xref: /linux/drivers/md/dm-crypt.c (revision 90d32e92011eaae8e70a9169b4e7acf4ca8f9d3a)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2003 Jana Saout <jana@saout.de>
4  * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
5  * Copyright (C) 2006-2020 Red Hat, Inc. All rights reserved.
6  * Copyright (C) 2013-2020 Milan Broz <gmazyland@gmail.com>
7  *
8  * This file is released under the GPL.
9  */
10 
11 #include <linux/completion.h>
12 #include <linux/err.h>
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/kernel.h>
16 #include <linux/key.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-integrity.h>
20 #include <linux/mempool.h>
21 #include <linux/slab.h>
22 #include <linux/crypto.h>
23 #include <linux/workqueue.h>
24 #include <linux/kthread.h>
25 #include <linux/backing-dev.h>
26 #include <linux/atomic.h>
27 #include <linux/scatterlist.h>
28 #include <linux/rbtree.h>
29 #include <linux/ctype.h>
30 #include <asm/page.h>
31 #include <asm/unaligned.h>
32 #include <crypto/hash.h>
33 #include <crypto/md5.h>
34 #include <crypto/skcipher.h>
35 #include <crypto/aead.h>
36 #include <crypto/authenc.h>
37 #include <crypto/utils.h>
38 #include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
39 #include <linux/key-type.h>
40 #include <keys/user-type.h>
41 #include <keys/encrypted-type.h>
42 #include <keys/trusted-type.h>
43 
44 #include <linux/device-mapper.h>
45 
46 #include "dm-audit.h"
47 
48 #define DM_MSG_PREFIX "crypt"
49 
50 static DEFINE_IDA(workqueue_ida);
51 
52 /*
53  * context holding the current state of a multi-part conversion
54  */
55 struct convert_context {
56 	struct completion restart;
57 	struct bio *bio_in;
58 	struct bvec_iter iter_in;
59 	struct bio *bio_out;
60 	struct bvec_iter iter_out;
61 	atomic_t cc_pending;
62 	u64 cc_sector;
63 	union {
64 		struct skcipher_request *req;
65 		struct aead_request *req_aead;
66 	} r;
67 	bool aead_recheck;
68 	bool aead_failed;
69 
70 };
71 
72 /*
73  * per bio private data
74  */
75 struct dm_crypt_io {
76 	struct crypt_config *cc;
77 	struct bio *base_bio;
78 	u8 *integrity_metadata;
79 	bool integrity_metadata_from_pool:1;
80 
81 	struct work_struct work;
82 
83 	struct convert_context ctx;
84 
85 	atomic_t io_pending;
86 	blk_status_t error;
87 	sector_t sector;
88 
89 	struct bvec_iter saved_bi_iter;
90 
91 	struct rb_node rb_node;
92 } CRYPTO_MINALIGN_ATTR;
93 
94 struct dm_crypt_request {
95 	struct convert_context *ctx;
96 	struct scatterlist sg_in[4];
97 	struct scatterlist sg_out[4];
98 	u64 iv_sector;
99 };
100 
101 struct crypt_config;
102 
103 struct crypt_iv_operations {
104 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
105 		   const char *opts);
106 	void (*dtr)(struct crypt_config *cc);
107 	int (*init)(struct crypt_config *cc);
108 	int (*wipe)(struct crypt_config *cc);
109 	int (*generator)(struct crypt_config *cc, u8 *iv,
110 			 struct dm_crypt_request *dmreq);
111 	int (*post)(struct crypt_config *cc, u8 *iv,
112 		    struct dm_crypt_request *dmreq);
113 };
114 
115 struct iv_benbi_private {
116 	int shift;
117 };
118 
119 #define LMK_SEED_SIZE 64 /* hash + 0 */
120 struct iv_lmk_private {
121 	struct crypto_shash *hash_tfm;
122 	u8 *seed;
123 };
124 
125 #define TCW_WHITENING_SIZE 16
126 struct iv_tcw_private {
127 	struct crypto_shash *crc32_tfm;
128 	u8 *iv_seed;
129 	u8 *whitening;
130 };
131 
132 #define ELEPHANT_MAX_KEY_SIZE 32
133 struct iv_elephant_private {
134 	struct crypto_skcipher *tfm;
135 };
136 
137 /*
138  * Crypt: maps a linear range of a block device
139  * and encrypts / decrypts at the same time.
140  */
141 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
142 	     DM_CRYPT_SAME_CPU, DM_CRYPT_HIGH_PRIORITY,
143 	     DM_CRYPT_NO_OFFLOAD, DM_CRYPT_NO_READ_WORKQUEUE,
144 	     DM_CRYPT_NO_WRITE_WORKQUEUE, DM_CRYPT_WRITE_INLINE };
145 
146 enum cipher_flags {
147 	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cipher */
148 	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
149 	CRYPT_ENCRYPT_PREPROCESS,	/* Must preprocess data for encryption (elephant) */
150 };
151 
152 /*
153  * The fields in here must be read only after initialization.
154  */
155 struct crypt_config {
156 	struct dm_dev *dev;
157 	sector_t start;
158 
159 	struct percpu_counter n_allocated_pages;
160 
161 	struct workqueue_struct *io_queue;
162 	struct workqueue_struct *crypt_queue;
163 
164 	spinlock_t write_thread_lock;
165 	struct task_struct *write_thread;
166 	struct rb_root write_tree;
167 
168 	char *cipher_string;
169 	char *cipher_auth;
170 	char *key_string;
171 
172 	const struct crypt_iv_operations *iv_gen_ops;
173 	union {
174 		struct iv_benbi_private benbi;
175 		struct iv_lmk_private lmk;
176 		struct iv_tcw_private tcw;
177 		struct iv_elephant_private elephant;
178 	} iv_gen_private;
179 	u64 iv_offset;
180 	unsigned int iv_size;
181 	unsigned short sector_size;
182 	unsigned char sector_shift;
183 
184 	union {
185 		struct crypto_skcipher **tfms;
186 		struct crypto_aead **tfms_aead;
187 	} cipher_tfm;
188 	unsigned int tfms_count;
189 	int workqueue_id;
190 	unsigned long cipher_flags;
191 
192 	/*
193 	 * Layout of each crypto request:
194 	 *
195 	 *   struct skcipher_request
196 	 *      context
197 	 *      padding
198 	 *   struct dm_crypt_request
199 	 *      padding
200 	 *   IV
201 	 *
202 	 * The padding is added so that dm_crypt_request and the IV are
203 	 * correctly aligned.
204 	 */
205 	unsigned int dmreq_start;
206 
207 	unsigned int per_bio_data_size;
208 
209 	unsigned long flags;
210 	unsigned int key_size;
211 	unsigned int key_parts;      /* independent parts in key buffer */
212 	unsigned int key_extra_size; /* additional keys length */
213 	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
214 
215 	unsigned int integrity_tag_size;
216 	unsigned int integrity_iv_size;
217 	unsigned int on_disk_tag_size;
218 
219 	/*
220 	 * pool for per bio private data, crypto requests,
221 	 * encryption requeusts/buffer pages and integrity tags
222 	 */
223 	unsigned int tag_pool_max_sectors;
224 	mempool_t tag_pool;
225 	mempool_t req_pool;
226 	mempool_t page_pool;
227 
228 	struct bio_set bs;
229 	struct mutex bio_alloc_lock;
230 
231 	u8 *authenc_key; /* space for keys in authenc() format (if used) */
232 	u8 key[] __counted_by(key_size);
233 };
234 
235 #define MIN_IOS		64
236 #define MAX_TAG_SIZE	480
237 #define POOL_ENTRY_SIZE	512
238 
239 static DEFINE_SPINLOCK(dm_crypt_clients_lock);
240 static unsigned int dm_crypt_clients_n;
241 static volatile unsigned long dm_crypt_pages_per_client;
242 #define DM_CRYPT_MEMORY_PERCENT			2
243 #define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_VECS * 16)
244 
245 static void crypt_endio(struct bio *clone);
246 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
247 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
248 					     struct scatterlist *sg);
249 
250 static bool crypt_integrity_aead(struct crypt_config *cc);
251 
252 /*
253  * Use this to access cipher attributes that are independent of the key.
254  */
255 static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
256 {
257 	return cc->cipher_tfm.tfms[0];
258 }
259 
260 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
261 {
262 	return cc->cipher_tfm.tfms_aead[0];
263 }
264 
265 /*
266  * Different IV generation algorithms:
267  *
268  * plain: the initial vector is the 32-bit little-endian version of the sector
269  *        number, padded with zeros if necessary.
270  *
271  * plain64: the initial vector is the 64-bit little-endian version of the sector
272  *        number, padded with zeros if necessary.
273  *
274  * plain64be: the initial vector is the 64-bit big-endian version of the sector
275  *        number, padded with zeros if necessary.
276  *
277  * essiv: "encrypted sector|salt initial vector", the sector number is
278  *        encrypted with the bulk cipher using a salt as key. The salt
279  *        should be derived from the bulk cipher's key via hashing.
280  *
281  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
282  *        (needed for LRW-32-AES and possible other narrow block modes)
283  *
284  * null: the initial vector is always zero.  Provides compatibility with
285  *       obsolete loop_fish2 devices.  Do not use for new devices.
286  *
287  * lmk:  Compatible implementation of the block chaining mode used
288  *       by the Loop-AES block device encryption system
289  *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
290  *       It operates on full 512 byte sectors and uses CBC
291  *       with an IV derived from the sector number, the data and
292  *       optionally extra IV seed.
293  *       This means that after decryption the first block
294  *       of sector must be tweaked according to decrypted data.
295  *       Loop-AES can use three encryption schemes:
296  *         version 1: is plain aes-cbc mode
297  *         version 2: uses 64 multikey scheme with lmk IV generator
298  *         version 3: the same as version 2 with additional IV seed
299  *                   (it uses 65 keys, last key is used as IV seed)
300  *
301  * tcw:  Compatible implementation of the block chaining mode used
302  *       by the TrueCrypt device encryption system (prior to version 4.1).
303  *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
304  *       It operates on full 512 byte sectors and uses CBC
305  *       with an IV derived from initial key and the sector number.
306  *       In addition, whitening value is applied on every sector, whitening
307  *       is calculated from initial key, sector number and mixed using CRC32.
308  *       Note that this encryption scheme is vulnerable to watermarking attacks
309  *       and should be used for old compatible containers access only.
310  *
311  * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
312  *        The IV is encrypted little-endian byte-offset (with the same key
313  *        and cipher as the volume).
314  *
315  * elephant: The extended version of eboiv with additional Elephant diffuser
316  *           used with Bitlocker CBC mode.
317  *           This mode was used in older Windows systems
318  *           https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
319  */
320 
321 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
322 			      struct dm_crypt_request *dmreq)
323 {
324 	memset(iv, 0, cc->iv_size);
325 	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
326 
327 	return 0;
328 }
329 
330 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
331 				struct dm_crypt_request *dmreq)
332 {
333 	memset(iv, 0, cc->iv_size);
334 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
335 
336 	return 0;
337 }
338 
339 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
340 				  struct dm_crypt_request *dmreq)
341 {
342 	memset(iv, 0, cc->iv_size);
343 	/* iv_size is at least of size u64; usually it is 16 bytes */
344 	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
345 
346 	return 0;
347 }
348 
349 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
350 			      struct dm_crypt_request *dmreq)
351 {
352 	/*
353 	 * ESSIV encryption of the IV is now handled by the crypto API,
354 	 * so just pass the plain sector number here.
355 	 */
356 	memset(iv, 0, cc->iv_size);
357 	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
358 
359 	return 0;
360 }
361 
362 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
363 			      const char *opts)
364 {
365 	unsigned int bs;
366 	int log;
367 
368 	if (crypt_integrity_aead(cc))
369 		bs = crypto_aead_blocksize(any_tfm_aead(cc));
370 	else
371 		bs = crypto_skcipher_blocksize(any_tfm(cc));
372 	log = ilog2(bs);
373 
374 	/*
375 	 * We need to calculate how far we must shift the sector count
376 	 * to get the cipher block count, we use this shift in _gen.
377 	 */
378 	if (1 << log != bs) {
379 		ti->error = "cypher blocksize is not a power of 2";
380 		return -EINVAL;
381 	}
382 
383 	if (log > 9) {
384 		ti->error = "cypher blocksize is > 512";
385 		return -EINVAL;
386 	}
387 
388 	cc->iv_gen_private.benbi.shift = 9 - log;
389 
390 	return 0;
391 }
392 
393 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
394 {
395 }
396 
397 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
398 			      struct dm_crypt_request *dmreq)
399 {
400 	__be64 val;
401 
402 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
403 
404 	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
405 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
406 
407 	return 0;
408 }
409 
410 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
411 			     struct dm_crypt_request *dmreq)
412 {
413 	memset(iv, 0, cc->iv_size);
414 
415 	return 0;
416 }
417 
418 static void crypt_iv_lmk_dtr(struct crypt_config *cc)
419 {
420 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
421 
422 	if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
423 		crypto_free_shash(lmk->hash_tfm);
424 	lmk->hash_tfm = NULL;
425 
426 	kfree_sensitive(lmk->seed);
427 	lmk->seed = NULL;
428 }
429 
430 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
431 			    const char *opts)
432 {
433 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
434 
435 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
436 		ti->error = "Unsupported sector size for LMK";
437 		return -EINVAL;
438 	}
439 
440 	lmk->hash_tfm = crypto_alloc_shash("md5", 0,
441 					   CRYPTO_ALG_ALLOCATES_MEMORY);
442 	if (IS_ERR(lmk->hash_tfm)) {
443 		ti->error = "Error initializing LMK hash";
444 		return PTR_ERR(lmk->hash_tfm);
445 	}
446 
447 	/* No seed in LMK version 2 */
448 	if (cc->key_parts == cc->tfms_count) {
449 		lmk->seed = NULL;
450 		return 0;
451 	}
452 
453 	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
454 	if (!lmk->seed) {
455 		crypt_iv_lmk_dtr(cc);
456 		ti->error = "Error kmallocing seed storage in LMK";
457 		return -ENOMEM;
458 	}
459 
460 	return 0;
461 }
462 
463 static int crypt_iv_lmk_init(struct crypt_config *cc)
464 {
465 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
466 	int subkey_size = cc->key_size / cc->key_parts;
467 
468 	/* LMK seed is on the position of LMK_KEYS + 1 key */
469 	if (lmk->seed)
470 		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
471 		       crypto_shash_digestsize(lmk->hash_tfm));
472 
473 	return 0;
474 }
475 
476 static int crypt_iv_lmk_wipe(struct crypt_config *cc)
477 {
478 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
479 
480 	if (lmk->seed)
481 		memset(lmk->seed, 0, LMK_SEED_SIZE);
482 
483 	return 0;
484 }
485 
486 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
487 			    struct dm_crypt_request *dmreq,
488 			    u8 *data)
489 {
490 	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
491 	SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
492 	struct md5_state md5state;
493 	__le32 buf[4];
494 	int i, r;
495 
496 	desc->tfm = lmk->hash_tfm;
497 
498 	r = crypto_shash_init(desc);
499 	if (r)
500 		return r;
501 
502 	if (lmk->seed) {
503 		r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
504 		if (r)
505 			return r;
506 	}
507 
508 	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
509 	r = crypto_shash_update(desc, data + 16, 16 * 31);
510 	if (r)
511 		return r;
512 
513 	/* Sector is cropped to 56 bits here */
514 	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
515 	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
516 	buf[2] = cpu_to_le32(4024);
517 	buf[3] = 0;
518 	r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
519 	if (r)
520 		return r;
521 
522 	/* No MD5 padding here */
523 	r = crypto_shash_export(desc, &md5state);
524 	if (r)
525 		return r;
526 
527 	for (i = 0; i < MD5_HASH_WORDS; i++)
528 		__cpu_to_le32s(&md5state.hash[i]);
529 	memcpy(iv, &md5state.hash, cc->iv_size);
530 
531 	return 0;
532 }
533 
534 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
535 			    struct dm_crypt_request *dmreq)
536 {
537 	struct scatterlist *sg;
538 	u8 *src;
539 	int r = 0;
540 
541 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
542 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
543 		src = kmap_local_page(sg_page(sg));
544 		r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
545 		kunmap_local(src);
546 	} else
547 		memset(iv, 0, cc->iv_size);
548 
549 	return r;
550 }
551 
552 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
553 			     struct dm_crypt_request *dmreq)
554 {
555 	struct scatterlist *sg;
556 	u8 *dst;
557 	int r;
558 
559 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
560 		return 0;
561 
562 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
563 	dst = kmap_local_page(sg_page(sg));
564 	r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
565 
566 	/* Tweak the first block of plaintext sector */
567 	if (!r)
568 		crypto_xor(dst + sg->offset, iv, cc->iv_size);
569 
570 	kunmap_local(dst);
571 	return r;
572 }
573 
574 static void crypt_iv_tcw_dtr(struct crypt_config *cc)
575 {
576 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
577 
578 	kfree_sensitive(tcw->iv_seed);
579 	tcw->iv_seed = NULL;
580 	kfree_sensitive(tcw->whitening);
581 	tcw->whitening = NULL;
582 
583 	if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
584 		crypto_free_shash(tcw->crc32_tfm);
585 	tcw->crc32_tfm = NULL;
586 }
587 
588 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
589 			    const char *opts)
590 {
591 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
592 
593 	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
594 		ti->error = "Unsupported sector size for TCW";
595 		return -EINVAL;
596 	}
597 
598 	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
599 		ti->error = "Wrong key size for TCW";
600 		return -EINVAL;
601 	}
602 
603 	tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
604 					    CRYPTO_ALG_ALLOCATES_MEMORY);
605 	if (IS_ERR(tcw->crc32_tfm)) {
606 		ti->error = "Error initializing CRC32 in TCW";
607 		return PTR_ERR(tcw->crc32_tfm);
608 	}
609 
610 	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
611 	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
612 	if (!tcw->iv_seed || !tcw->whitening) {
613 		crypt_iv_tcw_dtr(cc);
614 		ti->error = "Error allocating seed storage in TCW";
615 		return -ENOMEM;
616 	}
617 
618 	return 0;
619 }
620 
621 static int crypt_iv_tcw_init(struct crypt_config *cc)
622 {
623 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
624 	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
625 
626 	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
627 	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
628 	       TCW_WHITENING_SIZE);
629 
630 	return 0;
631 }
632 
633 static int crypt_iv_tcw_wipe(struct crypt_config *cc)
634 {
635 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
636 
637 	memset(tcw->iv_seed, 0, cc->iv_size);
638 	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
639 
640 	return 0;
641 }
642 
643 static int crypt_iv_tcw_whitening(struct crypt_config *cc,
644 				  struct dm_crypt_request *dmreq,
645 				  u8 *data)
646 {
647 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
648 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
649 	u8 buf[TCW_WHITENING_SIZE];
650 	SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
651 	int i, r;
652 
653 	/* xor whitening with sector number */
654 	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
655 	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
656 
657 	/* calculate crc32 for every 32bit part and xor it */
658 	desc->tfm = tcw->crc32_tfm;
659 	for (i = 0; i < 4; i++) {
660 		r = crypto_shash_digest(desc, &buf[i * 4], 4, &buf[i * 4]);
661 		if (r)
662 			goto out;
663 	}
664 	crypto_xor(&buf[0], &buf[12], 4);
665 	crypto_xor(&buf[4], &buf[8], 4);
666 
667 	/* apply whitening (8 bytes) to whole sector */
668 	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
669 		crypto_xor(data + i * 8, buf, 8);
670 out:
671 	memzero_explicit(buf, sizeof(buf));
672 	return r;
673 }
674 
675 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
676 			    struct dm_crypt_request *dmreq)
677 {
678 	struct scatterlist *sg;
679 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
680 	__le64 sector = cpu_to_le64(dmreq->iv_sector);
681 	u8 *src;
682 	int r = 0;
683 
684 	/* Remove whitening from ciphertext */
685 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
686 		sg = crypt_get_sg_data(cc, dmreq->sg_in);
687 		src = kmap_local_page(sg_page(sg));
688 		r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
689 		kunmap_local(src);
690 	}
691 
692 	/* Calculate IV */
693 	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
694 	if (cc->iv_size > 8)
695 		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
696 			       cc->iv_size - 8);
697 
698 	return r;
699 }
700 
701 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
702 			     struct dm_crypt_request *dmreq)
703 {
704 	struct scatterlist *sg;
705 	u8 *dst;
706 	int r;
707 
708 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
709 		return 0;
710 
711 	/* Apply whitening on ciphertext */
712 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
713 	dst = kmap_local_page(sg_page(sg));
714 	r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
715 	kunmap_local(dst);
716 
717 	return r;
718 }
719 
720 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
721 				struct dm_crypt_request *dmreq)
722 {
723 	/* Used only for writes, there must be an additional space to store IV */
724 	get_random_bytes(iv, cc->iv_size);
725 	return 0;
726 }
727 
728 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
729 			    const char *opts)
730 {
731 	if (crypt_integrity_aead(cc)) {
732 		ti->error = "AEAD transforms not supported for EBOIV";
733 		return -EINVAL;
734 	}
735 
736 	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
737 		ti->error = "Block size of EBOIV cipher does not match IV size of block cipher";
738 		return -EINVAL;
739 	}
740 
741 	return 0;
742 }
743 
744 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
745 			    struct dm_crypt_request *dmreq)
746 {
747 	struct crypto_skcipher *tfm = any_tfm(cc);
748 	struct skcipher_request *req;
749 	struct scatterlist src, dst;
750 	DECLARE_CRYPTO_WAIT(wait);
751 	unsigned int reqsize;
752 	int err;
753 	u8 *buf;
754 
755 	reqsize = sizeof(*req) + crypto_skcipher_reqsize(tfm);
756 	reqsize = ALIGN(reqsize, __alignof__(__le64));
757 
758 	req = kmalloc(reqsize + cc->iv_size, GFP_NOIO);
759 	if (!req)
760 		return -ENOMEM;
761 
762 	skcipher_request_set_tfm(req, tfm);
763 
764 	buf = (u8 *)req + reqsize;
765 	memset(buf, 0, cc->iv_size);
766 	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
767 
768 	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
769 	sg_init_one(&dst, iv, cc->iv_size);
770 	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
771 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
772 	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
773 	kfree_sensitive(req);
774 
775 	return err;
776 }
777 
778 static void crypt_iv_elephant_dtr(struct crypt_config *cc)
779 {
780 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
781 
782 	crypto_free_skcipher(elephant->tfm);
783 	elephant->tfm = NULL;
784 }
785 
786 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
787 			    const char *opts)
788 {
789 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
790 	int r;
791 
792 	elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
793 					      CRYPTO_ALG_ALLOCATES_MEMORY);
794 	if (IS_ERR(elephant->tfm)) {
795 		r = PTR_ERR(elephant->tfm);
796 		elephant->tfm = NULL;
797 		return r;
798 	}
799 
800 	r = crypt_iv_eboiv_ctr(cc, ti, NULL);
801 	if (r)
802 		crypt_iv_elephant_dtr(cc);
803 	return r;
804 }
805 
806 static void diffuser_disk_to_cpu(u32 *d, size_t n)
807 {
808 #ifndef __LITTLE_ENDIAN
809 	int i;
810 
811 	for (i = 0; i < n; i++)
812 		d[i] = le32_to_cpu((__le32)d[i]);
813 #endif
814 }
815 
816 static void diffuser_cpu_to_disk(__le32 *d, size_t n)
817 {
818 #ifndef __LITTLE_ENDIAN
819 	int i;
820 
821 	for (i = 0; i < n; i++)
822 		d[i] = cpu_to_le32((u32)d[i]);
823 #endif
824 }
825 
826 static void diffuser_a_decrypt(u32 *d, size_t n)
827 {
828 	int i, i1, i2, i3;
829 
830 	for (i = 0; i < 5; i++) {
831 		i1 = 0;
832 		i2 = n - 2;
833 		i3 = n - 5;
834 
835 		while (i1 < (n - 1)) {
836 			d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
837 			i1++; i2++; i3++;
838 
839 			if (i3 >= n)
840 				i3 -= n;
841 
842 			d[i1] += d[i2] ^ d[i3];
843 			i1++; i2++; i3++;
844 
845 			if (i2 >= n)
846 				i2 -= n;
847 
848 			d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
849 			i1++; i2++; i3++;
850 
851 			d[i1] += d[i2] ^ d[i3];
852 			i1++; i2++; i3++;
853 		}
854 	}
855 }
856 
857 static void diffuser_a_encrypt(u32 *d, size_t n)
858 {
859 	int i, i1, i2, i3;
860 
861 	for (i = 0; i < 5; i++) {
862 		i1 = n - 1;
863 		i2 = n - 2 - 1;
864 		i3 = n - 5 - 1;
865 
866 		while (i1 > 0) {
867 			d[i1] -= d[i2] ^ d[i3];
868 			i1--; i2--; i3--;
869 
870 			d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
871 			i1--; i2--; i3--;
872 
873 			if (i2 < 0)
874 				i2 += n;
875 
876 			d[i1] -= d[i2] ^ d[i3];
877 			i1--; i2--; i3--;
878 
879 			if (i3 < 0)
880 				i3 += n;
881 
882 			d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
883 			i1--; i2--; i3--;
884 		}
885 	}
886 }
887 
888 static void diffuser_b_decrypt(u32 *d, size_t n)
889 {
890 	int i, i1, i2, i3;
891 
892 	for (i = 0; i < 3; i++) {
893 		i1 = 0;
894 		i2 = 2;
895 		i3 = 5;
896 
897 		while (i1 < (n - 1)) {
898 			d[i1] += d[i2] ^ d[i3];
899 			i1++; i2++; i3++;
900 
901 			d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
902 			i1++; i2++; i3++;
903 
904 			if (i2 >= n)
905 				i2 -= n;
906 
907 			d[i1] += d[i2] ^ d[i3];
908 			i1++; i2++; i3++;
909 
910 			if (i3 >= n)
911 				i3 -= n;
912 
913 			d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
914 			i1++; i2++; i3++;
915 		}
916 	}
917 }
918 
919 static void diffuser_b_encrypt(u32 *d, size_t n)
920 {
921 	int i, i1, i2, i3;
922 
923 	for (i = 0; i < 3; i++) {
924 		i1 = n - 1;
925 		i2 = 2 - 1;
926 		i3 = 5 - 1;
927 
928 		while (i1 > 0) {
929 			d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
930 			i1--; i2--; i3--;
931 
932 			if (i3 < 0)
933 				i3 += n;
934 
935 			d[i1] -= d[i2] ^ d[i3];
936 			i1--; i2--; i3--;
937 
938 			if (i2 < 0)
939 				i2 += n;
940 
941 			d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
942 			i1--; i2--; i3--;
943 
944 			d[i1] -= d[i2] ^ d[i3];
945 			i1--; i2--; i3--;
946 		}
947 	}
948 }
949 
950 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
951 {
952 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
953 	u8 *es, *ks, *data, *data2, *data_offset;
954 	struct skcipher_request *req;
955 	struct scatterlist *sg, *sg2, src, dst;
956 	DECLARE_CRYPTO_WAIT(wait);
957 	int i, r;
958 
959 	req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
960 	es = kzalloc(16, GFP_NOIO); /* Key for AES */
961 	ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
962 
963 	if (!req || !es || !ks) {
964 		r = -ENOMEM;
965 		goto out;
966 	}
967 
968 	*(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
969 
970 	/* E(Ks, e(s)) */
971 	sg_init_one(&src, es, 16);
972 	sg_init_one(&dst, ks, 16);
973 	skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
974 	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
975 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
976 	if (r)
977 		goto out;
978 
979 	/* E(Ks, e'(s)) */
980 	es[15] = 0x80;
981 	sg_init_one(&dst, &ks[16], 16);
982 	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
983 	if (r)
984 		goto out;
985 
986 	sg = crypt_get_sg_data(cc, dmreq->sg_out);
987 	data = kmap_local_page(sg_page(sg));
988 	data_offset = data + sg->offset;
989 
990 	/* Cannot modify original bio, copy to sg_out and apply Elephant to it */
991 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
992 		sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
993 		data2 = kmap_local_page(sg_page(sg2));
994 		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
995 		kunmap_local(data2);
996 	}
997 
998 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
999 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1000 		diffuser_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1001 		diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1002 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1003 	}
1004 
1005 	for (i = 0; i < (cc->sector_size / 32); i++)
1006 		crypto_xor(data_offset + i * 32, ks, 32);
1007 
1008 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1009 		diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1010 		diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1011 		diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1012 		diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1013 	}
1014 
1015 	kunmap_local(data);
1016 out:
1017 	kfree_sensitive(ks);
1018 	kfree_sensitive(es);
1019 	skcipher_request_free(req);
1020 	return r;
1021 }
1022 
1023 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1024 			    struct dm_crypt_request *dmreq)
1025 {
1026 	int r;
1027 
1028 	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1029 		r = crypt_iv_elephant(cc, dmreq);
1030 		if (r)
1031 			return r;
1032 	}
1033 
1034 	return crypt_iv_eboiv_gen(cc, iv, dmreq);
1035 }
1036 
1037 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1038 				  struct dm_crypt_request *dmreq)
1039 {
1040 	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1041 		return crypt_iv_elephant(cc, dmreq);
1042 
1043 	return 0;
1044 }
1045 
1046 static int crypt_iv_elephant_init(struct crypt_config *cc)
1047 {
1048 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1049 	int key_offset = cc->key_size - cc->key_extra_size;
1050 
1051 	return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1052 }
1053 
1054 static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1055 {
1056 	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1057 	u8 key[ELEPHANT_MAX_KEY_SIZE];
1058 
1059 	memset(key, 0, cc->key_extra_size);
1060 	return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1061 }
1062 
1063 static const struct crypt_iv_operations crypt_iv_plain_ops = {
1064 	.generator = crypt_iv_plain_gen
1065 };
1066 
1067 static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1068 	.generator = crypt_iv_plain64_gen
1069 };
1070 
1071 static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1072 	.generator = crypt_iv_plain64be_gen
1073 };
1074 
1075 static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1076 	.generator = crypt_iv_essiv_gen
1077 };
1078 
1079 static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1080 	.ctr	   = crypt_iv_benbi_ctr,
1081 	.dtr	   = crypt_iv_benbi_dtr,
1082 	.generator = crypt_iv_benbi_gen
1083 };
1084 
1085 static const struct crypt_iv_operations crypt_iv_null_ops = {
1086 	.generator = crypt_iv_null_gen
1087 };
1088 
1089 static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1090 	.ctr	   = crypt_iv_lmk_ctr,
1091 	.dtr	   = crypt_iv_lmk_dtr,
1092 	.init	   = crypt_iv_lmk_init,
1093 	.wipe	   = crypt_iv_lmk_wipe,
1094 	.generator = crypt_iv_lmk_gen,
1095 	.post	   = crypt_iv_lmk_post
1096 };
1097 
1098 static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1099 	.ctr	   = crypt_iv_tcw_ctr,
1100 	.dtr	   = crypt_iv_tcw_dtr,
1101 	.init	   = crypt_iv_tcw_init,
1102 	.wipe	   = crypt_iv_tcw_wipe,
1103 	.generator = crypt_iv_tcw_gen,
1104 	.post	   = crypt_iv_tcw_post
1105 };
1106 
1107 static const struct crypt_iv_operations crypt_iv_random_ops = {
1108 	.generator = crypt_iv_random_gen
1109 };
1110 
1111 static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1112 	.ctr	   = crypt_iv_eboiv_ctr,
1113 	.generator = crypt_iv_eboiv_gen
1114 };
1115 
1116 static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1117 	.ctr	   = crypt_iv_elephant_ctr,
1118 	.dtr	   = crypt_iv_elephant_dtr,
1119 	.init	   = crypt_iv_elephant_init,
1120 	.wipe	   = crypt_iv_elephant_wipe,
1121 	.generator = crypt_iv_elephant_gen,
1122 	.post	   = crypt_iv_elephant_post
1123 };
1124 
1125 /*
1126  * Integrity extensions
1127  */
1128 static bool crypt_integrity_aead(struct crypt_config *cc)
1129 {
1130 	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1131 }
1132 
1133 static bool crypt_integrity_hmac(struct crypt_config *cc)
1134 {
1135 	return crypt_integrity_aead(cc) && cc->key_mac_size;
1136 }
1137 
1138 /* Get sg containing data */
1139 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1140 					     struct scatterlist *sg)
1141 {
1142 	if (unlikely(crypt_integrity_aead(cc)))
1143 		return &sg[2];
1144 
1145 	return sg;
1146 }
1147 
1148 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1149 {
1150 	struct bio_integrity_payload *bip;
1151 	unsigned int tag_len;
1152 	int ret;
1153 
1154 	if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1155 		return 0;
1156 
1157 	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1158 	if (IS_ERR(bip))
1159 		return PTR_ERR(bip);
1160 
1161 	tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1162 
1163 	bip->bip_iter.bi_sector = io->cc->start + io->sector;
1164 
1165 	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1166 				     tag_len, offset_in_page(io->integrity_metadata));
1167 	if (unlikely(ret != tag_len))
1168 		return -ENOMEM;
1169 
1170 	return 0;
1171 }
1172 
1173 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1174 {
1175 #ifdef CONFIG_BLK_DEV_INTEGRITY
1176 	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1177 	struct mapped_device *md = dm_table_get_md(ti->table);
1178 
1179 	/* From now we require underlying device with our integrity profile */
1180 	if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1181 		ti->error = "Integrity profile not supported.";
1182 		return -EINVAL;
1183 	}
1184 
1185 	if (bi->tag_size != cc->on_disk_tag_size ||
1186 	    bi->tuple_size != cc->on_disk_tag_size) {
1187 		ti->error = "Integrity profile tag size mismatch.";
1188 		return -EINVAL;
1189 	}
1190 	if (1 << bi->interval_exp != cc->sector_size) {
1191 		ti->error = "Integrity profile sector size mismatch.";
1192 		return -EINVAL;
1193 	}
1194 
1195 	if (crypt_integrity_aead(cc)) {
1196 		cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1197 		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1198 		       cc->integrity_tag_size, cc->integrity_iv_size);
1199 
1200 		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1201 			ti->error = "Integrity AEAD auth tag size is not supported.";
1202 			return -EINVAL;
1203 		}
1204 	} else if (cc->integrity_iv_size)
1205 		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1206 		       cc->integrity_iv_size);
1207 
1208 	if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1209 		ti->error = "Not enough space for integrity tag in the profile.";
1210 		return -EINVAL;
1211 	}
1212 
1213 	return 0;
1214 #else
1215 	ti->error = "Integrity profile not supported.";
1216 	return -EINVAL;
1217 #endif
1218 }
1219 
1220 static void crypt_convert_init(struct crypt_config *cc,
1221 			       struct convert_context *ctx,
1222 			       struct bio *bio_out, struct bio *bio_in,
1223 			       sector_t sector)
1224 {
1225 	ctx->bio_in = bio_in;
1226 	ctx->bio_out = bio_out;
1227 	if (bio_in)
1228 		ctx->iter_in = bio_in->bi_iter;
1229 	if (bio_out)
1230 		ctx->iter_out = bio_out->bi_iter;
1231 	ctx->cc_sector = sector + cc->iv_offset;
1232 	init_completion(&ctx->restart);
1233 }
1234 
1235 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1236 					     void *req)
1237 {
1238 	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1239 }
1240 
1241 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1242 {
1243 	return (void *)((char *)dmreq - cc->dmreq_start);
1244 }
1245 
1246 static u8 *iv_of_dmreq(struct crypt_config *cc,
1247 		       struct dm_crypt_request *dmreq)
1248 {
1249 	if (crypt_integrity_aead(cc))
1250 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1251 			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1252 	else
1253 		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1254 			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1255 }
1256 
1257 static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1258 		       struct dm_crypt_request *dmreq)
1259 {
1260 	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1261 }
1262 
1263 static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1264 		       struct dm_crypt_request *dmreq)
1265 {
1266 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1267 
1268 	return (__le64 *) ptr;
1269 }
1270 
1271 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1272 		       struct dm_crypt_request *dmreq)
1273 {
1274 	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1275 		  cc->iv_size + sizeof(uint64_t);
1276 
1277 	return (unsigned int *)ptr;
1278 }
1279 
1280 static void *tag_from_dmreq(struct crypt_config *cc,
1281 				struct dm_crypt_request *dmreq)
1282 {
1283 	struct convert_context *ctx = dmreq->ctx;
1284 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1285 
1286 	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1287 		cc->on_disk_tag_size];
1288 }
1289 
1290 static void *iv_tag_from_dmreq(struct crypt_config *cc,
1291 			       struct dm_crypt_request *dmreq)
1292 {
1293 	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1294 }
1295 
1296 static int crypt_convert_block_aead(struct crypt_config *cc,
1297 				     struct convert_context *ctx,
1298 				     struct aead_request *req,
1299 				     unsigned int tag_offset)
1300 {
1301 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1302 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1303 	struct dm_crypt_request *dmreq;
1304 	u8 *iv, *org_iv, *tag_iv, *tag;
1305 	__le64 *sector;
1306 	int r = 0;
1307 
1308 	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1309 
1310 	/* Reject unexpected unaligned bio. */
1311 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1312 		return -EIO;
1313 
1314 	dmreq = dmreq_of_req(cc, req);
1315 	dmreq->iv_sector = ctx->cc_sector;
1316 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1317 		dmreq->iv_sector >>= cc->sector_shift;
1318 	dmreq->ctx = ctx;
1319 
1320 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1321 
1322 	sector = org_sector_of_dmreq(cc, dmreq);
1323 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1324 
1325 	iv = iv_of_dmreq(cc, dmreq);
1326 	org_iv = org_iv_of_dmreq(cc, dmreq);
1327 	tag = tag_from_dmreq(cc, dmreq);
1328 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1329 
1330 	/* AEAD request:
1331 	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1332 	 *  | (authenticated) | (auth+encryption) |              |
1333 	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1334 	 */
1335 	sg_init_table(dmreq->sg_in, 4);
1336 	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1337 	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1338 	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1339 	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1340 
1341 	sg_init_table(dmreq->sg_out, 4);
1342 	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1343 	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1344 	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1345 	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1346 
1347 	if (cc->iv_gen_ops) {
1348 		/* For READs use IV stored in integrity metadata */
1349 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1350 			memcpy(org_iv, tag_iv, cc->iv_size);
1351 		} else {
1352 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1353 			if (r < 0)
1354 				return r;
1355 			/* Store generated IV in integrity metadata */
1356 			if (cc->integrity_iv_size)
1357 				memcpy(tag_iv, org_iv, cc->iv_size);
1358 		}
1359 		/* Working copy of IV, to be modified in crypto API */
1360 		memcpy(iv, org_iv, cc->iv_size);
1361 	}
1362 
1363 	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1364 	if (bio_data_dir(ctx->bio_in) == WRITE) {
1365 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1366 				       cc->sector_size, iv);
1367 		r = crypto_aead_encrypt(req);
1368 		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1369 			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1370 			       cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1371 	} else {
1372 		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1373 				       cc->sector_size + cc->integrity_tag_size, iv);
1374 		r = crypto_aead_decrypt(req);
1375 	}
1376 
1377 	if (r == -EBADMSG) {
1378 		sector_t s = le64_to_cpu(*sector);
1379 
1380 		ctx->aead_failed = true;
1381 		if (ctx->aead_recheck) {
1382 			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1383 				    ctx->bio_in->bi_bdev, s);
1384 			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1385 					 ctx->bio_in, s, 0);
1386 		}
1387 	}
1388 
1389 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1390 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1391 
1392 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1393 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1394 
1395 	return r;
1396 }
1397 
1398 static int crypt_convert_block_skcipher(struct crypt_config *cc,
1399 					struct convert_context *ctx,
1400 					struct skcipher_request *req,
1401 					unsigned int tag_offset)
1402 {
1403 	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1404 	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1405 	struct scatterlist *sg_in, *sg_out;
1406 	struct dm_crypt_request *dmreq;
1407 	u8 *iv, *org_iv, *tag_iv;
1408 	__le64 *sector;
1409 	int r = 0;
1410 
1411 	/* Reject unexpected unaligned bio. */
1412 	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1413 		return -EIO;
1414 
1415 	dmreq = dmreq_of_req(cc, req);
1416 	dmreq->iv_sector = ctx->cc_sector;
1417 	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1418 		dmreq->iv_sector >>= cc->sector_shift;
1419 	dmreq->ctx = ctx;
1420 
1421 	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1422 
1423 	iv = iv_of_dmreq(cc, dmreq);
1424 	org_iv = org_iv_of_dmreq(cc, dmreq);
1425 	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1426 
1427 	sector = org_sector_of_dmreq(cc, dmreq);
1428 	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1429 
1430 	/* For skcipher we use only the first sg item */
1431 	sg_in  = &dmreq->sg_in[0];
1432 	sg_out = &dmreq->sg_out[0];
1433 
1434 	sg_init_table(sg_in, 1);
1435 	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1436 
1437 	sg_init_table(sg_out, 1);
1438 	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1439 
1440 	if (cc->iv_gen_ops) {
1441 		/* For READs use IV stored in integrity metadata */
1442 		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1443 			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1444 		} else {
1445 			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1446 			if (r < 0)
1447 				return r;
1448 			/* Data can be already preprocessed in generator */
1449 			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1450 				sg_in = sg_out;
1451 			/* Store generated IV in integrity metadata */
1452 			if (cc->integrity_iv_size)
1453 				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1454 		}
1455 		/* Working copy of IV, to be modified in crypto API */
1456 		memcpy(iv, org_iv, cc->iv_size);
1457 	}
1458 
1459 	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1460 
1461 	if (bio_data_dir(ctx->bio_in) == WRITE)
1462 		r = crypto_skcipher_encrypt(req);
1463 	else
1464 		r = crypto_skcipher_decrypt(req);
1465 
1466 	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1467 		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1468 
1469 	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1470 	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1471 
1472 	return r;
1473 }
1474 
1475 static void kcryptd_async_done(void *async_req, int error);
1476 
1477 static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1478 				     struct convert_context *ctx)
1479 {
1480 	unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1);
1481 
1482 	if (!ctx->r.req) {
1483 		ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1484 		if (!ctx->r.req)
1485 			return -ENOMEM;
1486 	}
1487 
1488 	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1489 
1490 	/*
1491 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1492 	 * requests if driver request queue is full.
1493 	 */
1494 	skcipher_request_set_callback(ctx->r.req,
1495 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1496 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1497 
1498 	return 0;
1499 }
1500 
1501 static int crypt_alloc_req_aead(struct crypt_config *cc,
1502 				 struct convert_context *ctx)
1503 {
1504 	if (!ctx->r.req_aead) {
1505 		ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1506 		if (!ctx->r.req_aead)
1507 			return -ENOMEM;
1508 	}
1509 
1510 	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1511 
1512 	/*
1513 	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1514 	 * requests if driver request queue is full.
1515 	 */
1516 	aead_request_set_callback(ctx->r.req_aead,
1517 	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1518 	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1519 
1520 	return 0;
1521 }
1522 
1523 static int crypt_alloc_req(struct crypt_config *cc,
1524 			    struct convert_context *ctx)
1525 {
1526 	if (crypt_integrity_aead(cc))
1527 		return crypt_alloc_req_aead(cc, ctx);
1528 	else
1529 		return crypt_alloc_req_skcipher(cc, ctx);
1530 }
1531 
1532 static void crypt_free_req_skcipher(struct crypt_config *cc,
1533 				    struct skcipher_request *req, struct bio *base_bio)
1534 {
1535 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1536 
1537 	if ((struct skcipher_request *)(io + 1) != req)
1538 		mempool_free(req, &cc->req_pool);
1539 }
1540 
1541 static void crypt_free_req_aead(struct crypt_config *cc,
1542 				struct aead_request *req, struct bio *base_bio)
1543 {
1544 	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1545 
1546 	if ((struct aead_request *)(io + 1) != req)
1547 		mempool_free(req, &cc->req_pool);
1548 }
1549 
1550 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1551 {
1552 	if (crypt_integrity_aead(cc))
1553 		crypt_free_req_aead(cc, req, base_bio);
1554 	else
1555 		crypt_free_req_skcipher(cc, req, base_bio);
1556 }
1557 
1558 /*
1559  * Encrypt / decrypt data from one bio to another one (can be the same one)
1560  */
1561 static blk_status_t crypt_convert(struct crypt_config *cc,
1562 			 struct convert_context *ctx, bool atomic, bool reset_pending)
1563 {
1564 	unsigned int tag_offset = 0;
1565 	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1566 	int r;
1567 
1568 	/*
1569 	 * if reset_pending is set we are dealing with the bio for the first time,
1570 	 * else we're continuing to work on the previous bio, so don't mess with
1571 	 * the cc_pending counter
1572 	 */
1573 	if (reset_pending)
1574 		atomic_set(&ctx->cc_pending, 1);
1575 
1576 	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1577 
1578 		r = crypt_alloc_req(cc, ctx);
1579 		if (r) {
1580 			complete(&ctx->restart);
1581 			return BLK_STS_DEV_RESOURCE;
1582 		}
1583 
1584 		atomic_inc(&ctx->cc_pending);
1585 
1586 		if (crypt_integrity_aead(cc))
1587 			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1588 		else
1589 			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1590 
1591 		switch (r) {
1592 		/*
1593 		 * The request was queued by a crypto driver
1594 		 * but the driver request queue is full, let's wait.
1595 		 */
1596 		case -EBUSY:
1597 			if (in_interrupt()) {
1598 				if (try_wait_for_completion(&ctx->restart)) {
1599 					/*
1600 					 * we don't have to block to wait for completion,
1601 					 * so proceed
1602 					 */
1603 				} else {
1604 					/*
1605 					 * we can't wait for completion without blocking
1606 					 * exit and continue processing in a workqueue
1607 					 */
1608 					ctx->r.req = NULL;
1609 					ctx->cc_sector += sector_step;
1610 					tag_offset++;
1611 					return BLK_STS_DEV_RESOURCE;
1612 				}
1613 			} else {
1614 				wait_for_completion(&ctx->restart);
1615 			}
1616 			reinit_completion(&ctx->restart);
1617 			fallthrough;
1618 		/*
1619 		 * The request is queued and processed asynchronously,
1620 		 * completion function kcryptd_async_done() will be called.
1621 		 */
1622 		case -EINPROGRESS:
1623 			ctx->r.req = NULL;
1624 			ctx->cc_sector += sector_step;
1625 			tag_offset++;
1626 			continue;
1627 		/*
1628 		 * The request was already processed (synchronously).
1629 		 */
1630 		case 0:
1631 			atomic_dec(&ctx->cc_pending);
1632 			ctx->cc_sector += sector_step;
1633 			tag_offset++;
1634 			if (!atomic)
1635 				cond_resched();
1636 			continue;
1637 		/*
1638 		 * There was a data integrity error.
1639 		 */
1640 		case -EBADMSG:
1641 			atomic_dec(&ctx->cc_pending);
1642 			return BLK_STS_PROTECTION;
1643 		/*
1644 		 * There was an error while processing the request.
1645 		 */
1646 		default:
1647 			atomic_dec(&ctx->cc_pending);
1648 			return BLK_STS_IOERR;
1649 		}
1650 	}
1651 
1652 	return 0;
1653 }
1654 
1655 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1656 
1657 /*
1658  * Generate a new unfragmented bio with the given size
1659  * This should never violate the device limitations (but if it did then block
1660  * core should split the bio as needed).
1661  *
1662  * This function may be called concurrently. If we allocate from the mempool
1663  * concurrently, there is a possibility of deadlock. For example, if we have
1664  * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1665  * the mempool concurrently, it may deadlock in a situation where both processes
1666  * have allocated 128 pages and the mempool is exhausted.
1667  *
1668  * In order to avoid this scenario we allocate the pages under a mutex.
1669  *
1670  * In order to not degrade performance with excessive locking, we try
1671  * non-blocking allocations without a mutex first but on failure we fallback
1672  * to blocking allocations with a mutex.
1673  *
1674  * In order to reduce allocation overhead, we try to allocate compound pages in
1675  * the first pass. If they are not available, we fall back to the mempool.
1676  */
1677 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1678 {
1679 	struct crypt_config *cc = io->cc;
1680 	struct bio *clone;
1681 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1682 	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1683 	unsigned int remaining_size;
1684 	unsigned int order = MAX_PAGE_ORDER;
1685 
1686 retry:
1687 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1688 		mutex_lock(&cc->bio_alloc_lock);
1689 
1690 	clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1691 				 GFP_NOIO, &cc->bs);
1692 	clone->bi_private = io;
1693 	clone->bi_end_io = crypt_endio;
1694 	clone->bi_ioprio = io->base_bio->bi_ioprio;
1695 
1696 	remaining_size = size;
1697 
1698 	while (remaining_size) {
1699 		struct page *pages;
1700 		unsigned size_to_add;
1701 		unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT);
1702 		order = min(order, remaining_order);
1703 
1704 		while (order > 0) {
1705 			if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) +
1706 					(1 << order) > dm_crypt_pages_per_client))
1707 				goto decrease_order;
1708 			pages = alloc_pages(gfp_mask
1709 				| __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP,
1710 				order);
1711 			if (likely(pages != NULL)) {
1712 				percpu_counter_add(&cc->n_allocated_pages, 1 << order);
1713 				goto have_pages;
1714 			}
1715 decrease_order:
1716 			order--;
1717 		}
1718 
1719 		pages = mempool_alloc(&cc->page_pool, gfp_mask);
1720 		if (!pages) {
1721 			crypt_free_buffer_pages(cc, clone);
1722 			bio_put(clone);
1723 			gfp_mask |= __GFP_DIRECT_RECLAIM;
1724 			order = 0;
1725 			goto retry;
1726 		}
1727 
1728 have_pages:
1729 		size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size);
1730 		__bio_add_page(clone, pages, size_to_add, 0);
1731 		remaining_size -= size_to_add;
1732 	}
1733 
1734 	/* Allocate space for integrity tags */
1735 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1736 		crypt_free_buffer_pages(cc, clone);
1737 		bio_put(clone);
1738 		clone = NULL;
1739 	}
1740 
1741 	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1742 		mutex_unlock(&cc->bio_alloc_lock);
1743 
1744 	return clone;
1745 }
1746 
1747 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1748 {
1749 	struct folio_iter fi;
1750 
1751 	if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */
1752 		bio_for_each_folio_all(fi, clone) {
1753 			if (folio_test_large(fi.folio)) {
1754 				percpu_counter_sub(&cc->n_allocated_pages,
1755 						1 << folio_order(fi.folio));
1756 				folio_put(fi.folio);
1757 			} else {
1758 				mempool_free(&fi.folio->page, &cc->page_pool);
1759 			}
1760 		}
1761 	}
1762 }
1763 
1764 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1765 			  struct bio *bio, sector_t sector)
1766 {
1767 	io->cc = cc;
1768 	io->base_bio = bio;
1769 	io->sector = sector;
1770 	io->error = 0;
1771 	io->ctx.aead_recheck = false;
1772 	io->ctx.aead_failed = false;
1773 	io->ctx.r.req = NULL;
1774 	io->integrity_metadata = NULL;
1775 	io->integrity_metadata_from_pool = false;
1776 	atomic_set(&io->io_pending, 0);
1777 }
1778 
1779 static void crypt_inc_pending(struct dm_crypt_io *io)
1780 {
1781 	atomic_inc(&io->io_pending);
1782 }
1783 
1784 static void kcryptd_queue_read(struct dm_crypt_io *io);
1785 
1786 /*
1787  * One of the bios was finished. Check for completion of
1788  * the whole request and correctly clean up the buffer.
1789  */
1790 static void crypt_dec_pending(struct dm_crypt_io *io)
1791 {
1792 	struct crypt_config *cc = io->cc;
1793 	struct bio *base_bio = io->base_bio;
1794 	blk_status_t error = io->error;
1795 
1796 	if (!atomic_dec_and_test(&io->io_pending))
1797 		return;
1798 
1799 	if (likely(!io->ctx.aead_recheck) && unlikely(io->ctx.aead_failed) &&
1800 	    cc->on_disk_tag_size && bio_data_dir(base_bio) == READ) {
1801 		io->ctx.aead_recheck = true;
1802 		io->ctx.aead_failed = false;
1803 		io->error = 0;
1804 		kcryptd_queue_read(io);
1805 		return;
1806 	}
1807 
1808 	if (io->ctx.r.req)
1809 		crypt_free_req(cc, io->ctx.r.req, base_bio);
1810 
1811 	if (unlikely(io->integrity_metadata_from_pool))
1812 		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1813 	else
1814 		kfree(io->integrity_metadata);
1815 
1816 	base_bio->bi_status = error;
1817 
1818 	bio_endio(base_bio);
1819 }
1820 
1821 /*
1822  * kcryptd/kcryptd_io:
1823  *
1824  * Needed because it would be very unwise to do decryption in an
1825  * interrupt context.
1826  *
1827  * kcryptd performs the actual encryption or decryption.
1828  *
1829  * kcryptd_io performs the IO submission.
1830  *
1831  * They must be separated as otherwise the final stages could be
1832  * starved by new requests which can block in the first stages due
1833  * to memory allocation.
1834  *
1835  * The work is done per CPU global for all dm-crypt instances.
1836  * They should not depend on each other and do not block.
1837  */
1838 static void crypt_endio(struct bio *clone)
1839 {
1840 	struct dm_crypt_io *io = clone->bi_private;
1841 	struct crypt_config *cc = io->cc;
1842 	unsigned int rw = bio_data_dir(clone);
1843 	blk_status_t error = clone->bi_status;
1844 
1845 	if (io->ctx.aead_recheck && !error) {
1846 		kcryptd_queue_crypt(io);
1847 		return;
1848 	}
1849 
1850 	/*
1851 	 * free the processed pages
1852 	 */
1853 	if (rw == WRITE || io->ctx.aead_recheck)
1854 		crypt_free_buffer_pages(cc, clone);
1855 
1856 	bio_put(clone);
1857 
1858 	if (rw == READ && !error) {
1859 		kcryptd_queue_crypt(io);
1860 		return;
1861 	}
1862 
1863 	if (unlikely(error))
1864 		io->error = error;
1865 
1866 	crypt_dec_pending(io);
1867 }
1868 
1869 #define CRYPT_MAP_READ_GFP GFP_NOWAIT
1870 
1871 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1872 {
1873 	struct crypt_config *cc = io->cc;
1874 	struct bio *clone;
1875 
1876 	if (io->ctx.aead_recheck) {
1877 		if (!(gfp & __GFP_DIRECT_RECLAIM))
1878 			return 1;
1879 		crypt_inc_pending(io);
1880 		clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
1881 		if (unlikely(!clone)) {
1882 			crypt_dec_pending(io);
1883 			return 1;
1884 		}
1885 		clone->bi_iter.bi_sector = cc->start + io->sector;
1886 		crypt_convert_init(cc, &io->ctx, clone, clone, io->sector);
1887 		io->saved_bi_iter = clone->bi_iter;
1888 		dm_submit_bio_remap(io->base_bio, clone);
1889 		return 0;
1890 	}
1891 
1892 	/*
1893 	 * We need the original biovec array in order to decrypt the whole bio
1894 	 * data *afterwards* -- thanks to immutable biovecs we don't need to
1895 	 * worry about the block layer modifying the biovec array; so leverage
1896 	 * bio_alloc_clone().
1897 	 */
1898 	clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1899 	if (!clone)
1900 		return 1;
1901 	clone->bi_private = io;
1902 	clone->bi_end_io = crypt_endio;
1903 
1904 	crypt_inc_pending(io);
1905 
1906 	clone->bi_iter.bi_sector = cc->start + io->sector;
1907 
1908 	if (dm_crypt_integrity_io_alloc(io, clone)) {
1909 		crypt_dec_pending(io);
1910 		bio_put(clone);
1911 		return 1;
1912 	}
1913 
1914 	dm_submit_bio_remap(io->base_bio, clone);
1915 	return 0;
1916 }
1917 
1918 static void kcryptd_io_read_work(struct work_struct *work)
1919 {
1920 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1921 
1922 	crypt_inc_pending(io);
1923 	if (kcryptd_io_read(io, GFP_NOIO))
1924 		io->error = BLK_STS_RESOURCE;
1925 	crypt_dec_pending(io);
1926 }
1927 
1928 static void kcryptd_queue_read(struct dm_crypt_io *io)
1929 {
1930 	struct crypt_config *cc = io->cc;
1931 
1932 	INIT_WORK(&io->work, kcryptd_io_read_work);
1933 	queue_work(cc->io_queue, &io->work);
1934 }
1935 
1936 static void kcryptd_io_write(struct dm_crypt_io *io)
1937 {
1938 	struct bio *clone = io->ctx.bio_out;
1939 
1940 	dm_submit_bio_remap(io->base_bio, clone);
1941 }
1942 
1943 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1944 
1945 static int dmcrypt_write(void *data)
1946 {
1947 	struct crypt_config *cc = data;
1948 	struct dm_crypt_io *io;
1949 
1950 	while (1) {
1951 		struct rb_root write_tree;
1952 		struct blk_plug plug;
1953 
1954 		spin_lock_irq(&cc->write_thread_lock);
1955 continue_locked:
1956 
1957 		if (!RB_EMPTY_ROOT(&cc->write_tree))
1958 			goto pop_from_list;
1959 
1960 		set_current_state(TASK_INTERRUPTIBLE);
1961 
1962 		spin_unlock_irq(&cc->write_thread_lock);
1963 
1964 		if (unlikely(kthread_should_stop())) {
1965 			set_current_state(TASK_RUNNING);
1966 			break;
1967 		}
1968 
1969 		schedule();
1970 
1971 		spin_lock_irq(&cc->write_thread_lock);
1972 		goto continue_locked;
1973 
1974 pop_from_list:
1975 		write_tree = cc->write_tree;
1976 		cc->write_tree = RB_ROOT;
1977 		spin_unlock_irq(&cc->write_thread_lock);
1978 
1979 		BUG_ON(rb_parent(write_tree.rb_node));
1980 
1981 		/*
1982 		 * Note: we cannot walk the tree here with rb_next because
1983 		 * the structures may be freed when kcryptd_io_write is called.
1984 		 */
1985 		blk_start_plug(&plug);
1986 		do {
1987 			io = crypt_io_from_node(rb_first(&write_tree));
1988 			rb_erase(&io->rb_node, &write_tree);
1989 			kcryptd_io_write(io);
1990 			cond_resched();
1991 		} while (!RB_EMPTY_ROOT(&write_tree));
1992 		blk_finish_plug(&plug);
1993 	}
1994 	return 0;
1995 }
1996 
1997 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1998 {
1999 	struct bio *clone = io->ctx.bio_out;
2000 	struct crypt_config *cc = io->cc;
2001 	unsigned long flags;
2002 	sector_t sector;
2003 	struct rb_node **rbp, *parent;
2004 
2005 	if (unlikely(io->error)) {
2006 		crypt_free_buffer_pages(cc, clone);
2007 		bio_put(clone);
2008 		crypt_dec_pending(io);
2009 		return;
2010 	}
2011 
2012 	/* crypt_convert should have filled the clone bio */
2013 	BUG_ON(io->ctx.iter_out.bi_size);
2014 
2015 	clone->bi_iter.bi_sector = cc->start + io->sector;
2016 
2017 	if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
2018 	    test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
2019 		dm_submit_bio_remap(io->base_bio, clone);
2020 		return;
2021 	}
2022 
2023 	spin_lock_irqsave(&cc->write_thread_lock, flags);
2024 	if (RB_EMPTY_ROOT(&cc->write_tree))
2025 		wake_up_process(cc->write_thread);
2026 	rbp = &cc->write_tree.rb_node;
2027 	parent = NULL;
2028 	sector = io->sector;
2029 	while (*rbp) {
2030 		parent = *rbp;
2031 		if (sector < crypt_io_from_node(parent)->sector)
2032 			rbp = &(*rbp)->rb_left;
2033 		else
2034 			rbp = &(*rbp)->rb_right;
2035 	}
2036 	rb_link_node(&io->rb_node, parent, rbp);
2037 	rb_insert_color(&io->rb_node, &cc->write_tree);
2038 	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
2039 }
2040 
2041 static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
2042 				       struct convert_context *ctx)
2043 
2044 {
2045 	if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
2046 		return false;
2047 
2048 	/*
2049 	 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
2050 	 * constraints so they do not need to be issued inline by
2051 	 * kcryptd_crypt_write_convert().
2052 	 */
2053 	switch (bio_op(ctx->bio_in)) {
2054 	case REQ_OP_WRITE:
2055 	case REQ_OP_WRITE_ZEROES:
2056 		return true;
2057 	default:
2058 		return false;
2059 	}
2060 }
2061 
2062 static void kcryptd_crypt_write_continue(struct work_struct *work)
2063 {
2064 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2065 	struct crypt_config *cc = io->cc;
2066 	struct convert_context *ctx = &io->ctx;
2067 	int crypt_finished;
2068 	sector_t sector = io->sector;
2069 	blk_status_t r;
2070 
2071 	wait_for_completion(&ctx->restart);
2072 	reinit_completion(&ctx->restart);
2073 
2074 	r = crypt_convert(cc, &io->ctx, true, false);
2075 	if (r)
2076 		io->error = r;
2077 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2078 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2079 		/* Wait for completion signaled by kcryptd_async_done() */
2080 		wait_for_completion(&ctx->restart);
2081 		crypt_finished = 1;
2082 	}
2083 
2084 	/* Encryption was already finished, submit io now */
2085 	if (crypt_finished) {
2086 		kcryptd_crypt_write_io_submit(io, 0);
2087 		io->sector = sector;
2088 	}
2089 
2090 	crypt_dec_pending(io);
2091 }
2092 
2093 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2094 {
2095 	struct crypt_config *cc = io->cc;
2096 	struct convert_context *ctx = &io->ctx;
2097 	struct bio *clone;
2098 	int crypt_finished;
2099 	sector_t sector = io->sector;
2100 	blk_status_t r;
2101 
2102 	/*
2103 	 * Prevent io from disappearing until this function completes.
2104 	 */
2105 	crypt_inc_pending(io);
2106 	crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2107 
2108 	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2109 	if (unlikely(!clone)) {
2110 		io->error = BLK_STS_IOERR;
2111 		goto dec;
2112 	}
2113 
2114 	io->ctx.bio_out = clone;
2115 	io->ctx.iter_out = clone->bi_iter;
2116 
2117 	if (crypt_integrity_aead(cc)) {
2118 		bio_copy_data(clone, io->base_bio);
2119 		io->ctx.bio_in = clone;
2120 		io->ctx.iter_in = clone->bi_iter;
2121 	}
2122 
2123 	sector += bio_sectors(clone);
2124 
2125 	crypt_inc_pending(io);
2126 	r = crypt_convert(cc, ctx,
2127 			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2128 	/*
2129 	 * Crypto API backlogged the request, because its queue was full
2130 	 * and we're in softirq context, so continue from a workqueue
2131 	 * (TODO: is it actually possible to be in softirq in the write path?)
2132 	 */
2133 	if (r == BLK_STS_DEV_RESOURCE) {
2134 		INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2135 		queue_work(cc->crypt_queue, &io->work);
2136 		return;
2137 	}
2138 	if (r)
2139 		io->error = r;
2140 	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2141 	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2142 		/* Wait for completion signaled by kcryptd_async_done() */
2143 		wait_for_completion(&ctx->restart);
2144 		crypt_finished = 1;
2145 	}
2146 
2147 	/* Encryption was already finished, submit io now */
2148 	if (crypt_finished) {
2149 		kcryptd_crypt_write_io_submit(io, 0);
2150 		io->sector = sector;
2151 	}
2152 
2153 dec:
2154 	crypt_dec_pending(io);
2155 }
2156 
2157 static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2158 {
2159 	if (io->ctx.aead_recheck) {
2160 		if (!io->error) {
2161 			io->ctx.bio_in->bi_iter = io->saved_bi_iter;
2162 			bio_copy_data(io->base_bio, io->ctx.bio_in);
2163 		}
2164 		crypt_free_buffer_pages(io->cc, io->ctx.bio_in);
2165 		bio_put(io->ctx.bio_in);
2166 	}
2167 	crypt_dec_pending(io);
2168 }
2169 
2170 static void kcryptd_crypt_read_continue(struct work_struct *work)
2171 {
2172 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2173 	struct crypt_config *cc = io->cc;
2174 	blk_status_t r;
2175 
2176 	wait_for_completion(&io->ctx.restart);
2177 	reinit_completion(&io->ctx.restart);
2178 
2179 	r = crypt_convert(cc, &io->ctx, true, false);
2180 	if (r)
2181 		io->error = r;
2182 
2183 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2184 		kcryptd_crypt_read_done(io);
2185 
2186 	crypt_dec_pending(io);
2187 }
2188 
2189 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2190 {
2191 	struct crypt_config *cc = io->cc;
2192 	blk_status_t r;
2193 
2194 	crypt_inc_pending(io);
2195 
2196 	if (io->ctx.aead_recheck) {
2197 		io->ctx.cc_sector = io->sector + cc->iv_offset;
2198 		r = crypt_convert(cc, &io->ctx,
2199 				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2200 	} else {
2201 		crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2202 				   io->sector);
2203 
2204 		r = crypt_convert(cc, &io->ctx,
2205 				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2206 	}
2207 	/*
2208 	 * Crypto API backlogged the request, because its queue was full
2209 	 * and we're in softirq context, so continue from a workqueue
2210 	 */
2211 	if (r == BLK_STS_DEV_RESOURCE) {
2212 		INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2213 		queue_work(cc->crypt_queue, &io->work);
2214 		return;
2215 	}
2216 	if (r)
2217 		io->error = r;
2218 
2219 	if (atomic_dec_and_test(&io->ctx.cc_pending))
2220 		kcryptd_crypt_read_done(io);
2221 
2222 	crypt_dec_pending(io);
2223 }
2224 
2225 static void kcryptd_async_done(void *data, int error)
2226 {
2227 	struct dm_crypt_request *dmreq = data;
2228 	struct convert_context *ctx = dmreq->ctx;
2229 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2230 	struct crypt_config *cc = io->cc;
2231 
2232 	/*
2233 	 * A request from crypto driver backlog is going to be processed now,
2234 	 * finish the completion and continue in crypt_convert().
2235 	 * (Callback will be called for the second time for this request.)
2236 	 */
2237 	if (error == -EINPROGRESS) {
2238 		complete(&ctx->restart);
2239 		return;
2240 	}
2241 
2242 	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2243 		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2244 
2245 	if (error == -EBADMSG) {
2246 		sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2247 
2248 		ctx->aead_failed = true;
2249 		if (ctx->aead_recheck) {
2250 			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2251 				    ctx->bio_in->bi_bdev, s);
2252 			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2253 					 ctx->bio_in, s, 0);
2254 		}
2255 		io->error = BLK_STS_PROTECTION;
2256 	} else if (error < 0)
2257 		io->error = BLK_STS_IOERR;
2258 
2259 	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2260 
2261 	if (!atomic_dec_and_test(&ctx->cc_pending))
2262 		return;
2263 
2264 	/*
2265 	 * The request is fully completed: for inline writes, let
2266 	 * kcryptd_crypt_write_convert() do the IO submission.
2267 	 */
2268 	if (bio_data_dir(io->base_bio) == READ) {
2269 		kcryptd_crypt_read_done(io);
2270 		return;
2271 	}
2272 
2273 	if (kcryptd_crypt_write_inline(cc, ctx)) {
2274 		complete(&ctx->restart);
2275 		return;
2276 	}
2277 
2278 	kcryptd_crypt_write_io_submit(io, 1);
2279 }
2280 
2281 static void kcryptd_crypt(struct work_struct *work)
2282 {
2283 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2284 
2285 	if (bio_data_dir(io->base_bio) == READ)
2286 		kcryptd_crypt_read_convert(io);
2287 	else
2288 		kcryptd_crypt_write_convert(io);
2289 }
2290 
2291 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2292 {
2293 	struct crypt_config *cc = io->cc;
2294 
2295 	if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2296 	    (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2297 		/*
2298 		 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2299 		 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2300 		 * it is being executed with irqs disabled.
2301 		 */
2302 		if (in_hardirq() || irqs_disabled()) {
2303 			INIT_WORK(&io->work, kcryptd_crypt);
2304 			queue_work(system_bh_wq, &io->work);
2305 			return;
2306 		} else {
2307 			kcryptd_crypt(&io->work);
2308 			return;
2309 		}
2310 	}
2311 
2312 	INIT_WORK(&io->work, kcryptd_crypt);
2313 	queue_work(cc->crypt_queue, &io->work);
2314 }
2315 
2316 static void crypt_free_tfms_aead(struct crypt_config *cc)
2317 {
2318 	if (!cc->cipher_tfm.tfms_aead)
2319 		return;
2320 
2321 	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2322 		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2323 		cc->cipher_tfm.tfms_aead[0] = NULL;
2324 	}
2325 
2326 	kfree(cc->cipher_tfm.tfms_aead);
2327 	cc->cipher_tfm.tfms_aead = NULL;
2328 }
2329 
2330 static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2331 {
2332 	unsigned int i;
2333 
2334 	if (!cc->cipher_tfm.tfms)
2335 		return;
2336 
2337 	for (i = 0; i < cc->tfms_count; i++)
2338 		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2339 			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2340 			cc->cipher_tfm.tfms[i] = NULL;
2341 		}
2342 
2343 	kfree(cc->cipher_tfm.tfms);
2344 	cc->cipher_tfm.tfms = NULL;
2345 }
2346 
2347 static void crypt_free_tfms(struct crypt_config *cc)
2348 {
2349 	if (crypt_integrity_aead(cc))
2350 		crypt_free_tfms_aead(cc);
2351 	else
2352 		crypt_free_tfms_skcipher(cc);
2353 }
2354 
2355 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2356 {
2357 	unsigned int i;
2358 	int err;
2359 
2360 	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2361 				      sizeof(struct crypto_skcipher *),
2362 				      GFP_KERNEL);
2363 	if (!cc->cipher_tfm.tfms)
2364 		return -ENOMEM;
2365 
2366 	for (i = 0; i < cc->tfms_count; i++) {
2367 		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2368 						CRYPTO_ALG_ALLOCATES_MEMORY);
2369 		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2370 			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2371 			crypt_free_tfms(cc);
2372 			return err;
2373 		}
2374 	}
2375 
2376 	/*
2377 	 * dm-crypt performance can vary greatly depending on which crypto
2378 	 * algorithm implementation is used.  Help people debug performance
2379 	 * problems by logging the ->cra_driver_name.
2380 	 */
2381 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2382 	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2383 	return 0;
2384 }
2385 
2386 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2387 {
2388 	int err;
2389 
2390 	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2391 	if (!cc->cipher_tfm.tfms)
2392 		return -ENOMEM;
2393 
2394 	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2395 						CRYPTO_ALG_ALLOCATES_MEMORY);
2396 	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2397 		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2398 		crypt_free_tfms(cc);
2399 		return err;
2400 	}
2401 
2402 	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2403 	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2404 	return 0;
2405 }
2406 
2407 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2408 {
2409 	if (crypt_integrity_aead(cc))
2410 		return crypt_alloc_tfms_aead(cc, ciphermode);
2411 	else
2412 		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2413 }
2414 
2415 static unsigned int crypt_subkey_size(struct crypt_config *cc)
2416 {
2417 	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2418 }
2419 
2420 static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2421 {
2422 	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2423 }
2424 
2425 /*
2426  * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2427  * the key must be for some reason in special format.
2428  * This funcion converts cc->key to this special format.
2429  */
2430 static void crypt_copy_authenckey(char *p, const void *key,
2431 				  unsigned int enckeylen, unsigned int authkeylen)
2432 {
2433 	struct crypto_authenc_key_param *param;
2434 	struct rtattr *rta;
2435 
2436 	rta = (struct rtattr *)p;
2437 	param = RTA_DATA(rta);
2438 	param->enckeylen = cpu_to_be32(enckeylen);
2439 	rta->rta_len = RTA_LENGTH(sizeof(*param));
2440 	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2441 	p += RTA_SPACE(sizeof(*param));
2442 	memcpy(p, key + enckeylen, authkeylen);
2443 	p += authkeylen;
2444 	memcpy(p, key, enckeylen);
2445 }
2446 
2447 static int crypt_setkey(struct crypt_config *cc)
2448 {
2449 	unsigned int subkey_size;
2450 	int err = 0, i, r;
2451 
2452 	/* Ignore extra keys (which are used for IV etc) */
2453 	subkey_size = crypt_subkey_size(cc);
2454 
2455 	if (crypt_integrity_hmac(cc)) {
2456 		if (subkey_size < cc->key_mac_size)
2457 			return -EINVAL;
2458 
2459 		crypt_copy_authenckey(cc->authenc_key, cc->key,
2460 				      subkey_size - cc->key_mac_size,
2461 				      cc->key_mac_size);
2462 	}
2463 
2464 	for (i = 0; i < cc->tfms_count; i++) {
2465 		if (crypt_integrity_hmac(cc))
2466 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2467 				cc->authenc_key, crypt_authenckey_size(cc));
2468 		else if (crypt_integrity_aead(cc))
2469 			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2470 					       cc->key + (i * subkey_size),
2471 					       subkey_size);
2472 		else
2473 			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2474 						   cc->key + (i * subkey_size),
2475 						   subkey_size);
2476 		if (r)
2477 			err = r;
2478 	}
2479 
2480 	if (crypt_integrity_hmac(cc))
2481 		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2482 
2483 	return err;
2484 }
2485 
2486 #ifdef CONFIG_KEYS
2487 
2488 static bool contains_whitespace(const char *str)
2489 {
2490 	while (*str)
2491 		if (isspace(*str++))
2492 			return true;
2493 	return false;
2494 }
2495 
2496 static int set_key_user(struct crypt_config *cc, struct key *key)
2497 {
2498 	const struct user_key_payload *ukp;
2499 
2500 	ukp = user_key_payload_locked(key);
2501 	if (!ukp)
2502 		return -EKEYREVOKED;
2503 
2504 	if (cc->key_size != ukp->datalen)
2505 		return -EINVAL;
2506 
2507 	memcpy(cc->key, ukp->data, cc->key_size);
2508 
2509 	return 0;
2510 }
2511 
2512 static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2513 {
2514 	const struct encrypted_key_payload *ekp;
2515 
2516 	ekp = key->payload.data[0];
2517 	if (!ekp)
2518 		return -EKEYREVOKED;
2519 
2520 	if (cc->key_size != ekp->decrypted_datalen)
2521 		return -EINVAL;
2522 
2523 	memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2524 
2525 	return 0;
2526 }
2527 
2528 static int set_key_trusted(struct crypt_config *cc, struct key *key)
2529 {
2530 	const struct trusted_key_payload *tkp;
2531 
2532 	tkp = key->payload.data[0];
2533 	if (!tkp)
2534 		return -EKEYREVOKED;
2535 
2536 	if (cc->key_size != tkp->key_len)
2537 		return -EINVAL;
2538 
2539 	memcpy(cc->key, tkp->key, cc->key_size);
2540 
2541 	return 0;
2542 }
2543 
2544 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2545 {
2546 	char *new_key_string, *key_desc;
2547 	int ret;
2548 	struct key_type *type;
2549 	struct key *key;
2550 	int (*set_key)(struct crypt_config *cc, struct key *key);
2551 
2552 	/*
2553 	 * Reject key_string with whitespace. dm core currently lacks code for
2554 	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2555 	 */
2556 	if (contains_whitespace(key_string)) {
2557 		DMERR("whitespace chars not allowed in key string");
2558 		return -EINVAL;
2559 	}
2560 
2561 	/* look for next ':' separating key_type from key_description */
2562 	key_desc = strchr(key_string, ':');
2563 	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2564 		return -EINVAL;
2565 
2566 	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2567 		type = &key_type_logon;
2568 		set_key = set_key_user;
2569 	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2570 		type = &key_type_user;
2571 		set_key = set_key_user;
2572 	} else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2573 		   !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2574 		type = &key_type_encrypted;
2575 		set_key = set_key_encrypted;
2576 	} else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2577 		   !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2578 		type = &key_type_trusted;
2579 		set_key = set_key_trusted;
2580 	} else {
2581 		return -EINVAL;
2582 	}
2583 
2584 	new_key_string = kstrdup(key_string, GFP_KERNEL);
2585 	if (!new_key_string)
2586 		return -ENOMEM;
2587 
2588 	key = request_key(type, key_desc + 1, NULL);
2589 	if (IS_ERR(key)) {
2590 		kfree_sensitive(new_key_string);
2591 		return PTR_ERR(key);
2592 	}
2593 
2594 	down_read(&key->sem);
2595 
2596 	ret = set_key(cc, key);
2597 	if (ret < 0) {
2598 		up_read(&key->sem);
2599 		key_put(key);
2600 		kfree_sensitive(new_key_string);
2601 		return ret;
2602 	}
2603 
2604 	up_read(&key->sem);
2605 	key_put(key);
2606 
2607 	/* clear the flag since following operations may invalidate previously valid key */
2608 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2609 
2610 	ret = crypt_setkey(cc);
2611 
2612 	if (!ret) {
2613 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2614 		kfree_sensitive(cc->key_string);
2615 		cc->key_string = new_key_string;
2616 	} else
2617 		kfree_sensitive(new_key_string);
2618 
2619 	return ret;
2620 }
2621 
2622 static int get_key_size(char **key_string)
2623 {
2624 	char *colon, dummy;
2625 	int ret;
2626 
2627 	if (*key_string[0] != ':')
2628 		return strlen(*key_string) >> 1;
2629 
2630 	/* look for next ':' in key string */
2631 	colon = strpbrk(*key_string + 1, ":");
2632 	if (!colon)
2633 		return -EINVAL;
2634 
2635 	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2636 		return -EINVAL;
2637 
2638 	*key_string = colon;
2639 
2640 	/* remaining key string should be :<logon|user>:<key_desc> */
2641 
2642 	return ret;
2643 }
2644 
2645 #else
2646 
2647 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2648 {
2649 	return -EINVAL;
2650 }
2651 
2652 static int get_key_size(char **key_string)
2653 {
2654 	return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2655 }
2656 
2657 #endif /* CONFIG_KEYS */
2658 
2659 static int crypt_set_key(struct crypt_config *cc, char *key)
2660 {
2661 	int r = -EINVAL;
2662 	int key_string_len = strlen(key);
2663 
2664 	/* Hyphen (which gives a key_size of zero) means there is no key. */
2665 	if (!cc->key_size && strcmp(key, "-"))
2666 		goto out;
2667 
2668 	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2669 	if (key[0] == ':') {
2670 		r = crypt_set_keyring_key(cc, key + 1);
2671 		goto out;
2672 	}
2673 
2674 	/* clear the flag since following operations may invalidate previously valid key */
2675 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2676 
2677 	/* wipe references to any kernel keyring key */
2678 	kfree_sensitive(cc->key_string);
2679 	cc->key_string = NULL;
2680 
2681 	/* Decode key from its hex representation. */
2682 	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2683 		goto out;
2684 
2685 	r = crypt_setkey(cc);
2686 	if (!r)
2687 		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2688 
2689 out:
2690 	/* Hex key string not needed after here, so wipe it. */
2691 	memset(key, '0', key_string_len);
2692 
2693 	return r;
2694 }
2695 
2696 static int crypt_wipe_key(struct crypt_config *cc)
2697 {
2698 	int r;
2699 
2700 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2701 	get_random_bytes(&cc->key, cc->key_size);
2702 
2703 	/* Wipe IV private keys */
2704 	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2705 		r = cc->iv_gen_ops->wipe(cc);
2706 		if (r)
2707 			return r;
2708 	}
2709 
2710 	kfree_sensitive(cc->key_string);
2711 	cc->key_string = NULL;
2712 	r = crypt_setkey(cc);
2713 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2714 
2715 	return r;
2716 }
2717 
2718 static void crypt_calculate_pages_per_client(void)
2719 {
2720 	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2721 
2722 	if (!dm_crypt_clients_n)
2723 		return;
2724 
2725 	pages /= dm_crypt_clients_n;
2726 	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2727 		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2728 	dm_crypt_pages_per_client = pages;
2729 }
2730 
2731 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2732 {
2733 	struct crypt_config *cc = pool_data;
2734 	struct page *page;
2735 
2736 	/*
2737 	 * Note, percpu_counter_read_positive() may over (and under) estimate
2738 	 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2739 	 * but avoids potential spinlock contention of an exact result.
2740 	 */
2741 	if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2742 	    likely(gfp_mask & __GFP_NORETRY))
2743 		return NULL;
2744 
2745 	page = alloc_page(gfp_mask);
2746 	if (likely(page != NULL))
2747 		percpu_counter_add(&cc->n_allocated_pages, 1);
2748 
2749 	return page;
2750 }
2751 
2752 static void crypt_page_free(void *page, void *pool_data)
2753 {
2754 	struct crypt_config *cc = pool_data;
2755 
2756 	__free_page(page);
2757 	percpu_counter_sub(&cc->n_allocated_pages, 1);
2758 }
2759 
2760 static void crypt_dtr(struct dm_target *ti)
2761 {
2762 	struct crypt_config *cc = ti->private;
2763 
2764 	ti->private = NULL;
2765 
2766 	if (!cc)
2767 		return;
2768 
2769 	if (cc->write_thread)
2770 		kthread_stop(cc->write_thread);
2771 
2772 	if (cc->io_queue)
2773 		destroy_workqueue(cc->io_queue);
2774 	if (cc->crypt_queue)
2775 		destroy_workqueue(cc->crypt_queue);
2776 
2777 	if (cc->workqueue_id)
2778 		ida_free(&workqueue_ida, cc->workqueue_id);
2779 
2780 	crypt_free_tfms(cc);
2781 
2782 	bioset_exit(&cc->bs);
2783 
2784 	mempool_exit(&cc->page_pool);
2785 	mempool_exit(&cc->req_pool);
2786 	mempool_exit(&cc->tag_pool);
2787 
2788 	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2789 	percpu_counter_destroy(&cc->n_allocated_pages);
2790 
2791 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2792 		cc->iv_gen_ops->dtr(cc);
2793 
2794 	if (cc->dev)
2795 		dm_put_device(ti, cc->dev);
2796 
2797 	kfree_sensitive(cc->cipher_string);
2798 	kfree_sensitive(cc->key_string);
2799 	kfree_sensitive(cc->cipher_auth);
2800 	kfree_sensitive(cc->authenc_key);
2801 
2802 	mutex_destroy(&cc->bio_alloc_lock);
2803 
2804 	/* Must zero key material before freeing */
2805 	kfree_sensitive(cc);
2806 
2807 	spin_lock(&dm_crypt_clients_lock);
2808 	WARN_ON(!dm_crypt_clients_n);
2809 	dm_crypt_clients_n--;
2810 	crypt_calculate_pages_per_client();
2811 	spin_unlock(&dm_crypt_clients_lock);
2812 
2813 	dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2814 }
2815 
2816 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2817 {
2818 	struct crypt_config *cc = ti->private;
2819 
2820 	if (crypt_integrity_aead(cc))
2821 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2822 	else
2823 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2824 
2825 	if (cc->iv_size)
2826 		/* at least a 64 bit sector number should fit in our buffer */
2827 		cc->iv_size = max(cc->iv_size,
2828 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2829 	else if (ivmode) {
2830 		DMWARN("Selected cipher does not support IVs");
2831 		ivmode = NULL;
2832 	}
2833 
2834 	/* Choose ivmode, see comments at iv code. */
2835 	if (ivmode == NULL)
2836 		cc->iv_gen_ops = NULL;
2837 	else if (strcmp(ivmode, "plain") == 0)
2838 		cc->iv_gen_ops = &crypt_iv_plain_ops;
2839 	else if (strcmp(ivmode, "plain64") == 0)
2840 		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2841 	else if (strcmp(ivmode, "plain64be") == 0)
2842 		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2843 	else if (strcmp(ivmode, "essiv") == 0)
2844 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2845 	else if (strcmp(ivmode, "benbi") == 0)
2846 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2847 	else if (strcmp(ivmode, "null") == 0)
2848 		cc->iv_gen_ops = &crypt_iv_null_ops;
2849 	else if (strcmp(ivmode, "eboiv") == 0)
2850 		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2851 	else if (strcmp(ivmode, "elephant") == 0) {
2852 		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2853 		cc->key_parts = 2;
2854 		cc->key_extra_size = cc->key_size / 2;
2855 		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2856 			return -EINVAL;
2857 		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2858 	} else if (strcmp(ivmode, "lmk") == 0) {
2859 		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2860 		/*
2861 		 * Version 2 and 3 is recognised according
2862 		 * to length of provided multi-key string.
2863 		 * If present (version 3), last key is used as IV seed.
2864 		 * All keys (including IV seed) are always the same size.
2865 		 */
2866 		if (cc->key_size % cc->key_parts) {
2867 			cc->key_parts++;
2868 			cc->key_extra_size = cc->key_size / cc->key_parts;
2869 		}
2870 	} else if (strcmp(ivmode, "tcw") == 0) {
2871 		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2872 		cc->key_parts += 2; /* IV + whitening */
2873 		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2874 	} else if (strcmp(ivmode, "random") == 0) {
2875 		cc->iv_gen_ops = &crypt_iv_random_ops;
2876 		/* Need storage space in integrity fields. */
2877 		cc->integrity_iv_size = cc->iv_size;
2878 	} else {
2879 		ti->error = "Invalid IV mode";
2880 		return -EINVAL;
2881 	}
2882 
2883 	return 0;
2884 }
2885 
2886 /*
2887  * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2888  * The HMAC is needed to calculate tag size (HMAC digest size).
2889  * This should be probably done by crypto-api calls (once available...)
2890  */
2891 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2892 {
2893 	char *start, *end, *mac_alg = NULL;
2894 	struct crypto_ahash *mac;
2895 
2896 	if (!strstarts(cipher_api, "authenc("))
2897 		return 0;
2898 
2899 	start = strchr(cipher_api, '(');
2900 	end = strchr(cipher_api, ',');
2901 	if (!start || !end || ++start > end)
2902 		return -EINVAL;
2903 
2904 	mac_alg = kmemdup_nul(start, end - start, GFP_KERNEL);
2905 	if (!mac_alg)
2906 		return -ENOMEM;
2907 
2908 	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2909 	kfree(mac_alg);
2910 
2911 	if (IS_ERR(mac))
2912 		return PTR_ERR(mac);
2913 
2914 	cc->key_mac_size = crypto_ahash_digestsize(mac);
2915 	crypto_free_ahash(mac);
2916 
2917 	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2918 	if (!cc->authenc_key)
2919 		return -ENOMEM;
2920 
2921 	return 0;
2922 }
2923 
2924 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2925 				char **ivmode, char **ivopts)
2926 {
2927 	struct crypt_config *cc = ti->private;
2928 	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2929 	int ret = -EINVAL;
2930 
2931 	cc->tfms_count = 1;
2932 
2933 	/*
2934 	 * New format (capi: prefix)
2935 	 * capi:cipher_api_spec-iv:ivopts
2936 	 */
2937 	tmp = &cipher_in[strlen("capi:")];
2938 
2939 	/* Separate IV options if present, it can contain another '-' in hash name */
2940 	*ivopts = strrchr(tmp, ':');
2941 	if (*ivopts) {
2942 		**ivopts = '\0';
2943 		(*ivopts)++;
2944 	}
2945 	/* Parse IV mode */
2946 	*ivmode = strrchr(tmp, '-');
2947 	if (*ivmode) {
2948 		**ivmode = '\0';
2949 		(*ivmode)++;
2950 	}
2951 	/* The rest is crypto API spec */
2952 	cipher_api = tmp;
2953 
2954 	/* Alloc AEAD, can be used only in new format. */
2955 	if (crypt_integrity_aead(cc)) {
2956 		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2957 		if (ret < 0) {
2958 			ti->error = "Invalid AEAD cipher spec";
2959 			return ret;
2960 		}
2961 	}
2962 
2963 	if (*ivmode && !strcmp(*ivmode, "lmk"))
2964 		cc->tfms_count = 64;
2965 
2966 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2967 		if (!*ivopts) {
2968 			ti->error = "Digest algorithm missing for ESSIV mode";
2969 			return -EINVAL;
2970 		}
2971 		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2972 			       cipher_api, *ivopts);
2973 		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2974 			ti->error = "Cannot allocate cipher string";
2975 			return -ENOMEM;
2976 		}
2977 		cipher_api = buf;
2978 	}
2979 
2980 	cc->key_parts = cc->tfms_count;
2981 
2982 	/* Allocate cipher */
2983 	ret = crypt_alloc_tfms(cc, cipher_api);
2984 	if (ret < 0) {
2985 		ti->error = "Error allocating crypto tfm";
2986 		return ret;
2987 	}
2988 
2989 	if (crypt_integrity_aead(cc))
2990 		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2991 	else
2992 		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2993 
2994 	return 0;
2995 }
2996 
2997 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2998 				char **ivmode, char **ivopts)
2999 {
3000 	struct crypt_config *cc = ti->private;
3001 	char *tmp, *cipher, *chainmode, *keycount;
3002 	char *cipher_api = NULL;
3003 	int ret = -EINVAL;
3004 	char dummy;
3005 
3006 	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
3007 		ti->error = "Bad cipher specification";
3008 		return -EINVAL;
3009 	}
3010 
3011 	/*
3012 	 * Legacy dm-crypt cipher specification
3013 	 * cipher[:keycount]-mode-iv:ivopts
3014 	 */
3015 	tmp = cipher_in;
3016 	keycount = strsep(&tmp, "-");
3017 	cipher = strsep(&keycount, ":");
3018 
3019 	if (!keycount)
3020 		cc->tfms_count = 1;
3021 	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
3022 		 !is_power_of_2(cc->tfms_count)) {
3023 		ti->error = "Bad cipher key count specification";
3024 		return -EINVAL;
3025 	}
3026 	cc->key_parts = cc->tfms_count;
3027 
3028 	chainmode = strsep(&tmp, "-");
3029 	*ivmode = strsep(&tmp, ":");
3030 	*ivopts = tmp;
3031 
3032 	/*
3033 	 * For compatibility with the original dm-crypt mapping format, if
3034 	 * only the cipher name is supplied, use cbc-plain.
3035 	 */
3036 	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
3037 		chainmode = "cbc";
3038 		*ivmode = "plain";
3039 	}
3040 
3041 	if (strcmp(chainmode, "ecb") && !*ivmode) {
3042 		ti->error = "IV mechanism required";
3043 		return -EINVAL;
3044 	}
3045 
3046 	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
3047 	if (!cipher_api)
3048 		goto bad_mem;
3049 
3050 	if (*ivmode && !strcmp(*ivmode, "essiv")) {
3051 		if (!*ivopts) {
3052 			ti->error = "Digest algorithm missing for ESSIV mode";
3053 			kfree(cipher_api);
3054 			return -EINVAL;
3055 		}
3056 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3057 			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
3058 	} else {
3059 		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3060 			       "%s(%s)", chainmode, cipher);
3061 	}
3062 	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3063 		kfree(cipher_api);
3064 		goto bad_mem;
3065 	}
3066 
3067 	/* Allocate cipher */
3068 	ret = crypt_alloc_tfms(cc, cipher_api);
3069 	if (ret < 0) {
3070 		ti->error = "Error allocating crypto tfm";
3071 		kfree(cipher_api);
3072 		return ret;
3073 	}
3074 	kfree(cipher_api);
3075 
3076 	return 0;
3077 bad_mem:
3078 	ti->error = "Cannot allocate cipher strings";
3079 	return -ENOMEM;
3080 }
3081 
3082 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3083 {
3084 	struct crypt_config *cc = ti->private;
3085 	char *ivmode = NULL, *ivopts = NULL;
3086 	int ret;
3087 
3088 	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3089 	if (!cc->cipher_string) {
3090 		ti->error = "Cannot allocate cipher strings";
3091 		return -ENOMEM;
3092 	}
3093 
3094 	if (strstarts(cipher_in, "capi:"))
3095 		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3096 	else
3097 		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3098 	if (ret)
3099 		return ret;
3100 
3101 	/* Initialize IV */
3102 	ret = crypt_ctr_ivmode(ti, ivmode);
3103 	if (ret < 0)
3104 		return ret;
3105 
3106 	/* Initialize and set key */
3107 	ret = crypt_set_key(cc, key);
3108 	if (ret < 0) {
3109 		ti->error = "Error decoding and setting key";
3110 		return ret;
3111 	}
3112 
3113 	/* Allocate IV */
3114 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3115 		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3116 		if (ret < 0) {
3117 			ti->error = "Error creating IV";
3118 			return ret;
3119 		}
3120 	}
3121 
3122 	/* Initialize IV (set keys for ESSIV etc) */
3123 	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3124 		ret = cc->iv_gen_ops->init(cc);
3125 		if (ret < 0) {
3126 			ti->error = "Error initialising IV";
3127 			return ret;
3128 		}
3129 	}
3130 
3131 	/* wipe the kernel key payload copy */
3132 	if (cc->key_string)
3133 		memset(cc->key, 0, cc->key_size * sizeof(u8));
3134 
3135 	return ret;
3136 }
3137 
3138 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3139 {
3140 	struct crypt_config *cc = ti->private;
3141 	struct dm_arg_set as;
3142 	static const struct dm_arg _args[] = {
3143 		{0, 9, "Invalid number of feature args"},
3144 	};
3145 	unsigned int opt_params, val;
3146 	const char *opt_string, *sval;
3147 	char dummy;
3148 	int ret;
3149 
3150 	/* Optional parameters */
3151 	as.argc = argc;
3152 	as.argv = argv;
3153 
3154 	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3155 	if (ret)
3156 		return ret;
3157 
3158 	while (opt_params--) {
3159 		opt_string = dm_shift_arg(&as);
3160 		if (!opt_string) {
3161 			ti->error = "Not enough feature arguments";
3162 			return -EINVAL;
3163 		}
3164 
3165 		if (!strcasecmp(opt_string, "allow_discards"))
3166 			ti->num_discard_bios = 1;
3167 
3168 		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3169 			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3170 		else if (!strcasecmp(opt_string, "high_priority"))
3171 			set_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3172 
3173 		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3174 			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3175 		else if (!strcasecmp(opt_string, "no_read_workqueue"))
3176 			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3177 		else if (!strcasecmp(opt_string, "no_write_workqueue"))
3178 			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3179 		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3180 			if (val == 0 || val > MAX_TAG_SIZE) {
3181 				ti->error = "Invalid integrity arguments";
3182 				return -EINVAL;
3183 			}
3184 			cc->on_disk_tag_size = val;
3185 			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3186 			if (!strcasecmp(sval, "aead")) {
3187 				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3188 			} else if (strcasecmp(sval, "none")) {
3189 				ti->error = "Unknown integrity profile";
3190 				return -EINVAL;
3191 			}
3192 
3193 			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3194 			if (!cc->cipher_auth)
3195 				return -ENOMEM;
3196 		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3197 			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3198 			    cc->sector_size > 4096 ||
3199 			    (cc->sector_size & (cc->sector_size - 1))) {
3200 				ti->error = "Invalid feature value for sector_size";
3201 				return -EINVAL;
3202 			}
3203 			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3204 				ti->error = "Device size is not multiple of sector_size feature";
3205 				return -EINVAL;
3206 			}
3207 			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3208 		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
3209 			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3210 		else {
3211 			ti->error = "Invalid feature arguments";
3212 			return -EINVAL;
3213 		}
3214 	}
3215 
3216 	return 0;
3217 }
3218 
3219 #ifdef CONFIG_BLK_DEV_ZONED
3220 static int crypt_report_zones(struct dm_target *ti,
3221 		struct dm_report_zones_args *args, unsigned int nr_zones)
3222 {
3223 	struct crypt_config *cc = ti->private;
3224 
3225 	return dm_report_zones(cc->dev->bdev, cc->start,
3226 			cc->start + dm_target_offset(ti, args->next_sector),
3227 			args, nr_zones);
3228 }
3229 #else
3230 #define crypt_report_zones NULL
3231 #endif
3232 
3233 /*
3234  * Construct an encryption mapping:
3235  * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3236  */
3237 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3238 {
3239 	struct crypt_config *cc;
3240 	const char *devname = dm_table_device_name(ti->table);
3241 	int key_size, wq_id;
3242 	unsigned int align_mask;
3243 	unsigned int common_wq_flags;
3244 	unsigned long long tmpll;
3245 	int ret;
3246 	size_t iv_size_padding, additional_req_size;
3247 	char dummy;
3248 
3249 	if (argc < 5) {
3250 		ti->error = "Not enough arguments";
3251 		return -EINVAL;
3252 	}
3253 
3254 	key_size = get_key_size(&argv[1]);
3255 	if (key_size < 0) {
3256 		ti->error = "Cannot parse key size";
3257 		return -EINVAL;
3258 	}
3259 
3260 	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3261 	if (!cc) {
3262 		ti->error = "Cannot allocate encryption context";
3263 		return -ENOMEM;
3264 	}
3265 	cc->key_size = key_size;
3266 	cc->sector_size = (1 << SECTOR_SHIFT);
3267 	cc->sector_shift = 0;
3268 
3269 	ti->private = cc;
3270 
3271 	spin_lock(&dm_crypt_clients_lock);
3272 	dm_crypt_clients_n++;
3273 	crypt_calculate_pages_per_client();
3274 	spin_unlock(&dm_crypt_clients_lock);
3275 
3276 	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3277 	if (ret < 0)
3278 		goto bad;
3279 
3280 	/* Optional parameters need to be read before cipher constructor */
3281 	if (argc > 5) {
3282 		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3283 		if (ret)
3284 			goto bad;
3285 	}
3286 
3287 	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3288 	if (ret < 0)
3289 		goto bad;
3290 
3291 	if (crypt_integrity_aead(cc)) {
3292 		cc->dmreq_start = sizeof(struct aead_request);
3293 		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3294 		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3295 	} else {
3296 		cc->dmreq_start = sizeof(struct skcipher_request);
3297 		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3298 		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3299 	}
3300 	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3301 
3302 	if (align_mask < CRYPTO_MINALIGN) {
3303 		/* Allocate the padding exactly */
3304 		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3305 				& align_mask;
3306 	} else {
3307 		/*
3308 		 * If the cipher requires greater alignment than kmalloc
3309 		 * alignment, we don't know the exact position of the
3310 		 * initialization vector. We must assume worst case.
3311 		 */
3312 		iv_size_padding = align_mask;
3313 	}
3314 
3315 	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3316 	additional_req_size = sizeof(struct dm_crypt_request) +
3317 		iv_size_padding + cc->iv_size +
3318 		cc->iv_size +
3319 		sizeof(uint64_t) +
3320 		sizeof(unsigned int);
3321 
3322 	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3323 	if (ret) {
3324 		ti->error = "Cannot allocate crypt request mempool";
3325 		goto bad;
3326 	}
3327 
3328 	cc->per_bio_data_size = ti->per_io_data_size =
3329 		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3330 		      ARCH_DMA_MINALIGN);
3331 
3332 	ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3333 	if (ret) {
3334 		ti->error = "Cannot allocate page mempool";
3335 		goto bad;
3336 	}
3337 
3338 	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3339 	if (ret) {
3340 		ti->error = "Cannot allocate crypt bioset";
3341 		goto bad;
3342 	}
3343 
3344 	mutex_init(&cc->bio_alloc_lock);
3345 
3346 	ret = -EINVAL;
3347 	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3348 	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3349 		ti->error = "Invalid iv_offset sector";
3350 		goto bad;
3351 	}
3352 	cc->iv_offset = tmpll;
3353 
3354 	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3355 	if (ret) {
3356 		ti->error = "Device lookup failed";
3357 		goto bad;
3358 	}
3359 
3360 	ret = -EINVAL;
3361 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3362 		ti->error = "Invalid device sector";
3363 		goto bad;
3364 	}
3365 	cc->start = tmpll;
3366 
3367 	if (bdev_is_zoned(cc->dev->bdev)) {
3368 		/*
3369 		 * For zoned block devices, we need to preserve the issuer write
3370 		 * ordering. To do so, disable write workqueues and force inline
3371 		 * encryption completion.
3372 		 */
3373 		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3374 		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3375 
3376 		/*
3377 		 * All zone append writes to a zone of a zoned block device will
3378 		 * have the same BIO sector, the start of the zone. When the
3379 		 * cypher IV mode uses sector values, all data targeting a
3380 		 * zone will be encrypted using the first sector numbers of the
3381 		 * zone. This will not result in write errors but will
3382 		 * cause most reads to fail as reads will use the sector values
3383 		 * for the actual data locations, resulting in IV mismatch.
3384 		 * To avoid this problem, ask DM core to emulate zone append
3385 		 * operations with regular writes.
3386 		 */
3387 		DMDEBUG("Zone append operations will be emulated");
3388 		ti->emulate_zone_append = true;
3389 	}
3390 
3391 	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3392 		ret = crypt_integrity_ctr(cc, ti);
3393 		if (ret)
3394 			goto bad;
3395 
3396 		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3397 		if (!cc->tag_pool_max_sectors)
3398 			cc->tag_pool_max_sectors = 1;
3399 
3400 		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3401 			cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3402 		if (ret) {
3403 			ti->error = "Cannot allocate integrity tags mempool";
3404 			goto bad;
3405 		}
3406 
3407 		cc->tag_pool_max_sectors <<= cc->sector_shift;
3408 	}
3409 
3410 	wq_id = ida_alloc_min(&workqueue_ida, 1, GFP_KERNEL);
3411 	if (wq_id < 0) {
3412 		ti->error = "Couldn't get workqueue id";
3413 		ret = wq_id;
3414 		goto bad;
3415 	}
3416 	cc->workqueue_id = wq_id;
3417 
3418 	ret = -ENOMEM;
3419 	common_wq_flags = WQ_MEM_RECLAIM | WQ_SYSFS;
3420 	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3421 		common_wq_flags |= WQ_HIGHPRI;
3422 
3423 	cc->io_queue = alloc_workqueue("kcryptd_io-%s-%d", common_wq_flags, 1, devname, wq_id);
3424 	if (!cc->io_queue) {
3425 		ti->error = "Couldn't create kcryptd io queue";
3426 		goto bad;
3427 	}
3428 
3429 	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) {
3430 		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3431 						  common_wq_flags | WQ_CPU_INTENSIVE,
3432 						  1, devname, wq_id);
3433 	} else {
3434 		/*
3435 		 * While crypt_queue is certainly CPU intensive, the use of
3436 		 * WQ_CPU_INTENSIVE is meaningless with WQ_UNBOUND.
3437 		 */
3438 		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3439 						  common_wq_flags | WQ_UNBOUND,
3440 						  num_online_cpus(), devname, wq_id);
3441 	}
3442 	if (!cc->crypt_queue) {
3443 		ti->error = "Couldn't create kcryptd queue";
3444 		goto bad;
3445 	}
3446 
3447 	spin_lock_init(&cc->write_thread_lock);
3448 	cc->write_tree = RB_ROOT;
3449 
3450 	cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3451 	if (IS_ERR(cc->write_thread)) {
3452 		ret = PTR_ERR(cc->write_thread);
3453 		cc->write_thread = NULL;
3454 		ti->error = "Couldn't spawn write thread";
3455 		goto bad;
3456 	}
3457 	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3458 		set_user_nice(cc->write_thread, MIN_NICE);
3459 
3460 	ti->num_flush_bios = 1;
3461 	ti->limit_swap_bios = true;
3462 	ti->accounts_remapped_io = true;
3463 
3464 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3465 	return 0;
3466 
3467 bad:
3468 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3469 	crypt_dtr(ti);
3470 	return ret;
3471 }
3472 
3473 static int crypt_map(struct dm_target *ti, struct bio *bio)
3474 {
3475 	struct dm_crypt_io *io;
3476 	struct crypt_config *cc = ti->private;
3477 
3478 	/*
3479 	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3480 	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3481 	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3482 	 */
3483 	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3484 	    bio_op(bio) == REQ_OP_DISCARD)) {
3485 		bio_set_dev(bio, cc->dev->bdev);
3486 		if (bio_sectors(bio))
3487 			bio->bi_iter.bi_sector = cc->start +
3488 				dm_target_offset(ti, bio->bi_iter.bi_sector);
3489 		return DM_MAPIO_REMAPPED;
3490 	}
3491 
3492 	/*
3493 	 * Check if bio is too large, split as needed.
3494 	 */
3495 	if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) &&
3496 	    (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3497 		dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT));
3498 
3499 	/*
3500 	 * Ensure that bio is a multiple of internal sector encryption size
3501 	 * and is aligned to this size as defined in IO hints.
3502 	 */
3503 	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3504 		return DM_MAPIO_KILL;
3505 
3506 	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3507 		return DM_MAPIO_KILL;
3508 
3509 	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3510 	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3511 
3512 	if (cc->on_disk_tag_size) {
3513 		unsigned int tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3514 
3515 		if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3516 			io->integrity_metadata = NULL;
3517 		else
3518 			io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3519 
3520 		if (unlikely(!io->integrity_metadata)) {
3521 			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3522 				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3523 			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3524 			io->integrity_metadata_from_pool = true;
3525 		}
3526 	}
3527 
3528 	if (crypt_integrity_aead(cc))
3529 		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3530 	else
3531 		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3532 
3533 	if (bio_data_dir(io->base_bio) == READ) {
3534 		if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3535 			kcryptd_queue_read(io);
3536 	} else
3537 		kcryptd_queue_crypt(io);
3538 
3539 	return DM_MAPIO_SUBMITTED;
3540 }
3541 
3542 static char hex2asc(unsigned char c)
3543 {
3544 	return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3545 }
3546 
3547 static void crypt_status(struct dm_target *ti, status_type_t type,
3548 			 unsigned int status_flags, char *result, unsigned int maxlen)
3549 {
3550 	struct crypt_config *cc = ti->private;
3551 	unsigned int i, sz = 0;
3552 	int num_feature_args = 0;
3553 
3554 	switch (type) {
3555 	case STATUSTYPE_INFO:
3556 		result[0] = '\0';
3557 		break;
3558 
3559 	case STATUSTYPE_TABLE:
3560 		DMEMIT("%s ", cc->cipher_string);
3561 
3562 		if (cc->key_size > 0) {
3563 			if (cc->key_string)
3564 				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3565 			else {
3566 				for (i = 0; i < cc->key_size; i++) {
3567 					DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3568 					       hex2asc(cc->key[i] & 0xf));
3569 				}
3570 			}
3571 		} else
3572 			DMEMIT("-");
3573 
3574 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3575 				cc->dev->name, (unsigned long long)cc->start);
3576 
3577 		num_feature_args += !!ti->num_discard_bios;
3578 		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3579 		num_feature_args += test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3580 		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3581 		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3582 		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3583 		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3584 		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3585 		if (cc->on_disk_tag_size)
3586 			num_feature_args++;
3587 		if (num_feature_args) {
3588 			DMEMIT(" %d", num_feature_args);
3589 			if (ti->num_discard_bios)
3590 				DMEMIT(" allow_discards");
3591 			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3592 				DMEMIT(" same_cpu_crypt");
3593 			if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3594 				DMEMIT(" high_priority");
3595 			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3596 				DMEMIT(" submit_from_crypt_cpus");
3597 			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3598 				DMEMIT(" no_read_workqueue");
3599 			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3600 				DMEMIT(" no_write_workqueue");
3601 			if (cc->on_disk_tag_size)
3602 				DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3603 			if (cc->sector_size != (1 << SECTOR_SHIFT))
3604 				DMEMIT(" sector_size:%d", cc->sector_size);
3605 			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3606 				DMEMIT(" iv_large_sectors");
3607 		}
3608 		break;
3609 
3610 	case STATUSTYPE_IMA:
3611 		DMEMIT_TARGET_NAME_VERSION(ti->type);
3612 		DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3613 		DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3614 		DMEMIT(",high_priority=%c", test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags) ? 'y' : 'n');
3615 		DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3616 		       'y' : 'n');
3617 		DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3618 		       'y' : 'n');
3619 		DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3620 		       'y' : 'n');
3621 		DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3622 		       'y' : 'n');
3623 
3624 		if (cc->on_disk_tag_size)
3625 			DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3626 			       cc->on_disk_tag_size, cc->cipher_auth);
3627 		if (cc->sector_size != (1 << SECTOR_SHIFT))
3628 			DMEMIT(",sector_size=%d", cc->sector_size);
3629 		if (cc->cipher_string)
3630 			DMEMIT(",cipher_string=%s", cc->cipher_string);
3631 
3632 		DMEMIT(",key_size=%u", cc->key_size);
3633 		DMEMIT(",key_parts=%u", cc->key_parts);
3634 		DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3635 		DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3636 		DMEMIT(";");
3637 		break;
3638 	}
3639 }
3640 
3641 static void crypt_postsuspend(struct dm_target *ti)
3642 {
3643 	struct crypt_config *cc = ti->private;
3644 
3645 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3646 }
3647 
3648 static int crypt_preresume(struct dm_target *ti)
3649 {
3650 	struct crypt_config *cc = ti->private;
3651 
3652 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3653 		DMERR("aborting resume - crypt key is not set.");
3654 		return -EAGAIN;
3655 	}
3656 
3657 	return 0;
3658 }
3659 
3660 static void crypt_resume(struct dm_target *ti)
3661 {
3662 	struct crypt_config *cc = ti->private;
3663 
3664 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3665 }
3666 
3667 /* Message interface
3668  *	key set <key>
3669  *	key wipe
3670  */
3671 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3672 			 char *result, unsigned int maxlen)
3673 {
3674 	struct crypt_config *cc = ti->private;
3675 	int key_size, ret = -EINVAL;
3676 
3677 	if (argc < 2)
3678 		goto error;
3679 
3680 	if (!strcasecmp(argv[0], "key")) {
3681 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3682 			DMWARN("not suspended during key manipulation.");
3683 			return -EINVAL;
3684 		}
3685 		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3686 			/* The key size may not be changed. */
3687 			key_size = get_key_size(&argv[2]);
3688 			if (key_size < 0 || cc->key_size != key_size) {
3689 				memset(argv[2], '0', strlen(argv[2]));
3690 				return -EINVAL;
3691 			}
3692 
3693 			ret = crypt_set_key(cc, argv[2]);
3694 			if (ret)
3695 				return ret;
3696 			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3697 				ret = cc->iv_gen_ops->init(cc);
3698 			/* wipe the kernel key payload copy */
3699 			if (cc->key_string)
3700 				memset(cc->key, 0, cc->key_size * sizeof(u8));
3701 			return ret;
3702 		}
3703 		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3704 			return crypt_wipe_key(cc);
3705 	}
3706 
3707 error:
3708 	DMWARN("unrecognised message received.");
3709 	return -EINVAL;
3710 }
3711 
3712 static int crypt_iterate_devices(struct dm_target *ti,
3713 				 iterate_devices_callout_fn fn, void *data)
3714 {
3715 	struct crypt_config *cc = ti->private;
3716 
3717 	return fn(ti, cc->dev, cc->start, ti->len, data);
3718 }
3719 
3720 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3721 {
3722 	struct crypt_config *cc = ti->private;
3723 
3724 	limits->logical_block_size =
3725 		max_t(unsigned int, limits->logical_block_size, cc->sector_size);
3726 	limits->physical_block_size =
3727 		max_t(unsigned int, limits->physical_block_size, cc->sector_size);
3728 	limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size);
3729 	limits->dma_alignment = limits->logical_block_size - 1;
3730 }
3731 
3732 static struct target_type crypt_target = {
3733 	.name   = "crypt",
3734 	.version = {1, 26, 0},
3735 	.module = THIS_MODULE,
3736 	.ctr    = crypt_ctr,
3737 	.dtr    = crypt_dtr,
3738 	.features = DM_TARGET_ZONED_HM,
3739 	.report_zones = crypt_report_zones,
3740 	.map    = crypt_map,
3741 	.status = crypt_status,
3742 	.postsuspend = crypt_postsuspend,
3743 	.preresume = crypt_preresume,
3744 	.resume = crypt_resume,
3745 	.message = crypt_message,
3746 	.iterate_devices = crypt_iterate_devices,
3747 	.io_hints = crypt_io_hints,
3748 };
3749 module_dm(crypt);
3750 
3751 MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3752 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3753 MODULE_LICENSE("GPL");
3754