypt(req);
1118 	else
1119 		r = crypto_skcipher_decrypt(req);
1120 
1121 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1122 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1123 
1124 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1125 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1126 
1127 	return r;
1128 }
1129 
1130 static void kcryptd_async_done(struct crypto_async_request *async_req,
1131 			       int error);
1132 
1133 static void crypt_alloc_req_skcipher(struct crypt_config *cc,
1134 				     struct convert_context *ctx)
1135 {
1136 	unsigned key_index = ctx->cc_sector & (cc->tfms_count - 1);
1137 
1138 	if (!ctx->r.req)
1139 		ctx->r.req = mempool_alloc(&cc->req_pool, GFP_NOIO);
1140 
1141 	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1142 
1143 	/*
1144 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1145 	 * requests if driver request queue is full.
1146 	 */
1147 	skcipher_request_set_callback(ctx->r.req,
1148 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1149 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1150 }
1151 
1152 static void crypt_alloc_req_aead(struct crypt_config *cc,
1153 				 struct convert_context *ctx)
1154 {
1155 	if (!ctx->r.req_aead)
1156 		ctx->r.req_aead = mempool_alloc(&cc->req_pool, GFP_NOIO);
1157 
1158 	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1159 
1160 	/*
1161 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1162 	 * requests if driver request queue is full.
1163 	 */
1164 	aead_request_set_callback(ctx->r.req_aead,
1165 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1166 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1167 }
1168 
1169 static void crypt_alloc_req(struct crypt_config *cc,
1170 			    struct convert_context *ctx)
1171 {
1172 	if (crypt_integrity_aead(cc))
1173 		crypt_alloc_req_aead(cc, ctx);
1174 	else
1175 		crypt_alloc_req_skcipher(cc, ctx);
1176 }
1177 
1178 static void crypt_free_req_skcipher(struct crypt_config *cc,
1179 				    struct skcipher_request *req, struct bio *base_bio)
1180 {
1181 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1182 
1183 	if ((struct skcipher_request *)(io + 1) != req)
1184 		mempool_free(req, &cc->req_pool);
1185 }
1186 
1187 static void crypt_free_req_aead(struct crypt_config *cc,
1188 				struct aead_request *req, struct bio *base_bio)
1189 {
1190 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1191 
1192 	if ((struct aead_request *)(io + 1) != req)
1193 		mempool_free(req, &cc->req_pool);
1194 }
1195 
1196 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1197 {
1198 	if (crypt_integrity_aead(cc))
1199 		crypt_free_req_aead(cc, req, base_bio);
1200 	else
1201 		crypt_free_req_skcipher(cc, req, base_bio);
1202 }
1203 
1204 /*
1205  * Encrypt / decrypt data from one bio to another one (can be the same one)
1206  */
1207 static blk_status_t crypt_convert(struct crypt_config *cc,
1208 			 struct convert_context *ctx)
1209 {
1210 	unsigned int tag_offset = 0;
1211 	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1212 	int r;
1213 
1214 	atomic_set(&ctx->cc_pending, 1);
1215 
1216 	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1217 
1218 		crypt_alloc_req(cc, ctx);
1219 		atomic_inc(&ctx->cc_pending);
1220 
1221 		if (crypt_integrity_aead(cc))
1222 			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1223 		else
1224 			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1225 
1226 		switch (r) {
1227 		/*
1228 		 * The request was queued by a crypto driver
1229 		 * but the driver request queue is full, let's wait.
1230 		 */
1231 		case -EBUSY:
1232 			wait_for_completion(&ctx->restart);
1233 			reinit_completion(&ctx->restart);
1234 			/* fall through */
1235 		/*
1236 		 * The request is queued and processed asynchronously,
1237 		 * completion function kcryptd_async_done() will be called.
1238 		 */
1239 		case -EINPROGRESS:
1240 			ctx->r.req = NULL;
1241 			ctx->cc_sector += sector_step;
1242 			tag_offset++;
1243 			continue;
1244 		/*
1245 		 * The request was already processed (synchronously).
1246 		 */
1247 		case 0:
1248 			atomic_dec(&ctx->cc_pending);
1249 			ctx->cc_sector += sector_step;
1250 			tag_offset++;
1251 			cond_resched();
1252 			continue;
1253 		/*
1254 		 * There was a data integrity error.
1255 		 */
1256 		case -EBADMSG:
1257 			atomic_dec(&ctx->cc_pending);
1258 			return BLK_STS_PROTECTION;
1259 		/*
1260 		 * There was an error while processing the request.
1261 		 */
1262 		default:
1263 			atomic_dec(&ctx->cc_pending);
1264 			return BLK_STS_IOERR;
1265 		}
1266 	}
1267 
1268 	return 0;
1269 }
1270 
1271 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1272 
1273 /*
1274  * Generate a new unfragmented bio with the given size
1275  * This should never violate the device limitations (but only because
1276  * max_segment_size is being constrained to PAGE_SIZE).
1277  *
1278  * This function may be called concurrently. If we allocate from the mempool
1279  * concurrently, there is a possibility of deadlock. For example, if we have
1280  * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1281  * the mempool concurrently, it may deadlock in a situation where both processes
1282  * have allocated 128 pages and the mempool is exhausted.
1283  *
1284  * In order to avoid this scenario we allocate the pages under a mutex.
1285  *
1286  * In order to not degrade performance with excessive locking, we try
1287  * non-blocking allocations without a mutex first but on failure we fallback
1288  * to blocking allocations with a mutex.
1289  */
1290 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size)
1291 {
1292 	struct crypt_config *cc = io->cc;
1293 	struct bio *clone;
1294 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1295 	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1296 	unsigned i, len, remaining_size;
1297 	struct page *page;
1298 
1299 retry:
1300 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1301 		mutex_lock(&cc->bio_alloc_lock);
1302 
1303 	clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, &cc->bs);
1304 	if (!clone)
1305 		goto out;
1306 
1307 	clone_init(io, clone);
1308 
1309 	remaining_size = size;
1310 
1311 	for (i = 0; i < nr_iovecs; i++) {
1312 		page = mempool_alloc(&cc->page_pool, gfp_mask);
1313 		if (!page) {
1314 			crypt_free_buffer_pages(cc, clone);
1315 			bio_put(clone);
1316 			gfp_mask |= __GFP_DIRECT_RECLAIM;
1317 			goto retry;
1318 		}
1319 
1320 		len = (remaining_size > PAGE_SIZE) ? PAGE_SIZE : remaining_size;
1321 
1322 		bio_add_page(clone, page, len, 0);
1323 
1324 		remaining_size -= len;
1325 	}
1326 
1327 	/* Allocate space for integrity tags */
1328 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1329 		crypt_free_buffer_pages(cc, clone);
1330 		bio_put(clone);
1331 		clone = NULL;
1332 	}
1333 out:
1334 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1335 		mutex_unlock(&cc->bio_alloc_lock);
1336 
1337 	return clone;
1338 }
1339 
1340 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1341 {
1342 	struct bio_vec *bv;
1343 	struct bvec_iter_all iter_all;
1344 
1345 	bio_for_each_segment_all(bv, clone, iter_all) {
1346 		BUG_ON(!bv->bv_page);
1347 		mempool_free(bv->bv_page, &cc->page_pool);
1348 	}
1349 }
1350 
1351 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1352 			  struct bio *bio, sector_t sector)
1353 {
1354 	io->cc = cc;
1355 	io->base_bio = bio;
1356 	io->sector = sector;
1357 	io->error = 0;
1358 	io->ctx.r.req = NULL;
1359 	io->integrity_metadata = NULL;
1360 	io->integrity_metadata_from_pool = false;
1361 	atomic_set(&io->io_pending, 0);
1362 }
1363 
1364 static void crypt_inc_pending(struct dm_crypt_io *io)
1365 {
1366 	atomic_inc(&io->io_pending);
1367 }
1368 
1369 /*
1370  * One of the bios was finished. Check for completion of
1371  * the whole request and correctly clean up the buffer.
1372  */
1373 static void crypt_dec_pending(struct dm_crypt_io *io)
1374 {
1375 	struct crypt_config *cc = io->cc;
1376 	struct bio *base_bio = io->base_bio;
1377 	blk_status_t error = io->error;
1378 
1379 	if (!atomic_dec_and_test(&io->io_pending))
1380 		return;
1381 
1382 	if (io->ctx.r.req)
1383 		crypt_free_req(cc, io->ctx.r.req, base_bio);
1384 
1385 	if (unlikely(io->integrity_metadata_from_pool))
1386 		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1387 	else
1388 		kfree(io->integrity_metadata);
1389 
1390 	base_bio->bi_status = error;
1391 	bio_endio(base_bio);
1392 }
1393 
1394 /*
1395  * kcryptd/kcryptd_io:
1396  *
1397  * Needed because it would be very unwise to do decryption in an
1398  * interrupt context.
1399  *
1400  * kcryptd performs the actual encryption or decryption.
1401  *
1402  * kcryptd_io performs the IO submission.
1403  *
1404  * They must be separated as otherwise the final stages could be
1405  * starved by new requests which can block in the first stages due
1406  * to memory allocation.
1407  *
1408  * The work is done per CPU global for all dm-crypt instances.
1409  * They should not depend on each other and do not block.
1410  */
1411 static void crypt_endio(struct bio *clone)
1412 {
1413 	struct dm_crypt_io *io = clone->bi_private;
1414 	struct crypt_config *cc = io->cc;
1415 	unsigned rw = bio_data_dir(clone);
1416 	blk_status_t error;
1417 
1418 	/*
1419 	 * free the processed pages
1420 	 */
1421 	if (rw == WRITE)
1422 		crypt_free_buffer_pages(cc, clone);
1423 
1424 	error = clone->bi_status;
1425 	bio_put(clone);
1426 
1427 	if (rw == READ && !error) {
1428 		kcryptd_queue_crypt(io);
1429 		return;
1430 	}
1431 
1432 	if (unlikely(error))
1433 		io->error = error;
1434 
1435 	crypt_dec_pending(io);
1436 }
1437 
1438 static void clone_init(struct dm_crypt_io *io, struct bio *clone)
1439 {
1440 	struct crypt_config *cc = io->cc;
1441 
1442 	clone->bi_private = io;
1443 	clone->bi_end_io  = crypt_endio;
1444 	bio_set_dev(clone, cc->dev->bdev);
1445 	clone->bi_opf	  = io->base_bio->bi_opf;
1446 }
1447 
1448 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1449 {
1450 	struct crypt_config *cc = io->cc;
1451 	struct bio *clone;
1452 
1453 	/*
1454 	 * We need the original biovec array in order to decrypt
1455 	 * the whole bio data *afterwards* -- thanks to immutable
1456 	 * biovecs we don't need to worry about the block layer
1457 	 * modifying the biovec array; so leverage bio_clone_fast().
1458 	 */
1459 	clone = bio_clone_fast(io->base_bio, gfp, &cc->bs);
1460 	if (!clone)
1461 		return 1;
1462 
1463 	crypt_inc_pending(io);
1464 
1465 	clone_init(io, clone);
1466 	clone->bi_iter.bi_sector = cc->start + io->sector;
1467 
1468 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1469 		crypt_dec_pending(io);
1470 		bio_put(clone);
1471 		return 1;
1472 	}
1473 
1474 	generic_make_request(clone);
1475 	return 0;
1476 }
1477 
1478 static void kcryptd_io_read_work(struct work_struct *work)
1479 {
1480 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1481 
1482 	crypt_inc_pending(io);
1483 	if (kcryptd_io_read(io, GFP_NOIO))
1484 		io->error = BLK_STS_RESOURCE;
1485 	crypt_dec_pending(io);
1486 }
1487 
1488 static void kcryptd_queue_read(struct dm_crypt_io *io)
1489 {
1490 	struct crypt_config *cc = io->cc;
1491 
1492 	INIT_WORK(&io->work, kcryptd_io_read_work);
1493 	queue_work(cc->io_queue, &io->work);
1494 }
1495 
1496 static void kcryptd_io_write(struct dm_crypt_io *io)
1497 {
1498 	struct bio *clone = io->ctx.bio_out;
1499 
1500 	generic_make_request(clone);
1501 }
1502 
1503 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1504 
1505 static int dmcrypt_write(void *data)
1506 {
1507 	struct crypt_config *cc = data;
1508 	struct dm_crypt_io *io;
1509 
1510 	while (1) {
1511 		struct rb_root write_tree;
1512 		struct blk_plug plug;
1513 
1514 		spin_lock_irq(&cc->write_thread_lock);
1515 continue_locked:
1516 
1517 		if (!RB_EMPTY_ROOT(&cc->write_tree))
1518 			goto pop_from_list;
1519 
1520 		set_current_state(TASK_INTERRUPTIBLE);
1521 
1522 		spin_unlock_irq(&cc->write_thread_lock);
1523 
1524 		if (unlikely(kthread_should_stop())) {
1525 			set_current_state(TASK_RUNNING);
1526 			break;
1527 		}
1528 
1529 		schedule();
1530 
1531 		set_current_state(TASK_RUNNING);
1532 		spin_lock_irq(&cc->write_thread_lock);
1533 		goto continue_locked;
1534 
1535 pop_from_list:
1536 		write_tree = cc->write_tree;
1537 		cc->write_tree = RB_ROOT;
1538 		spin_unlock_irq(&cc->write_thread_lock);
1539 
1540 		BUG_ON(rb_parent(write_tree.rb_node));
1541 
1542 		/*
1543 		 * Note: we cannot walk the tree here with rb_next because
1544 		 * the structures may be freed when kcryptd_io_write is called.
1545 		 */
1546 		blk_start_plug(&plug);
1547 		do {
1548 			io = crypt_io_from_node(rb_first(&write_tree));
1549 			rb_erase(&io->rb_node, &write_tree);
1550 			kcryptd_io_write(io);
1551 		} while (!RB_EMPTY_ROOT(&write_tree));
1552 		blk_finish_plug(&plug);
1553 	}
1554 	return 0;
1555 }
1556 
1557 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1558 {
1559 	struct bio *clone = io->ctx.bio_out;
1560 	struct crypt_config *cc = io->cc;
1561 	unsigned long flags;
1562 	sector_t sector;
1563 	struct rb_node **rbp, *parent;
1564 
1565 	if (unlikely(io->error)) {
1566 		crypt_free_buffer_pages(cc, clone);
1567 		bio_put(clone);
1568 		crypt_dec_pending(io);
1569 		return;
1570 	}
1571 
1572 	/* crypt_convert should have filled the clone bio */
1573 	BUG_ON(io->ctx.iter_out.bi_size);
1574 
1575 	clone->bi_iter.bi_sector = cc->start + io->sector;
1576 
1577 	if (likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) {
1578 		generic_make_request(clone);
1579 		return;
1580 	}
1581 
1582 	spin_lock_irqsave(&cc->write_thread_lock, flags);
1583 	if (RB_EMPTY_ROOT(&cc->write_tree))
1584 		wake_up_process(cc->write_thread);
1585 	rbp = &cc->write_tree.rb_node;
1586 	parent = NULL;
1587 	sector = io->sector;
1588 	while (*rbp) {
1589 		parent = *rbp;
1590 		if (sector < crypt_io_from_node(parent)->sector)
1591 			rbp = &(*rbp)->rb_left;
1592 		else
1593 			rbp = &(*rbp)->rb_right;
1594 	}
1595 	rb_link_node(&io->rb_node, parent, rbp);
1596 	rb_insert_color(&io->rb_node, &cc->write_tree);
1597 	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
1598 }
1599 
1600 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
1601 {
1602 	struct crypt_config *cc = io->cc;
1603 	struct bio *clone;
1604 	int crypt_finished;
1605 	sector_t sector = io->sector;
1606 	blk_status_t r;
1607 
1608 	/*
1609 	 * Prevent io from disappearing until this function completes.
1610 	 */
1611 	crypt_inc_pending(io);
1612 	crypt_convert_init(cc, &io->ctx, NULL, io->base_bio, sector);
1613 
1614 	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
1615 	if (unlikely(!clone)) {
1616 		io->error = BLK_STS_IOERR;
1617 		goto dec;
1618 	}
1619 
1620 	io->ctx.bio_out = clone;
1621 	io->ctx.iter_out = clone->bi_iter;
1622 
1623 	sector += bio_sectors(clone);
1624 
1625 	crypt_inc_pending(io);
1626 	r = crypt_convert(cc, &io->ctx);
1627 	if (r)
1628 		io->error = r;
1629 	crypt_finished = atomic_dec_and_test(&io->ctx.cc_pending);
1630 
1631 	/* Encryption was already finished, submit io now */
1632 	if (crypt_finished) {
1633 		kcryptd_crypt_write_io_submit(io, 0);
1634 		io->sector = sector;
1635 	}
1636 
1637 dec:
1638 	crypt_dec_pending(io);
1639 }
1640 
1641 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
1642 {
1643 	crypt_dec_pending(io);
1644 }
1645 
1646 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
1647 {
1648 	struct crypt_config *cc = io->cc;
1649 	blk_status_t r;
1650 
1651 	crypt_inc_pending(io);
1652 
1653 	crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
1654 			   io->sector);
1655 
1656 	r = crypt_convert(cc, &io->ctx);
1657 	if (r)
1658 		io->error = r;
1659 
1660 	if (atomic_dec_and_test(&io->ctx.cc_pending))
1661 		kcryptd_crypt_read_done(io);
1662 
1663 	crypt_dec_pending(io);
1664 }
1665 
1666 static void kcryptd_async_done(struct crypto_async_request *async_req,
1667 			       int error)
1668 {
1669 	struct dm_crypt_request *dmreq = async_req->data;
1670 	struct convert_context *ctx = dmreq->ctx;
1671 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1672 	struct crypt_config *cc = io->cc;
1673 
1674 	/*
1675 	 * A request from crypto driver backlog is going to be processed now,
1676 	 * finish the completion and continue in crypt_convert().
1677 	 * (Callback will be called for the second time for this request.)
1678 	 */
1679 	if (error == -EINPROGRESS) {
1680 		complete(&ctx->restart);
1681 		return;
1682 	}
1683 
1684 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
1685 		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
1686 
1687 	if (error == -EBADMSG) {
1688 		char b[BDEVNAME_SIZE];
1689 		DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", bio_devname(ctx->bio_in, b),
1690 			    (unsigned long long)le64_to_cpu(*org_sector_of_dmreq(cc, dmreq)));
1691 		io->error = BLK_STS_PROTECTION;
1692 	} else if (error < 0)
1693 		io->error = BLK_STS_IOERR;
1694 
1695 	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
1696 
1697 	if (!atomic_dec_and_test(&ctx->cc_pending))
1698 		return;
1699 
1700 	if (bio_data_dir(io->base_bio) == READ)
1701 		kcryptd_crypt_read_done(io);
1702 	else
1703 		kcryptd_crypt_write_io_submit(io, 1);
1704 }
1705 
1706 static void kcryptd_crypt(struct work_struct *work)
1707 {
1708 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1709 
1710 	if (bio_data_dir(io->base_bio) == READ)
1711 		kcryptd_crypt_read_convert(io);
1712 	else
1713 		kcryptd_crypt_write_convert(io);
1714 }
1715 
1716 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
1717 {
1718 	struct crypt_config *cc = io->cc;
1719 
1720 	INIT_WORK(&io->work, kcryptd_crypt);
1721 	queue_work(cc->crypt_queue, &io->work);
1722 }
1723 
1724 static void crypt_free_tfms_aead(struct crypt_config *cc)
1725 {
1726 	if (!cc->cipher_tfm.tfms_aead)
1727 		return;
1728 
1729 	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
1730 		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
1731 		cc->cipher_tfm.tfms_aead[0] = NULL;
1732 	}
1733 
1734 	kfree(cc->cipher_tfm.tfms_aead);
1735 	cc->cipher_tfm.tfms_aead = NULL;
1736 }
1737 
1738 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
1739 {
1740 	unsigned i;
1741 
1742 	if (!cc->cipher_tfm.tfms)
1743 		return;
1744 
1745 	for (i = 0; i < cc->tfms_count; i++)
1746 		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
1747 			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
1748 			cc->cipher_tfm.tfms[i] = NULL;
1749 		}
1750 
1751 	kfree(cc->cipher_tfm.tfms);
1752 	cc->cipher_tfm.tfms = NULL;
1753 }
1754 
1755 static void crypt_free_tfms(struct crypt_config *cc)
1756 {
1757 	if (crypt_integrity_aead(cc))
1758 		crypt_free_tfms_aead(cc);
1759 	else
1760 		crypt_free_tfms_skcipher(cc);
1761 }
1762 
1763 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
1764 {
1765 	unsigned i;
1766 	int err;
1767 
1768 	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
1769 				      sizeof(struct crypto_skcipher *),
1770 				      GFP_KERNEL);
1771 	if (!cc->cipher_tfm.tfms)
1772 		return -ENOMEM;
1773 
1774 	for (i = 0; i < cc->tfms_count; i++) {
1775 		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0, 0);
1776 		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
1777 			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
1778 			crypt_free_tfms(cc);
1779 			return err;
1780 		}
1781 	}
1782 
1783 	/*
1784 	 * dm-crypt performance can vary greatly depending on which crypto
1785 	 * algorithm implementation is used.  Help people debug performance
1786 	 * problems by logging the ->cra_driver_name.
1787 	 */
1788 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
1789 	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
1790 	return 0;
1791 }
1792 
1793 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
1794 {
1795 	int err;
1796 
1797 	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
1798 	if (!cc->cipher_tfm.tfms)
1799 		return -ENOMEM;
1800 
1801 	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0, 0);
1802 	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
1803 		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
1804 		crypt_free_tfms(cc);
1805 		return err;
1806 	}
1807 
1808 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
1809 	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
1810 	return 0;
1811 }
1812 
1813 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
1814 {
1815 	if (crypt_integrity_aead(cc))
1816 		return crypt_alloc_tfms_aead(cc, ciphermode);
1817 	else
1818 		return crypt_alloc_tfms_skcipher(cc, ciphermode);
1819 }
1820 
1821 static unsigned crypt_subkey_size(struct crypt_config *cc)
1822 {
1823 	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
1824 }
1825 
1826 static unsigned crypt_authenckey_size(struct crypt_config *cc)
1827 {
1828 	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
1829 }
1830 
1831 /*
1832  * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
1833  * the key must be for some reason in special format.
1834  * This funcion converts cc->key to this special format.
1835  */
1836 static void crypt_copy_authenckey(char *p, const void *key,
1837 				  unsigned enckeylen, unsigned authkeylen)
1838 {
1839 	struct crypto_authenc_key_param *param;
1840 	struct rtattr *rta;
1841 
1842 	rta = (struct rtattr *)p;
1843 	param = RTA_DATA(rta);
1844 	param->enckeylen = cpu_to_be32(enckeylen);
1845 	rta->rta_len = RTA_LENGTH(sizeof(*param));
1846 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1847 	p += RTA_SPACE(sizeof(*param));
1848 	memcpy(p, key + enckeylen, authkeylen);
1849 	p += authkeylen;
1850 	memcpy(p, key, enckeylen);
1851 }
1852 
1853 static int crypt_setkey(struct crypt_config *cc)
1854 {
1855 	unsigned subkey_size;
1856 	int err = 0, i, r;
1857 
1858 	/* Ignore extra keys (which are used for IV etc) */
1859 	subkey_size = crypt_subkey_size(cc);
1860 
1861 	if (crypt_integrity_hmac(cc)) {
1862 		if (subkey_size < cc->key_mac_size)
1863 			return -EINVAL;
1864 
1865 		crypt_copy_authenckey(cc->authenc_key, cc->key,
1866 				      subkey_size - cc->key_mac_size,
1867 				      cc->key_mac_size);
1868 	}
1869 
1870 	for (i = 0; i < cc->tfms_count; i++) {
1871 		if (crypt_integrity_hmac(cc))
1872 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
1873 				cc->authenc_key, crypt_authenckey_size(cc));
1874 		else if (crypt_integrity_aead(cc))
1875 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
1876 					       cc->key + (i * subkey_size),
1877 					       subkey_size);
1878 		else
1879 			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
1880 						   cc->key + (i * subkey_size),
1881 						   subkey_size);
1882 		if (r)
1883 			err = r;
1884 	}
1885 
1886 	if (crypt_integrity_hmac(cc))
1887 		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
1888 
1889 	return err;
1890 }
1891 
1892 #ifdef CONFIG_KEYS
1893 
1894 static bool contains_whitespace(const char *str)
1895 {
1896 	while (*str)
1897 		if (isspace(*str++))
1898 			return true;
1899 	return false;
1900 }
1901 
1902 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
1903 {
1904 	char *new_key_string, *key_desc;
1905 	int ret;
1906 	struct key *key;
1907 	const struct user_key_payload *ukp;
1908 
1909 	/*
1910 	 * Reject key_string with whitespace. dm core currently lacks code for
1911 	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
1912 	 */
1913 	if (contains_whitespace(key_string)) {
1914 		DMERR("whitespace chars not allowed in key string");
1915 		return -EINVAL;
1916 	}
1917 
1918 	/* look for next ':' separating key_type from key_description */
1919 	key_desc = strpbrk(key_string, ":");
1920 	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
1921 		return -EINVAL;
1922 
1923 	if (strncmp(key_string, "logon:", key_desc - key_string + 1) &&
1924 	    strncmp(key_string, "user:", key_desc - key_string + 1))
1925 		return -EINVAL;
1926 
1927 	new_key_string = kstrdup(key_string, GFP_KERNEL);
1928 	if (!new_key_string)
1929 		return -ENOMEM;
1930 
1931 	key = request_key(key_string[0] == 'l' ? &key_type_logon : &key_type_user,
1932 			  key_desc + 1, NULL);
1933 	if (IS_ERR(key)) {
1934 		kzfree(new_key_string);
1935 		return PTR_ERR(key);
1936 	}
1937 
1938 	down_read(&key->sem);
1939 
1940 	ukp = user_key_payload_locked(key);
1941 	if (!ukp) {
1942 		up_read(&key->sem);
1943 		key_put(key);
1944 		kzfree(new_key_string);
1945 		return -EKEYREVOKED;
1946 	}
1947 
1948 	if (cc->key_size != ukp->datalen) {
1949 		up_read(&key->sem);
1950 		key_put(key);
1951 		kzfree(new_key_string);
1952 		return -EINVAL;
1953 	}
1954 
1955 	memcpy(cc->key, ukp->data, cc->key_size);
1956 
1957 	up_read(&key->sem);
1958 	key_put(key);
1959 
1960 	/* clear the flag since following operations may invalidate previously valid key */
1961 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
1962 
1963 	ret = crypt_setkey(cc);
1964 
1965 	if (!ret) {
1966 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
1967 		kzfree(cc->key_string);
1968 		cc->key_string = new_key_string;
1969 	} else
1970 		kzfree(new_key_string);
1971 
1972 	return ret;
1973 }
1974 
1975 static int get_key_size(char **key_string)
1976 {
1977 	char *colon, dummy;
1978 	int ret;
1979 
1980 	if (*key_string[0] != ':')
1981 		return strlen(*key_string) >> 1;
1982 
1983 	/* look for next ':' in key string */
1984 	colon = strpbrk(*key_string + 1, ":");
1985 	if (!colon)
1986 		return -EINVAL;
1987 
1988 	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
1989 		return -EINVAL;
1990 
1991 	*key_string = colon;
1992 
1993 	/* remaining key string should be :<logon|user>:<key_desc> */
1994 
1995 	return ret;
1996 }
1997 
1998 #else
1999 
2000 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2001 {
2002 	return -EINVAL;
2003 }
2004 
2005 static int get_key_size(char **key_string)
2006 {
2007 	return (*key_string[0] == ':') ? -EINVAL : strlen(*key_string) >> 1;
2008 }
2009 
2010 #endif
2011 
2012 static int crypt_set_key(struct crypt_config *cc, char *key)
2013 {
2014 	int r = -EINVAL;
2015 	int key_string_len = strlen(key);
2016 
2017 	/* Hyphen (which gives a key_size of zero) means there is no key. */
2018 	if (!cc->key_size && strcmp(key, "-"))
2019 		goto out;
2020 
2021 	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2022 	if (key[0] == ':') {
2023 		r = crypt_set_keyring_key(cc, key + 1);
2024 		goto out;
2025 	}
2026 
2027 	/* clear the flag since following operations may invalidate previously valid key */
2028 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2029 
2030 	/* wipe references to any kernel keyring key */
2031 	kzfree(cc->key_string);
2032 	cc->key_string = NULL;
2033 
2034 	/* Decode key from its hex representation. */
2035 	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2036 		goto out;
2037 
2038 	r = crypt_setkey(cc);
2039 	if (!r)
2040 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2041 
2042 out:
2043 	/* Hex key string not needed after here, so wipe it. */
2044 	memset(key, '0', key_string_len);
2045 
2046 	return r;
2047 }
2048 
2049 static int crypt_wipe_key(struct crypt_config *cc)
2050 {
2051 	int r;
2052 
2053 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2054 	get_random_bytes(&cc->key, cc->key_size);
2055 
2056 	/* Wipe IV private keys */
2057 	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2058 		r = cc->iv_gen_ops->wipe(cc);
2059 		if (r)
2060 			return r;
2061 	}
2062 
2063 	kzfree(cc->key_string);
2064 	cc->key_string = NULL;
2065 	r = crypt_setkey(cc);
2066 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2067 
2068 	return r;
2069 }
2070 
2071 static void crypt_calculate_pages_per_client(void)
2072 {
2073 	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2074 
2075 	if (!dm_crypt_clients_n)
2076 		return;
2077 
2078 	pages /= dm_crypt_clients_n;
2079 	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2080 		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2081 	dm_crypt_pages_per_client = pages;
2082 }
2083 
2084 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2085 {
2086 	struct crypt_config *cc = pool_data;
2087 	struct page *page;
2088 
2089 	if (unlikely(percpu_counter_compare(&cc->n_allocated_pages, dm_crypt_pages_per_client) >= 0) &&
2090 	    likely(gfp_mask & __GFP_NORETRY))
2091 		return NULL;
2092 
2093 	page = alloc_page(gfp_mask);
2094 	if (likely(page != NULL))
2095 		percpu_counter_add(&cc->n_allocated_pages, 1);
2096 
2097 	return page;
2098 }
2099 
2100 static void crypt_page_free(void *page, void *pool_data)
2101 {
2102 	struct crypt_config *cc = pool_data;
2103 
2104 	__free_page(page);
2105 	percpu_counter_sub(&cc->n_allocated_pages, 1);
2106 }
2107 
2108 static void crypt_dtr(struct dm_target *ti)
2109 {
2110 	struct crypt_config *cc = ti->private;
2111 
2112 	ti->private = NULL;
2113 
2114 	if (!cc)
2115 		return;
2116 
2117 	if (cc->write_thread)
2118 		kthread_stop(cc->write_thread);
2119 
2120 	if (cc->io_queue)
2121 		destroy_workqueue(cc->io_queue);
2122 	if (cc->crypt_queue)
2123 		destroy_workqueue(cc->crypt_queue);
2124 
2125 	crypt_free_tfms(cc);
2126 
2127 	bioset_exit(&cc->bs);
2128 
2129 	mempool_exit(&cc->page_pool);
2130 	mempool_exit(&cc->req_pool);
2131 	mempool_exit(&cc->tag_pool);
2132 
2133 	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2134 	percpu_counter_destroy(&cc->n_allocated_pages);
2135 
2136 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2137 		cc->iv_gen_ops->dtr(cc);
2138 
2139 	if (cc->dev)
2140 		dm_put_device(ti, cc->dev);
2141 
2142 	kzfree(cc->cipher_string);
2143 	kzfree(cc->key_string);
2144 	kzfree(cc->cipher_auth);
2145 	kzfree(cc->authenc_key);
2146 
2147 	mutex_destroy(&cc->bio_alloc_lock);
2148 
2149 	/* Must zero key material before freeing */
2150 	kzfree(cc);
2151 
2152 	spin_lock(&dm_crypt_clients_lock);
2153 	WARN_ON(!dm_crypt_clients_n);
2154 	dm_crypt_clients_n--;
2155 	crypt_calculate_pages_per_client();
2156 	spin_unlock(&dm_crypt_clients_lock);
2157 }
2158 
2159 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2160 {
2161 	struct crypt_config *cc = ti->private;
2162 
2163 	if (crypt_integrity_aead(cc))
2164 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2165 	else
2166 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2167 
2168 	if (cc->iv_size)
2169 		/* at least a 64 bit sector number should fit in our buffer */
2170 		cc->iv_size = max(cc->iv_size,
2171 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2172 	else if (ivmode) {
2173 		DMWARN("Selected cipher does not support IVs");
2174 		ivmode = NULL;
2175 	}
2176 
2177 	/* Choose ivmode, see comments at iv code. */
2178 	if (ivmode == NULL)
2179 		cc->iv_gen_ops = NULL;
2180 	else if (strcmp(ivmode, "plain") == 0)
2181 		cc->iv_gen_ops = &crypt_iv_plain_ops;
2182 	else if (strcmp(ivmode, "plain64") == 0)
2183 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2184 	else if (strcmp(ivmode, "plain64be") == 0)
2185 		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2186 	else if (strcmp(ivmode, "essiv") == 0)
2187 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2188 	else if (strcmp(ivmode, "benbi") == 0)
2189 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2190 	else if (strcmp(ivmode, "null") == 0)
2191 		cc->iv_gen_ops = &crypt_iv_null_ops;
2192 	else if (strcmp(ivmode, "eboiv") == 0)
2193 		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2194 	else if (strcmp(ivmode, "lmk") == 0) {
2195 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2196 		/*
2197 		 * Version 2 and 3 is recognised according
2198 		 * to length of provided multi-key string.
2199 		 * If present (version 3), last key is used as IV seed.
2200 		 * All keys (including IV seed) are always the same size.
2201 		 */
2202 		if (cc->key_size % cc->key_parts) {
2203 			cc->key_parts++;
2204 			cc->key_extra_size = cc->key_size / cc->key_parts;
2205 		}
2206 	} else if (strcmp(ivmode, "tcw") == 0) {
2207 		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2208 		cc->key_parts += 2; /* IV + whitening */
2209 		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2210 	} else if (strcmp(ivmode, "random") == 0) {
2211 		cc->iv_gen_ops = &crypt_iv_random_ops;
2212 		/* Need storage space in integrity fields. */
2213 		cc->integrity_iv_size = cc->iv_size;
2214 	} else {
2215 		ti->error = "Invalid IV mode";
2216 		return -EINVAL;
2217 	}
2218 
2219 	return 0;
2220 }
2221 
2222 /*
2223  * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2224  * The HMAC is needed to calculate tag size (HMAC digest size).
2225  * This should be probably done by crypto-api calls (once available...)
2226  */
2227 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2228 {
2229 	char *start, *end, *mac_alg = NULL;
2230 	struct crypto_ahash *mac;
2231 
2232 	if (!strstarts(cipher_api, "authenc("))
2233 		return 0;
2234 
2235 	start = strchr(cipher_api, '(');
2236 	end = strchr(cipher_api, ',');
2237 	if (!start || !end || ++start > end)
2238 		return -EINVAL;
2239 
2240 	mac_alg = kzalloc(end - start + 1, GFP_KERNEL);
2241 	if (!mac_alg)
2242 		return -ENOMEM;
2243 	strncpy(mac_alg, start, end - start);
2244 
2245 	mac = crypto_alloc_ahash(mac_alg, 0, 0);
2246 	kfree(mac_alg);
2247 
2248 	if (IS_ERR(mac))
2249 		return PTR_ERR(mac);
2250 
2251 	cc->key_mac_size = crypto_ahash_digestsize(mac);
2252 	crypto_free_ahash(mac);
2253 
2254 	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2255 	if (!cc->authenc_key)
2256 		return -ENOMEM;
2257 
2258 	return 0;
2259 }
2260 
2261 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2262 				char **ivmode, char **ivopts)
2263 {
2264 	struct crypt_config *cc = ti->private;
2265 	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2266 	int ret = -EINVAL;
2267 
2268 	cc->tfms_count = 1;
2269 
2270 	/*
2271 	 * New format (capi: prefix)
2272 	 * capi:cipher_api_spec-iv:ivopts
2273 	 */
2274 	tmp = &cipher_in[strlen("capi:")];
2275 
2276 	/* Separate IV options if present, it can contain another '-' in hash name */
2277 	*ivopts = strrchr(tmp, ':');
2278 	if (*ivopts) {
2279 		**ivopts = '\0';
2280 		(*ivopts)++;
2281 	}
2282 	/* Parse IV mode */
2283 	*ivmode = strrchr(tmp, '-');
2284 	if (*ivmode) {
2285 		**ivmode = '\0';
2286 		(*ivmode)++;
2287 	}
2288 	/* The rest is crypto API spec */
2289 	cipher_api = tmp;
2290 
2291 	/* Alloc AEAD, can be used only in new format. */
2292 	if (crypt_integrity_aead(cc)) {
2293 		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2294 		if (ret < 0) {
2295 			ti->error = "Invalid AEAD cipher spec";
2296 			return -ENOMEM;
2297 		}
2298 	}
2299 
2300 	if (*ivmode && !strcmp(*ivmode, "lmk"))
2301 		cc->tfms_count = 64;
2302 
2303 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2304 		if (!*ivopts) {
2305 			ti->error = "Digest algorithm missing for ESSIV mode";
2306 			return -EINVAL;
2307 		}
2308 		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2309 			       cipher_api, *ivopts);
2310 		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2311 			ti->error = "Cannot allocate cipher string";
2312 			return -ENOMEM;
2313 		}
2314 		cipher_api = buf;
2315 	}
2316 
2317 	cc->key_parts = cc->tfms_count;
2318 
2319 	/* Allocate cipher */
2320 	ret = crypt_alloc_tfms(cc, cipher_api);
2321 	if (ret < 0) {
2322 		ti->error = "Error allocating crypto tfm";
2323 		return ret;
2324 	}
2325 
2326 	if (crypt_integrity_aead(cc))
2327 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2328 	else
2329 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2330 
2331 	return 0;
2332 }
2333 
2334 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2335 				char **ivmode, char **ivopts)
2336 {
2337 	struct crypt_config *cc = ti->private;
2338 	char *tmp, *cipher, *chainmode, *keycount;
2339 	char *cipher_api = NULL;
2340 	int ret = -EINVAL;
2341 	char dummy;
2342 
2343 	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2344 		ti->error = "Bad cipher specification";
2345 		return -EINVAL;
2346 	}
2347 
2348 	/*
2349 	 * Legacy dm-crypt cipher specification
2350 	 * cipher[:keycount]-mode-iv:ivopts
2351 	 */
2352 	tmp = cipher_in;
2353 	keycount = strsep(&tmp, "-");
2354 	cipher = strsep(&keycount, ":");
2355 
2356 	if (!keycount)
2357 		cc->tfms_count = 1;
2358 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2359 		 !is_power_of_2(cc->tfms_count)) {
2360 		ti->error = "Bad cipher key count specification";
2361 		return -EINVAL;
2362 	}
2363 	cc->key_parts = cc->tfms_count;
2364 
2365 	chainmode = strsep(&tmp, "-");
2366 	*ivmode = strsep(&tmp, ":");
2367 	*ivopts = tmp;
2368 
2369 	/*
2370 	 * For compatibility with the original dm-crypt mapping format, if
2371 	 * only the cipher name is supplied, use cbc-plain.
2372 	 */
2373 	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2374 		chainmode = "cbc";
2375 		*ivmode = "plain";
2376 	}
2377 
2378 	if (strcmp(chainmode, "ecb") && !*ivmode) {
2379 		ti->error = "IV mechanism required";
2380 		return -EINVAL;
2381 	}
2382 
2383 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
2384 	if (!cipher_api)
2385 		goto bad_mem;
2386 
2387 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2388 		if (!*ivopts) {
2389 			ti->error = "Digest algorithm missing for ESSIV mode";
2390 			kfree(cipher_api);
2391 			return -EINVAL;
2392 		}
2393 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2394 			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
2395 	} else {
2396 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2397 			       "%s(%s)", chainmode, cipher);
2398 	}
2399 	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2400 		kfree(cipher_api);
2401 		goto bad_mem;
2402 	}
2403 
2404 	/* Allocate cipher */
2405 	ret = crypt_alloc_tfms(cc, cipher_api);
2406 	if (ret < 0) {
2407 		ti->error = "Error allocating crypto tfm";
2408 		kfree(cipher_api);
2409 		return ret;
2410 	}
2411 	kfree(cipher_api);
2412 
2413 	return 0;
2414 bad_mem:
2415 	ti->error = "Cannot allocate cipher strings";
2416 	return -ENOMEM;
2417 }
2418 
2419 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
2420 {
2421 	struct crypt_config *cc = ti->private;
2422 	char *ivmode = NULL, *ivopts = NULL;
2423 	int ret;
2424 
2425 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
2426 	if (!cc->cipher_string) {
2427 		ti->error = "Cannot allocate cipher strings";
2428 		return -ENOMEM;
2429 	}
2430 
2431 	if (strstarts(cipher_in, "capi:"))
2432 		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
2433 	else
2434 		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
2435 	if (ret)
2436 		return ret;
2437 
2438 	/* Initialize IV */
2439 	ret = crypt_ctr_ivmode(ti, ivmode);
2440 	if (ret < 0)
2441 		return ret;
2442 
2443 	/* Initialize and set key */
2444 	ret = crypt_set_key(cc, key);
2445 	if (ret < 0) {
2446 		ti->error = "Error decoding and setting key";
2447 		return ret;
2448 	}
2449 
2450 	/* Allocate IV */
2451 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
2452 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
2453 		if (ret < 0) {
2454 			ti->error = "Error creating IV";
2455 			return ret;
2456 		}
2457 	}
2458 
2459 	/* Initialize IV (set keys for ESSIV etc) */
2460 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
2461 		ret = cc->iv_gen_ops->init(cc);
2462 		if (ret < 0) {
2463 			ti->error = "Error initialising IV";
2464 			return ret;
2465 		}
2466 	}
2467 
2468 	/* wipe the kernel key payload copy */
2469 	if (cc->key_string)
2470 		memset(cc->key, 0, cc->key_size * sizeof(u8));
2471 
2472 	return ret;
2473 }
2474 
2475 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
2476 {
2477 	struct crypt_config *cc = ti->private;
2478 	struct dm_arg_set as;
2479 	static const struct dm_arg _args[] = {
2480 		{0, 6, "Invalid number of feature args"},
2481 	};
2482 	unsigned int opt_params, val;
2483 	const char *opt_string, *sval;
2484 	char dummy;
2485 	int ret;
2486 
2487 	/* Optional parameters */
2488 	as.argc = argc;
2489 	as.argv = argv;
2490 
2491 	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
2492 	if (ret)
2493 		return ret;
2494 
2495 	while (opt_params--) {
2496 		opt_string = dm_shift_arg(&as);
2497 		if (!opt_string) {
2498 			ti->error = "Not enough feature arguments";
2499 			return -EINVAL;
2500 		}
2501 
2502 		if (!strcasecmp(opt_string, "allow_discards"))
2503 			ti->num_discard_bios = 1;
2504 
2505 		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
2506 			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
2507 
2508 		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
2509 			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
2510 		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
2511 			if (val == 0 || val > MAX_TAG_SIZE) {
2512 				ti->error = "Invalid integrity arguments";
2513 				return -EINVAL;
2514 			}
2515 			cc->on_disk_tag_size = val;
2516 			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
2517 			if (!strcasecmp(sval, "aead")) {
2518 				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
2519 			} else  if (strcasecmp(sval, "none")) {
2520 				ti->error = "Unknown integrity profile";
2521 				return -EINVAL;
2522 			}
2523 
2524 			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
2525 			if (!cc->cipher_auth)
2526 				return -ENOMEM;
2527 		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
2528 			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
2529 			    cc->sector_size > 4096 ||
2530 			    (cc->sector_size & (cc->sector_size - 1))) {
2531 				ti->error = "Invalid feature value for sector_size";
2532 				return -EINVAL;
2533 			}
2534 			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
2535 				ti->error = "Device size is not multiple of sector_size feature";
2536 				return -EINVAL;
2537 			}
2538 			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
2539 		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
2540 			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
2541 		else {
2542 			ti->error = "Invalid feature arguments";
2543 			return -EINVAL;
2544 		}
2545 	}
2546 
2547 	return 0;
2548 }
2549 
2550 /*
2551  * Construct an encryption mapping:
2552  * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
2553  */
2554 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2555 {
2556 	struct crypt_config *cc;
2557 	const char *devname = dm_table_device_name(ti->table);
2558 	int key_size;
2559 	unsigned int align_mask;
2560 	unsigned long long tmpll;
2561 	int ret;
2562 	size_t iv_size_padding, additional_req_size;
2563 	char dummy;
2564 
2565 	if (argc < 5) {
2566 		ti->error = "Not enough arguments";
2567 		return -EINVAL;
2568 	}
2569 
2570 	key_size = get_key_size(&argv[1]);
2571 	if (key_size < 0) {
2572 		ti->error = "Cannot parse key size";
2573 		return -EINVAL;
2574 	}
2575 
2576 	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
2577 	if (!cc) {
2578 		ti->error = "Cannot allocate encryption context";
2579 		return -ENOMEM;
2580 	}
2581 	cc->key_size = key_size;
2582 	cc->sector_size = (1 << SECTOR_SHIFT);
2583 	cc->sector_shift = 0;
2584 
2585 	ti->private = cc;
2586 
2587 	spin_lock(&dm_crypt_clients_lock);
2588 	dm_crypt_clients_n++;
2589 	crypt_calculate_pages_per_client();
2590 	spin_unlock(&dm_crypt_clients_lock);
2591 
2592 	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
2593 	if (ret < 0)
2594 		goto bad;
2595 
2596 	/* Optional parameters need to be read before cipher constructor */
2597 	if (argc > 5) {
2598 		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
2599 		if (ret)
2600 			goto bad;
2601 	}
2602 
2603 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
2604 	if (ret < 0)
2605 		goto bad;
2606 
2607 	if (crypt_integrity_aead(cc)) {
2608 		cc->dmreq_start = sizeof(struct aead_request);
2609 		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
2610 		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
2611 	} else {
2612 		cc->dmreq_start = sizeof(struct skcipher_request);
2613 		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
2614 		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
2615 	}
2616 	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
2617 
2618 	if (align_mask < CRYPTO_MINALIGN) {
2619 		/* Allocate the padding exactly */
2620 		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
2621 				& align_mask;
2622 	} else {
2623 		/*
2624 		 * If the cipher requires greater alignment than kmalloc
2625 		 * alignment, we don't know the exact position of the
2626 		 * initialization vector. We must assume worst case.
2627 		 */
2628 		iv_size_padding = align_mask;
2629 	}
2630 
2631 	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
2632 	additional_req_size = sizeof(struct dm_crypt_request) +
2633 		iv_size_padding + cc->iv_size +
2634 		cc->iv_size +
2635 		sizeof(uint64_t) +
2636 		sizeof(unsigned int);
2637 
2638 	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
2639 	if (ret) {
2640 		ti->error = "Cannot allocate crypt request mempool";
2641 		goto bad;
2642 	}
2643 
2644 	cc->per_bio_data_size = ti->per_io_data_size =
2645 		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
2646 		      ARCH_KMALLOC_MINALIGN);
2647 
2648 	ret = mempool_init(&cc->page_pool, BIO_MAX_PAGES, crypt_page_alloc, crypt_page_free, cc);
2649 	if (ret) {
2650 		ti->error = "Cannot allocate page mempool";
2651 		goto bad;
2652 	}
2653 
2654 	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
2655 	if (ret) {
2656 		ti->error = "Cannot allocate crypt bioset";
2657 		goto bad;
2658 	}
2659 
2660 	mutex_init(&cc->bio_alloc_lock);
2661 
2662 	ret = -EINVAL;
2663 	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
2664 	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
2665 		ti->error = "Invalid iv_offset sector";
2666 		goto bad;
2667 	}
2668 	cc->iv_offset = tmpll;
2669 
2670 	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
2671 	if (ret) {
2672 		ti->error = "Device lookup failed";
2673 		goto bad;
2674 	}
2675 
2676 	ret = -EINVAL;
2677 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
2678 		ti->error = "Invalid device sector";
2679 		goto bad;
2680 	}
2681 	cc->start = tmpll;
2682 
2683 	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
2684 		ret = crypt_integrity_ctr(cc, ti);
2685 		if (ret)
2686 			goto bad;
2687 
2688 		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
2689 		if (!cc->tag_pool_max_sectors)
2690 			cc->tag_pool_max_sectors = 1;
2691 
2692 		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
2693 			cc->tag_pool_max_sectors * cc->on_disk_tag_size);
2694 		if (ret) {
2695 			ti->error = "Cannot allocate integrity tags mempool";
2696 			goto bad;
2697 		}
2698 
2699 		cc->tag_pool_max_sectors <<= cc->sector_shift;
2700 	}
2701 
2702 	ret = -ENOMEM;
2703 	cc->io_queue = alloc_workqueue("kcryptd_io/%s",
2704 				       WQ_HIGHPRI | WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
2705 				       1, devname);
2706 	if (!cc->io_queue) {
2707 		ti->error = "Couldn't create kcryptd io queue";
2708 		goto bad;
2709 	}
2710 
2711 	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
2712 		cc->crypt_queue = alloc_workqueue("kcryptd/%s",
2713 						  WQ_HIGHPRI | WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
2714 						  1, devname);
2715 	else
2716 		cc->crypt_queue = alloc_workqueue("kcryptd/%s",
2717 						  WQ_HIGHPRI | WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
2718 						  num_online_cpus(), devname);
2719 	if (!cc->crypt_queue) {
2720 		ti->error = "Couldn't create kcryptd queue";
2721 		goto bad;
2722 	}
2723 
2724 	spin_lock_init(&cc->write_thread_lock);
2725 	cc->write_tree = RB_ROOT;
2726 
2727 	cc->write_thread = kthread_create(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
2728 	if (IS_ERR(cc->write_thread)) {
2729 		ret = PTR_ERR(cc->write_thread);
2730 		cc->write_thread = NULL;
2731 		ti->error = "Couldn't spawn write thread";
2732 		goto bad;
2733 	}
2734 	wake_up_process(cc->write_thread);
2735 
2736 	ti->num_flush_bios = 1;
2737 
2738 	return 0;
2739 
2740 bad:
2741 	crypt_dtr(ti);
2742 	return ret;
2743 }
2744 
2745 static int crypt_map(struct dm_target *ti, struct bio *bio)
2746 {
2747 	struct dm_crypt_io *io;
2748 	struct crypt_config *cc = ti->private;
2749 
2750 	/*
2751 	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
2752 	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
2753 	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
2754 	 */
2755 	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
2756 	    bio_op(bio) == REQ_OP_DISCARD)) {
2757 		bio_set_dev(bio, cc->dev->bdev);
2758 		if (bio_sectors(bio))
2759 			bio->bi_iter.bi_sector = cc->start +
2760 				dm_target_offset(ti, bio->bi_iter.bi_sector);
2761 		return DM_MAPIO_REMAPPED;
2762 	}
2763 
2764 	/*
2765 	 * Check if bio is too large, split as needed.
2766 	 */
2767 	if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_PAGES << PAGE_SHIFT)) &&
2768 	    (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
2769 		dm_accept_partial_bio(bio, ((BIO_MAX_PAGES << PAGE_SHIFT) >> SECTOR_SHIFT));
2770 
2771 	/*
2772 	 * Ensure that bio is a multiple of internal sector encryption size
2773 	 * and is aligned to this size as defined in IO hints.
2774 	 */
2775 	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
2776 		return DM_MAPIO_KILL;
2777 
2778 	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
2779 		return DM_MAPIO_KILL;
2780 
2781 	io = dm_per_bio_data(bio, cc->per_bio_data_size);
2782 	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
2783 
2784 	if (cc->on_disk_tag_size) {
2785 		unsigned tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
2786 
2787 		if (unlikely(tag_len > KMALLOC_MAX_SIZE) ||
2788 		    unlikely(!(io->integrity_metadata = kmalloc(tag_len,
2789 				GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN)))) {
2790 			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
2791 				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
2792 			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
2793 			io->integrity_metadata_from_pool = true;
2794 		}
2795 	}
2796 
2797 	if (crypt_integrity_aead(cc))
2798 		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
2799 	else
2800 		io->ctx.r.req = (struct skcipher_request *)(io + 1);
2801 
2802 	if (bio_data_dir(io->base_bio) == READ) {
2803 		if (kcryptd_io_read(io, GFP_NOWAIT))
2804 			kcryptd_queue_read(io);
2805 	} else
2806 		kcryptd_queue_crypt(io);
2807 
2808 	return DM_MAPIO_SUBMITTED;
2809 }
2810 
2811 static void crypt_status(struct dm_target *ti, status_type_t type,
2812 			 unsigned status_flags, char *result, unsigned maxlen)
2813 {
2814 	struct crypt_config *cc = ti->private;
2815 	unsigned i, sz = 0;
2816 	int num_feature_args = 0;
2817 
2818 	switch (type) {
2819 	case STATUSTYPE_INFO:
2820 		result[0] = '\0';
2821 		break;
2822 
2823 	case STATUSTYPE_TABLE:
2824 		DMEMIT("%s ", cc->cipher_string);
2825 
2826 		if (cc->key_size > 0) {
2827 			if (cc->key_string)
2828 				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
2829 			else
2830 				for (i = 0; i < cc->key_size; i++)
2831 					DMEMIT("%02x", cc->key[i]);
2832 		} else
2833 			DMEMIT("-");
2834 
2835 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
2836 				cc->dev->name, (unsigned long long)cc->start);
2837 
2838 		num_feature_args += !!ti->num_discard_bios;
2839 		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
2840 		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
2841 		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
2842 		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
2843 		if (cc->on_disk_tag_size)
2844 			num_feature_args++;
2845 		if (num_feature_args) {
2846 			DMEMIT(" %d", num_feature_args);
2847 			if (ti->num_discard_bios)
2848 				DMEMIT(" allow_discards");
2849 			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
2850 				DMEMIT(" same_cpu_crypt");
2851 			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
2852 				DMEMIT(" submit_from_crypt_cpus");
2853 			if (cc->on_disk_tag_size)
2854 				DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
2855 			if (cc->sector_size != (1 << SECTOR_SHIFT))
2856 				DMEMIT(" sector_size:%d", cc->sector_size);
2857 			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
2858 				DMEMIT(" iv_large_sectors");
2859 		}
2860 
2861 		break;
2862 	}
2863 }
2864 
2865 static void crypt_postsuspend(struct dm_target *ti)
2866 {
2867 	struct crypt_config *cc = ti->private;
2868 
2869 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
2870 }
2871 
2872 static int crypt_preresume(struct dm_target *ti)
2873 {
2874 	struct crypt_config *cc = ti->private;
2875 
2876 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
2877 		DMERR("aborting resume - crypt key is not set.");
2878 		return -EAGAIN;
2879 	}
2880 
2881 	return 0;
2882 }
2883 
2884 static void crypt_resume(struct dm_target *ti)
2885 {
2886 	struct crypt_config *cc = ti->private;
2887 
2888 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
2889 }
2890 
2891 /* Message interface
2892  *	key set <key>
2893  *	key wipe
2894  */
2895 static int crypt_message(struct dm_target *ti, unsigned argc, char **argv,
2896 			 char *result, unsigned maxlen)
2897 {
2898 	struct crypt_config *cc = ti->private;
2899 	int key_size, ret = -EINVAL;
2900 
2901 	if (argc < 2)
2902 		goto error;
2903 
2904 	if (!strcasecmp(argv[0], "key")) {
2905 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
2906 			DMWARN("not suspended during key manipulation.");
2907 			return -EINVAL;
2908 		}
2909 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
2910 			/* The key size may not be changed. */
2911 			key_size = get_key_size(&argv[2]);
2912 			if (key_size < 0 || cc->key_size != key_size) {
2913 				memset(argv[2], '0', strlen(argv[2]));
2914 				return -EINVAL;
2915 			}
2916 
2917 			ret = crypt_set_key(cc, argv[2]);
2918 			if (ret)
2919 				return ret;
2920 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
2921 				ret = cc->iv_gen_ops->init(cc);
2922 			/* wipe the kernel key payload copy */
2923 			if (cc->key_string)
2924 				memset(cc->key, 0, cc->key_size * sizeof(u8));
2925 			return ret;
2926 		}
2927 		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
2928 			return crypt_wipe_key(cc);
2929 	}
2930 
2931 error:
2932 	DMWARN("unrecognised message received.");
2933 	return -EINVAL;
2934 }
2935 
2936 static int crypt_iterate_devices(struct dm_target *ti,
2937 				 iterate_devices_callout_fn fn, void *data)
2938 {
2939 	struct crypt_config *cc = ti->private;
2940 
2941 	return fn(ti, cc->dev, cc->start, ti->len, data);
2942 }
2943 
2944 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
2945 {
2946 	struct crypt_config *cc = ti->private;
2947 
2948 	/*
2949 	 * Unfortunate constraint that is required to avoid the potential
2950 	 * for exceeding underlying device's max_segments limits -- due to
2951 	 * crypt_alloc_buffer() possibly allocating pages for the encryption
2952 	 * bio that are not as physically contiguous as the original bio.
2953 	 */
2954 	limits->max_segment_size = PAGE_SIZE;
2955 
2956 	limits->logical_block_size =
2957 		max_t(unsigned short, limits->logical_block_size, cc->sector_size);
2958 	limits->physical_block_size =
2959 		max_t(unsigned, limits->physical_block_size, cc->sector_size);
2960 	limits->io_min = max_t(unsigned, limits->io_min, cc->sector_size);
2961 }
2962 
2963 static struct target_type crypt_target = {
2964 	.name   = "crypt",
2965 	.version = {1, 19, 0},
2966 	.module = THIS_MODULE,
2967 	.ctr    = crypt_ctr,
2968 	.dtr    = crypt_dtr,
2969 	.map    = crypt_map,
2970 	.status = crypt_status,
2971 	.postsuspend = crypt_postsuspend,
2972 	.preresume = crypt_preresume,
2973 	.resume = crypt_resume,
2974 	.message = crypt_message,
2975 	.iterate_devices = crypt_iterate_devices,
2976 	.io_hints = crypt_io_hints,
2977 };
2978 
2979 static int __init dm_crypt_init(void)
2980 {
2981 	int r;
2982 
2983 	r = dm_register_target(&crypt_target);
2984 	if (r < 0)
2985 		DMERR("register failed %d", r);
2986 
2987 	return r;
2988 }
2989 
2990 static void __exit dm_crypt_exit(void)
2991 {
2992 	dm_unregister_target(&crypt_target);
2993 }
2994 
2995 module_init(dm_crypt_init);
2996 module_exit(dm_crypt_exit);
2997 
2998 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
2999 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3000 MODULE_LICENSE("GPL");
3001